Recsperts - Recommender Systems Experts

Note: This transcript has been generated automatically using OpenAI's whisper and may contain inaccuracies or errors. We recommend listening to the audio for a better understanding of the content. Please feel free to reach out if you spot any corrections that need to be made. Thank you for your understanding.

You cannot plan serendipity, you cannot engineer serendipity.
I think the only thing you can do is try to design for serendipity and try to design your technology, your environment in a way that increases the likelihood of serendipity to happen.
I don't think that commercial objectives are per definition against serendipity.
We have not often asked the question, what is the value of serendipity for someone else than the ones who is experiencing serendipity.
So a lot of research focuses on serendipity in relation to a user interacting with an item or someone discovering something and for them that's valuable.
And I think that is also the right way to look at serendipity.
But in order to encourage designers and other stakeholders to design for serendipity, we of course also need to understand what's in it for them.
And I think we've not often asked that question, so we simply don't know what could be in it for other stakeholders.
When people expect to be surprised and they're not surprised, their experience is way worse than actually when they were not really expecting to be surprised.
After all the research that we have done, an idea that I really want to have out of the world is that serendipity bursts filter bubbles.
Hello and welcome to this new episode of RECSPERTSs, recommender systems experts.
It's been a while, but the RecSys journey must continue and I can tell you it will continue.
And it will continue with today's episode since for today we are talking about something unexpected, something fresh, something novel and something that might bring back recent tastes.
And from these descriptions you might already get an understanding of what the topic is for today's episode.
Today we are talking about serendipity in recommender systems.
And for this talk I'm very happy to be joined by a real expert on serendipity and recommender systems.
And therefore I'm very happy to welcome Professor Annelien Smets to the episode.
Hello Annelien.
Hello, thank you for having me.
Yeah, it's nice that you joined for this episode and I've been planning to do this for quite a long time.
And also apologies to my interview candidates and to my listeners because I have been relocating this year and it was also quite a lot of work for me.
But I'm very motivated to continue doing this podcast and delivering more exciting topics from recommender systems to this growing audience, which I'm very happy about.
And as always, before we kickstart the discussion and before I hand over to my guest to introduce herself and also into the topic, I will just start with a short introduction of Professor Annelien Smets.
So Professor Annelien Smets is a professor at Vrije Universiteit Brussel and she is also a senior researcher at IMAG-SMIT, which is Studies in Media, Information and Technology, co-located with Free University of Brussels.
Her research centers around personalization and recommender systems and their value in media markets.
She has a specific interest in discovery through recommenders and we will learn today how serendipity can actually help with that.
Besides being a professor at Vrije Universiteit Brussel, she also holds a Ph.D. in Media and Communication Studies and actual two master's degrees.
One that is in Artificial Intelligence with a specific focus on speech and language and a master degree in Business Information Management and also a bachelor's in Business Economics.
So we do have a guest with a broad set of experiences and interests.
Besides that, and interestingly though, and I guess she is the best person to be qualified for this, is the fact that she is the co-founder and chair of the Serendipity Academic Research Network, SARN.
That is part of the Serendipity Society.
She actually has published research at conferences such as Humab, RecSys, also in different journals, especially the transactions and recommender systems.
And in 2023 she also received the Women in RecSys Journal Paper of the Year Award.
So Annelien, congrats to that. Again, welcome to the episode and tell us more about yourself and your excitement for recommender systems and personalization and more specifically for Serendipity.
Yeah, thank you for the nice introduction.
So currently I'm affiliated with a research group in Media and Communication Studies departments.
And actually the nice thing about Media and Communication Studies is that it's a very broad discipline.
So we are concerned with a lot of methods, but also with a lot of topics.
And in our research group, so at SMIT we particularly focus on technology and media.
And that's also how I got interested in recommender systems, because recommender systems have become a very important technology in media and media consumption.
And from communication studies, when you talk about recommender systems, especially a few years ago, the first thing that came to your mind was filter bubbles.
So initially when I started to do my research, I looked a lot into filter bubbles and how they relate to recommender systems.
But then actually from the question, like, how can we make use of these technologies, of these recommender systems, but without having these negative consequences of filter bubbles?
Because in general, we are not very fond of having filter bubbles.
And so that became like the general question of my PhD, like, how can we use recommender systems without introducing filter bubbles?
And there actually my supervisor all of a sudden throw up this word serendipity.
And I had never heard of the term before.
So that was a very interesting situation where I was dealing with serendipity, which I didn't know, which turned out to be a very complex concept.
Okay, so breaking up the term serendipity was in itself a serendipitous experience for you.
Well, in hindsight, since it turns out to be a good experience, it was indeed quite serendipitous.
So, yeah, that's how the story evolved.
So that's how I got to really, first of all, trying to understand serendipity and especially serendipity in the context of recommender systems.
And I've also seen your doctoral thesis that I guess you concluded in 2022, if I'm not mistaken.
It was actually on urban recommender systems.
So can you share with us what are urban recommender systems or what is the purpose of urban recommender systems?
What do they recommend?
Indeed.
Well, often they recommend points of interests.
So I use the term urban recommender systems to refer to recommender systems that you use in an urban setting and actually in an offline setting.
And there actually, so my initially when I started to do research, my work was a lot on smart cities and on thinking about using technologies in cities.
And there we had this idea that if we start using technologies and personalized technologies like recommender systems to recommend those places to go, restaurants to have dinner, activities to do.
Can we also end up in an offline filter bubble?
Could it be that there will be an urban filter bubble?
So that's actually how I connected it to urban recommender systems.
So there we need to look at, yeah, recommenders that recommend points of interest, that could be activities to do, monuments to go see, places to have dinner, like a very broad range of categories.
And possibly I didn't really get this from the introduction or I misunderstood it, but was this thesis actually your first touch point with recommender systems or was there some prior touch point with personalization or also like with recommender systems as a means for personalization?
No, apart from, so when I was studying, I did have some courses on machine learning and their recommender systems probably have been like one course or like one lesson, but that was it basically.
So when I started the PhD and when I started to do the research, it was like my first proper encounter with recommender systems.
And how come that basically like from zero to hero, you went into recommender systems because yeah, the advanced masters in artificial intelligence, it provided already some touch points, I guess.
And there was this economics background, but was it more like also coincidence bringing you to Turexis or before you started with the PhD, was there something in specific?
No, actually it's a good question. I think it's just a real coincidence that there was this clear link with filter bubbles with this whole idea and this person.
And I also am generally interested in this technology and how it can help bring people in contact with content, with items. I think also from a non-professional account, I'm interested in it.
Sounds definitely cool. Like, I mean, from there you turned into a professor nowadays guiding research, teaching others on this amazing topic.
And you were on my RecSys RECSPERTSs Raider quite some time and I missed to reach out and was busy also with lots of things.
But if I ever want to do an episode on serendipity, it's definitely you who are sharing this and also having great collaborations with other researchers.
