#25: RecSys 2024 Special

Hello and welcome to RECSPIRTS, a recommended systems experts. Today's episode is going to be slightly different from all of our previous episodes. And it's due to the fact that we are looking forward to recommender systems conference short rexas taking place next week in Vari, Italy, starting Monday, October 14th and concluding on Friday, October 18th. For this episode I have decided to provide a little bit of a special where I quickly walk you through some key facts about the upcoming rexas. And I've also invited a guest to the show. However this time's guest is a person that you possibly know from one of my previous episodes. We are going to quickly talk about why this person is going to rexas, what this person is most excited about, and just have a quick chat on the upcoming rexas next week in Vari. So stay tuned for this special episode of RECOMENDERSystemsXpert to know all that you need to know for the rexas next week in Vari in Italy. And there's also one special thing. If we meet each other, please reach out, let's have a chat and I have also some nice rexas for you that I'm very eager to share and if you want it you can get it. So without further ado, let's take a look what next week's rexas is all about. Maybe as a bit of background. The first time rexas took place was actually 2007 in Minneapolis, Minnesota, USA. So the rexas that we are looking forward to take place next week is actually the 18th installment of the ACM conference on RECOMENDERSystems. This time taking place in the beautiful lovely country Italy and actually in South Italy so it will be pretty warm, pretty sunny and we are expecting a very packed program presenting research in industry and academia for the whole next week. This is also becoming a rather short episode so just three things I will quickly talk about the program and the venue and provide some stats and figures. So the conference is taking place from Monday until Friday and there will be different conference venues so there is the Polytechnical DeBari and there is actually the Petrucelli theater as well as the Chamber of Commerce which are both located next to each other. On Monday and Friday we will have tutorials and workshops. On Monday we will also be having the doctoral symposium. All of these, so the tutorials, workshops and the doctoral symposium on Monday and Friday are taking place at Polytechnical DeBari and the main conference which takes place from Tuesday until Thursday is at the main conference venue which is the Petrucelli theater as well as the Chamber of Commerce. As for the program, there is a bunch of tutorials and workshops. There are six tutorials, there is a total of 21 workshops taking place that are spread across Monday and Friday so there is plenty of stuff that you could get hyped about. And it's a pretty early bird conference given that we are in Italy times are definitely not Italian once I would say and as we will also talk about in our quick interview with my guest you should not miss out on all the opportunities to reach out to people, talk and discuss with them. The main conference is going to be opened on Tuesday at 9 a.m. with the opening remarks and continues through Wednesday and Thursday. Every main conference they will feature a keynote speech along with sessions that are filled with paper presentations around certain topics like large language models, bias and fairness, collaborative filtering, multitask learning, cold start, sequential recommendation, graph learning, optimization and evaluation, robust rex's of policy learning as well as women in rex's. The sessions of this single track conference are not the only thing that you shouldn't miss out on. There is also a lot of social events going on around the conference. Everything starts on Monday evening after a whole day of tutorials and workshops. The day is going to close with a welcome reception that is also taking place at Polytechnical DiBari and starting at 7 p.m. in the evening. And first main conference day so Tuesday there won't be any specific events planned in the evening. However, there is plenty of room to reach out to people, have a chat, have a drink and of course also have some proper dinner and explore the nice city of Bari. On Wednesday there is the social dinner that takes place at the Villa Rumanazi Carducci and is taking place at 8 p.m. And on the last conference day for those who were lucky enough to obtain a ticket, there is the Erasmus Orchestra concert in the very same theater the conference is taking place. So the Petr Celi theater. On the last day of the rex's so Friday there aren't any special events in the evening. However, also again as for Tuesday everybody is free and so you will definitely find some nice things to do with the people and apart from those official rex's events there will always be Karaoke so watch out for it you surely don't want to miss out on that. So let's continue with a couple of statistics regarding the number of papers also the acceptance rates. Just a couple of days ago on October 8 the proceedings of the rex's were published you also find a link in the show notes for easy access. I'm going to cover just a couple of details that you can find in the proceedings about the accepted contributions but also to get you hyped about what you can expect from the conference whether you are on site or whether you are attending virtually. So let's get started. Overall there are 97 full and short papers. 