I guess one of my prior guests, Lian, is also collaborating with you or part of your research lab.
There was this work, I guess it was already in summer by one of your PhD students, if I'm not mistaken, Brett Binst, who shared some work, reminding me of, oh gosh, I need to do this episode.
I need to talk with her and we need to talk about serendipity. So I'm very happy that we finally made it and that I'm also continuing podcasting.
And as I mentioned in the very beginning, today's focus topic should be serendipity.
And there, as I said, you were the go-to person for me and I also at the beginning asked myself where in my almost 10 years of developing recommender models and developing systems and still continuing excitement, where have I actually getting in touch with serendipity?
And I was reminded of one specific paper. It was actually a paper by Google, like 2022, 2023, on the values of user exploration that mentioned these beyond accuracy metrics and part of them being serendipity in the very likely YouTube video recommendation setting.
And I was, okay, this is a metric. This is a metric that is somewhat related to user retention in the long term.
But when I was reading your papers, I found the concept of serendipity is very inconsistently defined, applied, and there even consists such a thing as the paradox of serendipity.
Let's start with it. Can you share with us, elaborate and describe what this paradox is about?
Yeah. So in order to discuss the paradox of serendipity, I should maybe discuss very briefly what we understand when we talk about serendipity in general.
When you ask someone what is serendipity, I think the first things that come to mind is like randomness and unexpected things, surprise at things that happen just by accident.
A lot of people when they start thinking and it goes, it's not just a question that has been asked today.
It's actually already since the end of the 90s that they started to think about serendipity and technologies and computers.
Yeah, when you start to think about, okay, what is a computer doing or what are algorithms doing? Well, they're trying to plan things. They're trying to program things.
They're trying to predict things, which just semantically already seems to be opposing each other.
Serendipity is about things that happen randomly without anyone having control over it.
And when we start thinking of adding serendipity to algorithms or to technologies, it feels like something that we should be able to design, to construct, to engineer.
So that's often why some people say that you cannot engineer serendipity. And to be fair, I agree with that.
I really think that you cannot plan serendipity. You cannot engineer serendipity.
I think the only thing you can do is try to design for serendipity and try to design your technology, your environment in a way that increases the likelihood of serendipity to happen.
Yeah, as you are saying this new ideas or new resemblances of this in the real world start to manifest.
You could either stay at home and not have serendipitous encounters or you could go to a cafe alone and this might increase your likelihood of having serendipitous encounters.
Would this be a real world analog to what you mean when you say design for serendipity?
Yes, I think that's a very personal thing you can do. That's putting yourself in situations where serendipity is more likely to happen.
So when I explain this idea of designing for serendipity, I like to make the analogy to something that's often, I mean, a lot of people know is actually when we think about healthy behavior.
When we cannot make sure that people behave in the most healthy way, that they take the stairs instead of the elevators, for example.
But what we can do is try to design the environment in such a way that taking the stairs is actually more attractive compared to taking the elevator.
Like sometimes you have this thing where stairs are covered in piano notes and when you take them they make music.
So that's actually a way to design the environment to kind of nudge people to perform this more healthy behavior.
You can think about serendipity in the same way. We can try to design the environment in such a way that people who are in that environment pay attention maybe to small details or have the opportunity to see things that maybe otherwise they wouldn't have seen.
So you can start thinking about how should we design the environment in such a way that serendipity can happen.
When you say designing the environment, in this episode, we are also going to learn more about that this goes beyond the pure recommendation model, but really into a more systematic view of the overall system in which a model is embedded.
And that simply coining serendipity as a single narrow metric might not be sufficient to the needs.
But when we say needs, this of course brings up a very apparent question, which is why does it actually need serendipity?
So why should we design for serendipity or why should companies design for serendipity?
So I think that's a question that we often don't ask. I think in general when people in technology or broader think about serendipity, it often feels like a word you encounter on a Pinterest board in quotes and not something you encounter in an art core technology environment, for example.
Or it's something that's related to public values and something that you have to achieve to reach the broader societal goals because serendipity is about creativity, innovation.
So in general, that's often like a broad idea related to serendipity.
And I was really intrigued by the fact that you remembered the Google paper from I think a few years ago because they indeed exactly showed that serendipity, including serendipity in their design, had positive outcomes for them.
I think they showed that the long-term user satisfaction increased and that it really had a positive effect on the overall user experience.
And I think that's important. So understanding the consequences of experiencing serendipity, I think that's an important thing we should understand better and know better.
Also to encourage people to include it in their designs.
So would you agree saying that serendipity is an undervalued concept in recommender systems?
I'm very happy that you put it in words, but I totally agree.
I think it also relates to the fact that it's such a, I mean, I'm the first to say that serendipity is a very vague concept and that it's very clear.
I think that there are so many definitions and so many different metrics to measure it and ways to implement it that it feels like, yeah, a never-ending thing when you start reading about it and you don't have like clear instructions on this is how to do it.
But I definitely think that there is a lot of value in it. And if we do it in a proper way, it can yield a lot of value for a lot of stakeholders who are interacting with the product.
That's also the way we look at serendipity in the context of recommender systems is we really try to understand it from this multistakeholder perspective.
Like, okay, you can think of introducing this concept of serendipity to your products.
Yeah. Why do you want to do that? Is it just because you think serendipity is a very cool metric or is it because you want to create some added value for your end user?
Or is it maybe because you have a very large item catalog and you have a very long tail?
So maybe a lot of items are not recommended often.
Then you can also start thinking about using this concept of serendipity to maybe increase those long tail items.
Or are there maybe other reasons why you would want to include serendipity? So I think that's the, yeah, in general, the very broad question we look at serendipity.
Maybe it's already too early to ask that question, but staying a bit with the why questions here.
Why do you think it is undervalued?
I think it is undervalued because we have not often asked the question, what is the value of serendipity for someone else than the ones who is experiencing serendipity?
So a lot of research focuses on serendipity in relation to a user interacting with an item or someone discovering something and for them that's valuable.
And I think that is also the right way to look at serendipity. But in order to encourage designers and other stakeholders to design for serendipity, we of course also need to understand what's in it for them.
And I think we've not often asked that question. So we simply don't know what could be in it for other stakeholders.
Or like those who advocate for it haven't really done their best job in explaining the benefits of serendipity for different stakeholders.
I mean, even this user centric motivation for serendipity is already like a bit hard to consistently define.
But then if you go beyond it and you have mentioned the aspect of multi stakeholders, so looking at the providers of items or looking at the platform itself, and especially as a platform has interest.
If we stay with the industrial context, then it might be right to say that users like serendipitous encounters.
But this is not sufficient, of course, to convince an organization to invest in this, because for this you also need to answer the question, like, what is the benefit that the company has from investing into designing for serendipity?
Which might be something that Google has helped us to understand better within their paper, saying like, okay, we want users to stay with the platform.
The way we measure this is retention. And we see that serendipity is correlated with retention and maybe it's also causal for retention partially.
And then you have going back to this multi-stakeholder view, also the argument of we stay like with popularity bias, and I guess you mentioned this also in one of your papers, that serendipity can also be a means to counter popularity bias, like surfacing more from the from the long tail items to be shown to the user, but also thereby granting those providers the ability to be seen and consumed.
I mean, for me asking this question, why do you want serendipity or why does that sound an interesting concept?
For me, that's also a way to start making it a less vague concept, because I really think that serendipity is experienced in many, many different ways.
So even when you're interacting with the same product, like for example, if I'm opening a streaming service and I see a movie that I didn't know about, but somehow it appeals me and I start watching it and I like it.
You could say that is serendipity. It was something that I didn't know before. But maybe when interacting with that same service, but maybe on another day and I see a movie that actually I've watched like 10 years ago, but I had forgotten about and I watch it.
That can also give me an experience of serendipity. So I think it really happens in a lot of flavors.
And of course, if you want to design for serendipity, the first thing is asking, OK, but what type of serendipity are we actually interested in?
Because if we were talking about novel items or items that the user actually knows, but had forgotten about, if you want to turn this into practice, those are two very, very different things you actually want to recommend.
So I think by first of all starting to think about, OK, serendipity kind of makes sense to us.
But what does this actually mean and why do we want to do it? Is this indeed to make our end users happier or is this to make item providers happier, which could also be the case?
But I think started thinking about this and think, OK, what does this mean and how should we move this forward?
It can help you to make this fake concept into something that is way more tangible and practical to actually use.
And I'm happy to be joined by somebody who is definitely helping our listeners today to get this clearer and better understanding.
And for this, I was taking as a starting point to cut through your research a paper that was published in 2025.
So this year, at the very end of the year that we are recording this, and I hope this time I will be a bit faster with editing this episode so that we can also publish it more timely.
And it was a paper that was published in Ethics and Information Technology.
And the name of the paper intended, afforded and experienced serendipity, overcoming the paradox of artificial serendipity.
Let's take this as a starting point and talk about this intended serendipity that brings us closer to what are the actual objectives and goals we try to serve with serendipity.
So can you give us an introduction to the paper and can we take it from there?
Yeah, sure. So this paper is actually kind of a, you could call it a summary of all the work that we have been doing in the last year.
And it really explains how we look at serendipity and how we try to make it into something that you can actually design for.
We break up serendipity and indeed in these three concepts, intended serendipity, which relates to why do we actually want to design for serendipity and which value do you want to create for whom?
And start thinking about your different stakeholders and how this effect that an end user is experiencing serendipity, how that might be relevant for maybe other parties as well.
If you have established that, it means that you have an idea of what serendipity will be about.
And then the next question is actually asking, okay, imagine we want to surface these long tail items.
That's our goal. We want to make sure that our catalog has a better coverage and that also long tail items are being served.
So this could be a possible intention.
Yeah. I mean, actually, if we want to go a bit deeper into that, in another paper, I identified four different intentions to design for serendipity.
Very broadly, it's also quite theoretically.
The first one is serendipity as an ideal.
I think it's also kind of a bit of the most common interpretation of serendipity.
So when you think of serendipity as an ideal, there are the ideas that as a designer, you want to design for serendipity because you really care about the valuable encounter that your user will have with the items.
That's the only thing you care about.
You really want to make sure that your users encounter new ideas, new items, and that they that it brings value for them and that it enriches their lives.
Like a very typical example, there would be a library, not really technological, but libraries in essence.
What do they want? They want you to be able to find the things you're looking for, but also to have new ideas to open up your perspectives.
So that would be like the ideal serendipity type.
So like to think about an ideal also human being as someone who is broadly interested and if like somebody has an oral interest, at least like give them the cues in a daily life, for example, through the means of a library to facilitate this process of also engaging with new content or new topics, whatever.
Yeah.
So that's and that's the only thing that in that case you care about.
So everything you do is really focusing on making sure that they encounter these interesting items.
And then another type is what I termed serendipity as a common good.
And there it's quite similar, but you look a bit at a more high level perspective and you think about the aggregate effects of a lot of users.
Experiencing serendipity and getting new ideas.
Again, a stereotypical example is making sure that people encounter diverse viewpoints, that they are open to different perspectives.
And then on a societal level, we hope that this strengthens democracy and that it allows for a good discussion and people understanding one another.
So that's you could also think of serendipity as a way to achieve that kind of high level impact and effects.
Yeah.
Yeah. Could a corresponding like very technical and in the RecSys world example really be this case of YouTube promoting serendipity through adjustments of the recommender model and then seeing that users are interacting with more diverse content?
Could this be like a corresponding thing?
Yeah. Yes. Kind of. I mean, again, this is this four types are very theoretical. So I think in practice they often co-occur.
And I think, for example, in the case of YouTube, OK, they will add some value to having people encountering diverse perspectives, but most likely they will also be very interested in the fact that maybe that then relates to longer time spent on the platform.
Yeah, for sure. For sure. Retention. And I mean, it's not a nonprofit organization, as we all know.
Exactly. Yeah. Yeah. But I think that's also important to realize. Like I had a history teacher in high school and she said money makes the world go round.
I mean, it's a very bold statement, but indeed, in history, money and power, it often plays a role. And I mean, we should not be naive to host a big platform to make sure that it works.
Probably you need money. So I don't think that's in essence, economic incentives are not bad. Of course, we should make sure to balance them with with other values.
But that's also a bit what what I tried to do with understanding this why of serendipity and the value of serendipity.
I don't think that commercial objectives are per definition against serendipity.
I often that's that's a discourse that you often hear that as soon as there is a commercial incentive, serendipity probably will not be in scope.
But I know from having had discussions on serendipity with a lot of people working in commercial organizations, they do care about this idea of serendipity because like Google showed, it also has some value for them.
When people discover new things, when they know that they and they come to the platform and they will see something new.
It also has a commercial side of it, and I think we should be aware of that and also especially that's also what we try to do.
Think of ways to safeguard the essence of serendipity to make sure that serendipity is still something good and valuable for end users, but also acknowledging that there can be some commercial aspects related to it.
Yeah, yeah. So dear listeners, you hear that there are two people talking with each other who both have a background in.
Okay, okay. So you mentioned the second intention is the common good and how does it continue?
And then you have the one we've been talking about already, which I call serendipity as a mediator, where you use or you think about introducing serendipity to achieve something else, a goal that you're interested in.
For example, we want to introduce serendipity in our recommender system because actually what we care about is surfacing these long tail items.
And we will think about serendipity as a means to achieve this outcome.
And there the example from Google with what they saw with the long-term user experiences is also, I would categorize that also as serendipity as a mediator.