14 reproducibility papers. 11 industry talks and 23 industry posters. Next demonstrations 18 late breaking results papers in terms of the number of valid submitted papers. We saw 266 full papers and 174 short papers that were submitted and considered valid among those 58 out of the 266 full papers were accepted and 39 out of 174 short papers were accepted. This results in a 22% overall acceptance rate which is slightly above the long term average of 20%. As for the industry submissions we see contributions from Amazon, Criteria, Google, Netflix, Pinterest and others. If we look into the industry papers then there's an amount of 34 out of 73 submitted proposals which results in acceptance rate of 46% and those are divided into 11 talks and 23 industry posters. And right afterwards our interview on how my guests are looking forward to Rex's 2020 for in Bari. And again as I said meet me at the conference and I will have some nice Rex birds swank for you. Yeah, with that. I have with me Kim Falk now on this very early morning two days before Rex is going to start. Dear listeners you can be very lucky about the fact that this is not a video podcast because we both look quite tired. At least I have my coffee in front of me. It looks good but I haven't taken a sip right now. So I'm also not caffeinated properly right now, so please apologize for that. And I guess unfortunately the very same applies to Kim, who is or who has made it already to Rome, right? Yes, I'm in Rome in an apartment, sadly without coffee, but as soon as we finish here, I'm heading down to one of the great many coffee bars in Rome to have a good cover chain. Okay, so you all here like we also do have a little bit of an incentive to make this not too long. And as I already mentioned in the introduction, this is going to be a rather short episode or just a short preparation episode to get you hyped about Rex's. So yeah, Kim, you were my first guest on Rex's words. I guess, thank you. It was also an honor for me to talk to you and also to have had the chance to work with you in the past. I guess also plenty of people who might be listening to this podcast are also quite eagerly expecting the second edition of your book practical recommender systems. But I guess we will come to that question in the end. As Rome won't be the goal of your journey, but it will rather be Barry and the Rex's tell us maybe quickly about yourself and why you go to Rex's. So I have been working with recommender systems since around 2010 and in that period, of course you can learn a lot by simply working with recommender systems and during it in industry and in practice. But if you really want to understand recommender systems, I feel that Rex is the premier place to go and get those vibes. I think one of the important things that a lot of people out there working on recommender systems. So it's one big, deep come together to talk about recommender systems. And it's great because you don't have to explain everybody what collaborative filtering are before you start actually talking about the things you find is interesting. So Rex is just the greatest place to go. If you like recommender systems and want to learn both for beginners because there are a lot of people you can talk to and talk about what issues you have, but also as you get more advanced, there are always new ideas, new questions to ask that people will have asked in the presentations, but also when you when you meet them and talk with them. So you should like recommender systems, you should definitely go to Rex's. What you say only if people like recommender systems or what about all the other activities that make Rex is so special. Of course you should go to body because body it's a it's an Italy, it's a beautiful place. I think you should go. Have you been to body before? No, but that's what people are telling me. Okay, okay, that counts. Yeah, but all of easily is beautiful. So why not body? Good point. Yeah, of course it's important with all the research and the industry is coming together, but it's also very social events where you meet a lot of people. Of course people that's as I said that you work you can come to work with that has the same type of experience as you and you can discuss that with them. But of course it's also a place to create new concepts. We do have this fancy banquet. We do have the reception on Monday. There's actually opera on Thursday, which I actually didn't manage to get a ticket for. No, we need a. Okay, so we will be the ones who are asking whether some people might have gotten too many tickets. So let's see. So this is now the official call out. Kim Falk and Masayak-Roski are still in search for two tickets for the opera on Thursday evening. So everybody who is hearing that please reach out and do something