Okay, okay. I understand. And there you already see like those correspondence because perceiving or taking the intent of common good with keeping in mind the consequence of lively, diverse democracy could also be seen as kind of a mediator.
So in that sense, there's no like very 100% clear delineation between those concepts, but some of them are more or less pronounced.
Yeah, that's, I mean, in academia, we try to have like categories and nicely delineated between those in practice if it works out a little bit differently.
Yeah, yeah.
And then the final one is serendipity as a feature.
And there serendipity, the fact that people experience serendipity is actually your value proposition.
So the product that I'm making, you're using it because you want to experience serendipity.
It's like AI nowadays, right? So my toothbrush has AI where people don't take AI anymore as like a means to achieve something or to solve a problem, but as a feature in itself.
Yeah, yeah.
Sorry, I might be ridiculing it a bit. This was not my intent, but.
No, but I mean, I think it's also the least, it's also the hardest to find good examples of.
Like, for example, there are some mostly research projects. For example, it's an app there. We come to the urban part as well.
And it's an app that you use with your family and they can at random points in the city leave a movie clip.
For example, my nephew, he walks through the city and it's at a bench. He records a video and he sort of stores it there.
And when I walk by that bench and I don't know, but then the app will give me a notification and I will see the movie that my nephew recorded.
And there experiencing serendipity is really part of the concept of the application, part of the idea of the product.
Reminds me a bit of geocaching, just without the aspect that you like search for some specific point and try to discover something.
But there the serendipitous aspect is definitely far greater in the scenario in the app that you are mentioning.
And there are also, there is, I think it's particularly in the Netherlands, there is a company called Seeds to Meet.
It's a coworking community, so you can go there and have some space to sit and work in the cohort.
And their value proposition is also serendipity because what they offer you is serendipitous encounters with others.
They also have an app with an algorithm that will match you to people with different interests or slightly related interests.
So if you look at their websites, they're really trying to sell you serendipity.
That's really their value proposition. So their serendipity is really a feature.
If you take that out, it kind of collapses.
Yes, this reminds me of an app integration that we use in Slack, which is the random coffee bot where you can preselect some certain channels and then your availability.
And on a regular basis, it assigns you to a person to have a coffee chat with so that you also have this serendipity or serendipitous encounters as a feature there.
And like you subscribe to it and then you say like which channels and which people are like pools of people you basically want to make it select from.
And then you have also this aspect of randomization there.
So I'm not sure like how much like features play a role there. I guess not at all.
So it may be, I guess, probably random uniformly samples from these channels and then from the people in those channels that also subscribe to it.
But yet they would see something like corresponding to this.
Yeah. And I mean, this might also be maybe relevant to add serendipity and random.
It's often it's like a question like serendipity equal to random and how much randomness should there be in serendipity.
I mean, it really depends on the context and on the case you're looking at. But in general, the first person who used the term serendipity, he described it as making discoveries by accidents and sagacity.
So accidents and sagacity.
So sagacity.
So it's being, yeah, seeing things and connecting the dots. And so it's sagacity refers to a personal component.
Yeah. And I think that's very, very important to understand that there is this randomness or unexpected things or accidents happening around you.
But then to make it serendipity and to really start talking about serendipity, there also should be this sagacity from a person.
There needs to be someone who sees the relevance of what's happening.
So imagine that I'm randomly paired to someone in this slack coffee pot thing.
I mean, there is the accident that we meet.
But if none of us realizes that perhaps we have a connection and maybe we have we share a hobby and but we don't know or whatever.
Then it's hard to talk about serendipity because it's just random and there hasn't been this aha or this relevance notion.
Yeah, actually the term relevance. I mean, all of us RecSys experts are haunted by the term relevance.
But yeah, this was like popping up when I when I was processing what you what you said, like, yeah, there should be a certain amount of relevance that is also even part of serendipity or making serendipity actually valuable.
Because it would simply be random and therefore not relevant on average, then there would probably no value.
Yes and no. I'm mixing many terms right now. I know. Sorry.
I mean, of course, there needs to be a kind of relevant part, like somehow it should relate to something that you were looking for or maybe not looking for.
But you see that it connects to something that you're interested in.
But I also think that relevance can only be defined in the boundaries of what we know.
And this sounds very philosophical, but imagine that I'm using two platforms to order takeaway foods and maybe one of them is, yeah, somehow has a lot of appeal with Japanese restaurants.
So the offering of Japanese cuisine is very nice on that platform.
But maybe the other one is much more locally embedded or whatsoever.
And maybe there the traditional Belgian cuisine is more popular and there are more offers.
So maybe depending on what I'm interested in, I will use either of the two platforms to order food.
So for the Japanese, maybe they also have Belgian cuisine, but maybe I don't know that these restaurants are offering their foods there.
And since I'm only using this platform to order Japanese cuisine, it will think that for me, relevant is Japanese food.
But actually, when you look at me outside those platforms, my relevant food tastes are much broader.
So relevance is only defined in this relationship that I have with each of these platforms.
OK, so we do have these four different intentions, like the ideal, the common good, the mediator, the feature.
This provides us with the starting point for engaging in creating serendipitous encounters for users.
What's next?
So once you have established and you know why you actually want to design for serendipity, what's in it for you and for those who care about them in whatever way, then I think it's time to start thinking about, OK, and then what does this serendipity experience looks like in our context?
How will it because for me, serendipity is eventually a user experience.
It's what happens when this user with their sagacity is in this accidental environment.
How does this look like in our context, in our case?
And then we are with the experienced serendipity. That's what we how we term that.
All right. And there was a paper also in 2025 that was actually published at UMAP conference.
This was a collaboration between Fred Binst, Lien Mihielz and you.
And in this paper, you were diving into this notion of experience serendipity.
So now we talked about intended serendipity.
And so this next thing is experience serendipity. And the title was What is Serendipity?
An interview study to conceptualize experience serendipity in recommender systems.
And I guess one that I could say already is that this paper actually brought up very well-founded definition of what you could refer to as experience serendipity or even serendipity.
Am I right there or?
Yes, yes, yes.
Partially, maybe, or maybe not.
OK, so Annelien, share with us, like, what was the idea and what were the insights of this paper?
So what is experience serendipity?
Yes, experience serendipity is ultimately about what does the user describe as a serendipitous experience?
And this basically started because the three of us have been reading a lot about serendipity.
And also, especially with Lien, we're trying to make it happen through recommender systems.
But once you start looking at it, you see so many definitions, so many different ways of how to do it.
And yeah, what we were actually missing was, OK, but how do users experience it and how do they describe this?
So in that, and then Gretz did really most of most of the great work.
So he did interviews, very long interviews with users of different platforms.
So going from clothing as a Londo Pinterest music, so a very broad range of platforms.
And through these interviews, he actually just really tried to grasp and understand, OK, what is it actually that makes an experience serendipitous?
What are the necessary components of experiencing this positive feeling of serendipity?
And doing all that work, he identified three main components.
The three of them should be present for users to really experience serendipity and acknowledge this as a serendipitous encounter.
And of course, the three components are not very surprising and they resonate with in general what we what we understand about the concept.
So it's about the fact that it's a fortuitous encounter. It's something unintended or unintentional, especially.
It's something that's refreshing, which relates to novel content, unusual content of rediscovering a past interest.
And it should be enriching. So there should be this positive value to it.
So in general, I think no one is surprised by these three concepts. But more interestingly is then that actually he found that each of these three components, they can express very differently depending on the context and depending on the user.
And I think that's also why there is so much conceptual ambiguity when we talk about serendipity. It's because it really exists in all of these many flavors.
People experience it in very different ways. And often it depends on the context.
I think that's what this work really shows is that.
You could think of it as like a little box. You have different blocks. And depending on how you combine each of these three components, you get a different flavor of serendipity.
And I think that's the nice thing about showing like how it exactly relates or unfolds differently.
And as you say this, I have the paper on my screen and I could just very much encourage all the listeners to also take a look at this paper, which among all of the material we are discussing today, we will put into the show notes.
Because in the middle of that paper in table two, we can already see what you just mentioned.
So what these three main components are that taken together constitute serendipitous encounters.
So also a new term for me, fortuitous.
Yeah, we learned that we like to use different difficult words to pronounce and we circled it ourselves as well.
Like serendipity and serendipitous is often challenging, but this one is again a new thing to our list.
Yeah, actually to admit that throughout this reading of a couple of your papers, I definitely learned a couple of new words.
So it was fortuitous, refreshing and enriching.
And what I particularly liked about this table, because sometimes you can describe certain ideas with describing it with analogous ideas or metaphors or describing it taking synonyms.
But what you also did there, you actually put in the effort of describing what it is not.
And in that sense, the main components, so it should not be anticipated, it should not be boring, it should not be brain rot.
So yeah, this was very helpful. And as I'm looking at this, like you described this condition.
So every of these components has like certain conditions which could be perceived as subcomponents, right?
Yeah, so the idea is that every component should be in there, but how the component is operationalized, so to say, can differ.
So this refreshing, it can be an encounter with novel content, but it can also be something that is just very unusual to your taste and novel and unusual are not the same.
Something can be novel to you, but very much within your existing preferences.
But still it can be a novel, let's say a novel song within a genre you already know.
And I think that's the kind of nuance and the kind of detail that we can bring with this work that it's refreshing can be many, many things.
But it should be refreshing.
Yeah, you conducted these 17 interviews over a time of about half a year and then constantly refining these concepts over the course of those interviews.
And what is kind of refreshing for a RecSys paper, which I mean, it wasn't published at RecSys, but I guess it could qualify or it could be seen as a RecSys paper, is the fact that you are bringing up actual quotes from the participants of those interviews.
And I guess if we go through it, then we could also remember certain conversations among practitioners and academics at RecSys conference who are complaining about when and where a certain product didn't work out for them because the experience was simply bad, or it was good because they found certain aspects to really make them happy or what were kind of these experiences or things that participants shared here, which you took to derive this definition, this concept.
So first of all, there were many, many, many examples.
So one of the hardest thing to write this paper was to make sure that it's fitted within the page limits.
By the way, all the interviews are also shared in open access.
So the longer transcripts are publicly available for anyone interested in reading more.
So what, for example, I think two interesting examples, I think that that also helped us to differentiate some contexts is, for example, the difference that we observed between taste broadening serendipity.
So the idea where you broaden your taste, where you explore more genres versus taste deepening serendipity, where you kind of dive deeper within a particular interest you already have.
And that was amongst others because one of the participants shared a story where they were looking for shoes on Zalando.
We know there are many, many, many shoes. And all of a sudden, he got a recommendation that actually didn't really fit his style.
It was actually a type of shoes that he wouldn't buy by itself.
But it was recommended and yeah, they actually didn't look that bad.
And it was a good deal. So he decided to buy them anyway. In case he didn't like it, he could return it.
So the costs were relatively low. And actually, when he tried them on, he liked them.
They were fitting really well. And they also didn't look that bad. They liked them.
And now eventually he actually has expanded his clothing style to more also this kind of shoes and the entire look.
So that's like for us a real example of yeah, it broadened your taste quite literally.
But then on the contrary, there were also a lot of participants reporting that apps like, for example, Spotify also helped them to explore a genre, a music genre that they already are very fond of much deeper.
For example, by using this song radio where you, OK, I like this song and recommend me songs that are in the same genre.
So a lot of participants brought that up as a way to engage more deeper with that interest. And they discovered also new things.
And that's what we call taste deepening. So when you engage actually much closer with an interest that you already have.
Since you are talking about music recommendations or music platform there, I was particularly excited about this example of the Taylor Swift participant who mentioned that he came across a song of Taylor Swift and then listened to it.
And then this basically was a starting point of her whole like even avalanche of engagement with Taylor Swift.
So not just only listening to songs on the platform, but also buying a piece and finally ending up visiting a concert.
So it's really like that one spark can really trigger a huge aftermath of engagement, of user engagement with some content.
Yeah, exactly. And there was even a participant who actually said that she became vegetarian because of one documentary that was recommended to her and that inspired her, really like impacted her.
So when sometimes all it needs is like this little spark to sometimes have very, very impactful consequences of that one item you encountered.
And when I read through this overview and particularly tied to the refreshing and boring component, I was actually remembered of something additional and maybe connecting this to also some slight shameless self-promotion.
Because again, in this music domain, I found something where, yeah, you really hit the nail on its head, which was the condition.
So the subcomponent of taste reincarnation. And there I actually was unhappy with Spotify at a certain point.
And this led to a hobby project that I did in 2022 and also presented at Europython in Dublin there.
And it was my problem with my liked songs playlist, which ended up collecting more than two thousands of songs.
OK, at some point during my whole consumption, I said I like the song that much that I want it to be added to this liked songs playlist, but it's hard to like scroll through a 2000 song playlist.
So I was starting to use my GDPR data that I requested from Spotify and derive a taste profile of my own from the past couple of weeks of listening behavior and to use this to retrieve from this long list of liked songs.
The topmost like 20 or 30 that are most similar to my recent listening behavior patterns so that I could bring back these songs.
And I in the end ended up coining it the rediscover weekly.
That was a fun project. But when I was reading this paper, I was really like, yeah, taste reincarnation.
So I somehow solved it myself. At least this aspect.
Yeah, I think we should start using that example when we start talking about taste reincarnation because it really fits very well.
The idea of that serendipity can also like help you rediscover something that you had forgotten about.
I mean, it's important that you also think about that because often serendipity is about something new, something you didn't know.
But very often the interest that you had in the past might already have a very important meaning to you.
So when you then rediscover it, it's even more valuable than maybe something.
And there is also an instant positive value. Well, maybe with other when you hear a new song, maybe it takes some time before you really get into it.
But when you rediscover something, you already had started to build a connection with it.
So in the end of this whole work for me, as the one sentence takeaway from that paper was the sentence like you conceptualize experience serendipity as a user experience in which a user unintentionally encounters content that feels fortuitous, refreshing and enriching.
And there we have these three components. I mean, this has been some more recent work of yours.
But has there been some feedback or some catch up on this, possibly also by industry who really said like this helped us and this is like how we took it further to, for example, put it into practice or such thing?
Or have you seen this already like put into practice somewhere?
Not that I have been talked to about. Maybe at the conference, but I've had more discussions on that.
But I think in general, the feedback we get from people who read the work or who we talk about is like what you said, that it really helps you to unpack and to understand the different flavors of serendipity and also to make it more concrete and more specific.
And I think that's really what we should aim for, like to also look at a very specific context and see in that context how do users experience serendipity and let's go from there to work with it.
But what I think is also very important and also an important takeaway for people who want to include serendipity in their products and technologies is that it's really important that those three components should be there.
There is not a trade off to be made. Like we also discuss examples where perhaps it was an unintentional encounter and it was novel or unusual.
But when it was not enriching or when there was not really this positive aspect, then well, user said, yeah, OK, it was unintentional and unusual.
But yeah, it didn't really do the thing or especially and it most often happens with the unintentional thing when something is unusual and relevant.
But it's actually very obvious that it would show up. Then it's also not experienced serendipity.
I think one of the participants gave the example of Avatar at one point being recommended.
I don't remember the specific context, but at that point, Avatar, the movie was a big hype.
There was a lot of fuzz about it also outside the recommender. So it didn't really come as a big surprise or unexpected element that Avatar would be recommended.
The three components should really be there. And I think especially when we look at how serendipity is being measured in recommender systems, you could question if that is the case in how we try to design the metrics, because often it's the sum of some components, a metric that measures the relevance, that measures the unexpectedness.
Yeah, but maybe they level each other out. So I think that's also important to take from that that the three components should really be there for it to be experienced serendipity.
Great remark. Good to keep that in mind. So they can have different facets or be resembled through different subcomponents, but all of the three should be there.
So it's not an or, it's an and. We have already also been thinking about practical aspects, but let's focus on that a little more and also to complete this picture of intended and experienced serendipity with the aspect of afforded serendipity, which is another word that I learned throughout reading those papers, because first I was, okay, affording to afford something.
Yes, I know that word, but this was not really the meaning in that context. And then I was looking up, okay, what are affordances? And sometimes with these words, it's like they are very similarly written in German sometimes.
And then I found the word like affordance, which is this word in German. And then I said like, okay, now I start to get the idea. It's not about to afford something, but it's a totally different thing.
So let's dive into it and maybe first share with us what are affordances and what is afforded serendipity about?
Okay, so affordances actually, it's a concept that is more common in UX and design sounds.
Because it's about what the environment affords you to do with it. So let me give an example. A doorknob is a typical example that you can use, because when it's designed well, it should signal to you if you need to push the door, if you need to put down the handle to open the door, or if you should pull the door.
So affordances are characteristics of an item or of an environment or an object. And it shows actually what the user or someone else could do with it.
And it also depends on, and there again, we come back to this interaction between the environment and the user. Because affordances, it depends also on the abilities of the user.
Like if I'm a child and I'm very small and I cannot reach the door handle because I'm too small. Well, in that case, the handle does not afford me to open the door.
But for an adult who's tall enough to open the door, it does afford. So that's also why we, in serendipity, it's a broader concept that's been used affordances.
Because it really well resonates to this idea that there is this interplay between what an environment affords and eventually what a user can or decides to do with it.
So it's like this interplay of what was it again, this three aspects of an affordance?
It's the capacities of the environment. So it's the environment that shows some features for you to maybe do something with it. The capabilities of a user to actually do something with it.
And then what can be done with it is like the third aspect and it's an outcome. So it's kind of a triad relationship between the environment, the user and the eventual outcome.
And we look at serendipity as experience. Serendipity is a potential outcome when a user interacts with an environment.
All right. And there you have actually published another work in that domain that was published at the Joint Workshop on Interfaces and Human Decision Making for Recommender Systems, the INTRS workshop.
So the name of this paper that was published in 2022 was Serendipity in Recommender Systems Beyond the Algorithm, a feature repository and experimental design.
And in this paper, you actually talk about the, I would refer to it as kind of their practical means to enable serendipitous experiences.
And we also learn that they actually go far beyond the sole recommender model.
And one thing that I still remember quite vividly from this paper is that one of the many misconceptions of serendipity has also been that we have been treating it too narrowly just as the attribute of a recommender algorithm.
Can you share with us what this repository is about and how it helps to make the idea of serendipity more practical?
Yes. So in general, when I think of a recommender system, I think of not just the algorithm, and I especially think also about the user interface.
And then I also work with colleagues at other universities and they also tell me, yeah, but there is also like a big data aspect because, I mean, there is also data that you need to do your recommenders.
And I think that's the first main idea in that paper that, okay, when you think about introducing serendipity to your recommender systems, there are actually many ways in which you can do so.
And most of the work in this paper also builds on prior work by Lenape Jornebon, who has done a lot of research on serendipity and especially in libraries, because again, libraries are ideal type of serendipity.
But actually from that work, he identified three big affordances for serendipity, like design principles that in general will increase the likelihood that serendipity is happening.
And it's about the diversifiability of an environment. How diverse is it? The traversability of the environment. How easy is it to get across in the environment?
And the sense of your ability. How much is it appealing to your senses? And there are sub-affordances to make it more specific.
So actually what we did was, okay, these are these design principles for serendipity. If you want to design for serendipity, these are the principles you could be inspired by.
Let's see how they actually relate to the design of a recommender system and how we can connect them to the design decisions that you can make when you are working on a recommender system.
Again, we categorised it. As academics, we like to do that based on the content. So the actual items, what is in the recommender, the user interface, and then the information access, like how do users actually get to the information.
The idea is actually to have, and it's on our wish list, to one day have this feature repository where we have a collection of different features, be it in content interface or information access, to really have clear examples of how this can afford serendipity.
So it's like this is our catalogue of all the different ways you could think of introducing serendipity, and maybe there are more, but how we use it in practice today is to have really a set of examples, how you could design for serendipity, and work with product teams to think about how could they implement this design principle in their products.
So, for example, one of the design principles is slowability. When your environment allows people to slow down, it actually invites them also to look around. And in an urban setting, the example would be a bench.
And it's really a design principle that they do in practice, so architects, they install places to sit because people sit down, they look around, maybe they see someone, they start talking to someone.
So it really invites exploration, spontaneous interactions.
So then, for example, my question to you would be like, how could you introduce slowability, this idea of a bench, in your products? Would it make sense? Would it be something you want to do because maybe you don't want to do it?
So that's the idea of this feature repository, to have clear examples to try to make these design principles more practical and usable.
So basically, in a company, an industrial context, would agree, like, we want to pursue serendipity, and we want to have more serendipitous encounters for our users.
Then, like, what we could start with is, for example, a cross-functional workshop with designers, with our data engineers, with data scientists, applied scientists, UX designers.
And then we could basically take a look at this repository and say, okay, what are the features of this three categories?
So information access, user interface, content, and we could possibly, like, reflect on where we are and say, like, what could be possible measures, for example, to pursue slowability or to, yeah, invite users for the serendipity encounters through this affordance of slowability.
To then say, okay, these might be 10 or 20 different measures.
And then we could do, like, the classical stuff, effort and impact estimates and whatnot.
And in the end, come up with a plan like these are the top five measures that we are trying to invest into to achieve this goal.
Yes, partly, like, I would advise to first think about why do you want to design for serendipity and how do users experience it?
And then, indeed, as a third step, think about, okay, and how can we make this happen and then be inspired by the features that we list there or others, of course.
Good that you take this as an opportunity to remind, please start first with the why and, like, how users experiences.
So maybe at this point, like, what could possibly go wrong if you simply jump to solutions, simply jump to measures without having answered the why?
I think two things come to mind and they might be related, but I think the first one is that you're maybe pursuing an experience, a user experience that actually has no value for you or your stakeholders, or maybe not even in the context of your product.
Maybe you have read this amazing paper on serendipity somewhere and they have a clear approach on how to do it, but maybe they studied it in the context of a book recommender and you're building a food recommender and then books and foods.
I don't know how much overlap there is. There might be some, but so it could be just, let's say, bluntly irrelevant for your case.
Another danger, and I think that's maybe something that we don't think about often. Too much serendipity can also be very frustrating.
If I try to find something and I know I want to find a book or I want to find a movie or I want to find a song and you keep on recommending me all the interests that I have or maybe unusual things.
I might also be very frustrated with not being able to find what I was actually looking for. I think there a clear example is again from libraries. Libraries, they care about serendipity. It's an important aspect of what a library is.
And one of the things that have been found to trigger serendipity in the library is the fact that sometimes people take a book and then they just put it on a table and they don't put it back in the in the shelf on the right place.
And then there are just a few books wandering around. I mean, that could lead to serendipity because when I walk by, I can see the book and I that's interesting. I take it with me.
But if I'm the one looking for that book and the website tells me it's in the library, but it's not in the place where it should be, then it's also very frustrating for me that I can't find it. So it can be sometimes it can also be counter productive actually in a way.
So this is definitely a good example in the world of real things. If we turn our focus to the virtual world where there is not just one movie, but there's there's data and data could be shipped to several people at the very same time.
And I guess you also brought an example for this to the table for today. So do you want to share it where maybe serendipity has gone wrong?
Yes. Yes. So I think one example that really intrigues me and if anyone knows more about it, contact me and then talk to me about it. So I think in 2021.
So at that point, I was fully in PhD mode. So everything was serendipity. Netflix, Netflix launched this play something feature.
And it was I think it was in your home screen. You could just click this feature, play something and it would start a series or start a movie.
I guess the surprise me button. Yeah. So a bit later they rebranded it. They rebranded it to surprise me. I don't know why, but again, so I'm assuming it tapped into a real user need because we know there is a lot of research on that as well.
That on average people needs, I think, around 15 minutes before they start a show on Netflix.
So they need 15 minutes to browse through the catalog, which is sometimes also quite frustrating.
So I think display something feature was to help users find content more easily.
And then with the rebranding to surprise me also surprised them in a way.
But I think early 2023, so only two years after it was launched, the feature disappeared and they didn't communicate much about it, but they said that it had very low usage.
So that users didn't use it. I mean, I think that's an interesting example because in other contexts, we actually see that these kind of surprise me buttons work quite well.
Wikipedia has this surprise me button to bring on the random Wikipedia page. You could also argue that actually Spotify's discover weekly as a very similar concept.
Because the in case of Netflix, it would start a series based on your interests and a bit beyond that.
So they still thought it would be relevant. So I think it's very interesting to see why I didn't work.
So when there is not a lot of research on there on the topic, but also just this year, a study was published where they actually asked and again, which users talked about this surprise me function.
And they actually learned that when users saw this label, surprise me, they were really expecting to be surprised.
And even the recommendation or the item that started playing was not very surprising because I think in half of the cases, people actually said that they could recall that that show had been recommended to them before.
So they had seen it before. Yeah, since it wasn't really surprising, they actually really didn't like it.
So they also tested that experimentally and they really found that when people expect to be surprised and they're not surprised, their experience is way worse than actually when they were not really expecting to be surprised.
Frustration about a false promise.
Yeah, so there is this tension between expectations and what they then actually perceive, which kind of can have a negative effect apparently.
Yeah, or like a case of over promised and under delivered.
Yeah, so I think, I mean, when you think about serendipity or recommender systems, you should also think about expectation management to include there.
That's good to notice. So in many cases, but also in this, there can be too much of it, but things can also go wrong.
So it's worth doing it sensibly and like with a proper framework in mind.
And I guess for this, this brings us back to the starting point paper, which then basically weaves all these three aspects into each other.
And in that sense, I find your whole research is highly structured because you can really like go up and down in the different hierarchies and bringing us back to this intended, experienced and afforded serendipity.
And if we go back to this initial paper, you bring these three aspects into an overall framework.
And maybe can you give us a short overview of how this framework helps addressing serendipity?
For me, what it offers in practice, it's really just a way to start thinking about it.
Why do you want to design for it? And that can also already guides you in a particular direction and then narrowing down the flatora of different serendipity definitions that you could have.
And then start thinking about understanding your user, talking to your user or finding other means to figure out how they experience it.
And then also very practically think about, okay, and how are we going to do this? How are we going to design for it? And how are we going to measure it, of course?
And I think there having thought very well about what is exactly what we want to achieve and how we think users will experience it, will also help you to think about your evaluation approach.
Because, I mean, I think you should be pragmatic there. You can only measure what you can measure.
Like the example of the participants who turned into a vegetarian lifestyle because of a recommendation.
Of course, you probably are not going to able to measure that ultimate impact.
So thinking very clearly about what do we want to achieve and how and what does it mean will, I think, also help you to narrow how you do it and how you evaluate it.
In that sense, as you bring up those measurements, and I mean, this is also what we started with before we learned that this is a lot more to consider.
And that this concept goes a bit deeper than simply hunting lists of recommended items that are dissimilar from a user test profile or whatnot.
So what I mean is it seems easy to come up with metrics that possibly measure serendipity.
But in the end, it's, of course, something that we somewhere need, because even though this user study makes sense, it's, of course, quite also a very big time effort.
It's possibly expensive. And now if you want to iterate on this idea, you want to possibly use some other means of tracking whether you're on the right way or not with what you are doing.
So thinking about our typical settings of conducting offline evaluation, conducting online evaluation.
The problem that I see there is if you would simply like hand out the five core metrics to track serendipity, then everybody wrongly assumes that this is now solved if I just measure these things, but never think about the why and the how and whatnot.
But on the other hand, is there something so just thinking out loud, would a metrics repository be a promising addition to your affordance repository?
I think we could definitely start looking for what metrics do relate to or are good proxy measures for certain types of experience serendipity.
And I think that's also what we hope to do in the future is to when we have this range of different flavors of serendipity experiences, when we really understand what each of them are, what are probably good metrics to relate to each of them.
For example, we talked about this taste deepening and taste broadening.
Yeah, in the case of taste deepening, I mean, it's about a user's current preferences that they want to explore further.
So then it probably makes a lot of sense to look for metrics that look at the user profile and how related is this to the user profile.
So I think we could work towards it, but not to say that there is one serendipity metric, but I think that we could work towards metrics that are most likely good proxies for each of these different flavors.
But then again, also important to take into account, it also depends on the context.
Maybe in the case of book recommenders, this metric is a better proxy than in the case of foods or floats or whatever.
Yeah, yeah. So this is one possible direction of research in the setting of our serendipity for or within recommender systems.
Which other avenues of research tied to this do you see and which possible challenges?
I mean, we talked about a couple of challenges, but what other things are there on your research agenda that are coming up and that you want to investigate further?
So one of the things that I'm really interested about is what is the impact of the type of content that you're actually looking at?
For example, again, that's the example of movies or series. My assumption is that one of the ways that it seems to be a bit easier to include serendipity in music recommenders is because listening to a song that in the end is not really my thing.
So trying something new in the case of music is not very costly.
I mean, at most it costed me three minutes of my time and most likely after 10 seconds, I already know if this is something for me or not.
So recommending something a bit out of your usual taste is less costly than in the case of, for example, a movie because a movie I probably need like 90 minutes or more to know if I liked it.
And I mean, or at least maybe 20 minutes or I don't know how long it takes to.
And I think there is a lot of interesting questions we can think about what works, but also how does that relate to the type of content that you're actually talking about?
Maybe secondly, and there we go again to the economics and to the platform economics. I think a bit broader is not only looking at the technical aspects themselves, but often their recommender systems, they work on a platform.
One of my other research interests is on platform economics. So we could also start thinking about is your business model actually allowing for serendipity?
For example, we work with a theater, not very technical, but they included subscriptions. So now people take a subscription and they can go to, I don't know, say 10 events without going to one more or less.
It doesn't change anything to the to the price they're paying. And after a year, they really see that people attend more diverse events.
So really having a different business model apparently also invites people to explore a bit more. So I think it's also interesting to look at that.
Annelien, was there anything that we probably have missed or do you think this provides a good overview of the topic or is there something?
Maybe there is something and it circles back to the entire beginning. So it's maybe a nice close.
So I started to look at serendipity with the idea of it's a means to fight filter bubbles.
And I think after all the research that we have done, an idea that I really want to have out of the world is that serendipity bursts filter bubbles.
Okay. I really think that people can experience serendipity while they are in a filter bubble.
And it relates to the taste deepening example that we discussed. Like you can really be surprised or be pleasantly surprised by seeing things that you already knew.
We can have another discussion on that, but I think it's maybe something to think about whoever is interested in filter bubbles and serendipity.
Okay. So meaning like if I identify, I do have filter bubbles and if I also identify them as critical and whatever system platform we are talking about.
And then I want to also fight them, then just implementing. And I will say just that plainly implementing serendipity is not necessarily solving it.
So like after having implemented serendipity, if I even may say so, bluntly, you can't consider filter bubble as being successfully fought.
No. And then it brings back to, okay, why you are designing for serendipity is because you want to fight filter bubbles.
So you want to increase the diversity and then that becomes a very core design aspect in your entire serendipity journey onwards.
So I think it also illustrates why it's important to think about the why and then it triples down to everything to think about how the users experience it and how can we do it.
Definitely worth mentioning. Thanks for this reminder.
One last thing there that came to my mind often where we were talking.
I think they're really what I like about our work and also the communities that it reaches is really the fact that Brett and Dean and I, we work together.
The three of us have very different backgrounds. And I think it also helps us to talk about these issues and also try to make sure that hopefully our audiences also understand it and that it makes sense to them.
And I think that maybe we also learn them some new words sometimes. But I think there is this, you can call it an interdisciplinary collaboration.
It really, I think it makes our work to what it is. And I also definitely couldn't have done it without them and with the other collaborators that I work with.
So yeah, looking forward to the great research that we have been or not looking forward.
I mean, this is looking backwards, but I guess given all of that, we can also look forward to more research to come.
What is there already something that you are working on that you are striving for next year's RecSys?
I mean, we are now at the end of 2025 and we are possibly already looking forward to RecSys 2026.
Bringing RecSys somewhat back home, if I may say so.
What are your plans? Will you be there? Is there some topics that you want to propose, some workshops you are engaged with?
No, we are planning a submission, but it won't be about serendipity. It will be about news recommenders.
I don't think I will be there myself, but Hana, one of my other PhD students, hopefully will be there.
She's working on news recommenders. Yeah. So, yeah, there is something cooking there.
Great. Annelien, it was really nice talking to you and I am pretty sure that this will also be a very enriching episode.
Maybe one last, hopefully also a bit serendipitous question. What is the serendipity engine about?
Well, it's a research project with different universities in Flanders.
And actually a lot of the work that we have been discussing here today is part of that research project.
So it's a four year project. It ends in a year.
And we are trying to think of how we can design for serendipity and how we can keep a serendipity engine running.
So it's a collaborative project. Cool.
Then we are also definitely going to provide a link for that and then hopefully going to know more about it soonish.
For the moment, I want to express my biggest thanks. So thanks for participating.
Thanks for being my guest and sharing all of that stuff. Taking the time for it was really a great pleasure.
And I hope you enjoyed as well.
Definitely. Thank you for also going through all the papers and reading it and thinking along with me.
It was also very, very inspiring actually. Cool.
Then thank you again and have a nice rest of the day and see you soon somewhere or at RecSys.
Thank you so much for listening to this episode of RECSPERTSs, Recommender Systems Expert, the podcast that brings you the experts in recommender systems.
If you enjoyed this podcast, please subscribe to it on your favorite podcast player and please share it with anybody you think might benefit from it.
If you have questions or recommendation for an interesting expert you want to have in my show or any other suggestions, drop me a message on Twitter or send me an email.
Thank you again for listening and sharing and make sure not to miss the next episode because people who listen to this also listen to the next episode.
Goodbye.