for the community. Okay, yeah, Kim, it's actually not your first rex's obviously and you have been to too many of them. So we met also there a couple of times that I also still remember rex's in Amsterdam pretty well or actually my first one in Como. Given all these rex's that you have attended, what were kind of your most memorable rex's moments or your most memorable rex's moment. I heard that there are some that you can or are willing to share and some that you are not. So maybe the one that you are willing to share. I think each rex's each year has brought its own set of experiences. When you asked me this question yesterday, one of the first things that I thought about was actually YouTube presenting their multi states recommend that which has something that has diffused into basically every recommender system since then. I was industry chair for two years in 2019 and 20 if memory serves me great, which was a great experience to stand on the states in front of so many people and not actually be the one who's presenting anything and still try to be interesting. It was really interesting when the one in Rio that was 21 because of the COVID kind of went from being this huge conference that everybody wanted to go to because it was in Rio and then it went online instead because of the pandemic. It was a great experience to have the chance to share with so many intelligent people sitting home in my own living room also. But a lot of the things that you interact with every day like Netflix, like Amazon, I met most of the people who has provided the ideas and the engineering to actually make those things happen. And so if I should say the greatest thing about Rex is that I met all my recommender system heroes and talk to them and in some way I feel I also discussed things with them that they brought back home and hopefully also included in their product. That sounds great. All the Rex's heroes and then having also some kind of a say or even maybe some influence or ideation that then as you said makes it also into those products. No, that sounds definitely cool. I wonder a bit why you didn't mention Karaoke. What's your take on Karaoke at Rex's? Yeah, so I love Karaoke because it's a great evening where everybody goes out and and and you see all these intelligent people in a very different way. And a lot of them are actually blessed with a musical sense. Certainly I'm not one of them. So it's also a bit scary to go to the karaoke sessions. I think people are still laughing about my performance with Morton in Amsterdam. And so those kind of experiences I will never forget them for sure, but maybe I don't want to repeat. I'm still not aware of the specifics of how easily I could add some some music snippets to this podcast so I would rather not do that. That was definitely awesome. We got a lot of the force afterwards, but I'm not sure it was due to the quality of the singing. But therefore it was great entertainment and it was people had a lot of fun. And on the social side of things I should also mention that there is the Rex's running club. So at every Rex's we go we meet up at every Rex's it's a new time and it's always difficult to figure out when it is actually, but it's very nice to go for a run with the Rex's group. I hope that I can make it this time because I'm really like into running and as I already mentioned before so today is Saturday the day we are recording this and hopefully also already directly releasing it. And I'm almost also on my way to Barry but with a short stop in between actually in Munich where the half marathon is taking place tomorrow. I'm running that one before then taking the flight to Barry and unfortunately even though that I'm a lot into running I've never made it to one of those gatherings for the runs in the morning even though I also heard that they are quite popular. And I have all my stuff with me so I should definitely make it at least once but I have also I mean taking a look at the schedule already and I would say for a thousand Europe it's pretty tough. Yeah so I mean on Monday we are starting at nine and registration even opens at eight and if I look at the main conference days and I see the conference starting I mean everything opens around eight a.m. as well. But then we start on nine a.m. on Tuesday 830 on Wednesday and on nine or for the women's breakfast at eight even on Thursday and Friday I guess we also started nine so always starting at 839 is I would say pretty tough because you need to get to the venue. want to do your own, you possibly also want to have some breakfast. So I guess that's going to be tough given also the gatherings and having a wine, chatting, discussing that takes place in the evening. So I guess it's definitely becoming a week that will drown a bit everybody's batteries, but I guess that will be worth it. It's a very tough week and you come home quite destroyed, at least I do, but I'm also trying to burn the candles in every possible end while I'm there. Okay, that's that's fun. So I guess that you also have already taken a look at the schedule and what might be possible or not possibly what is going to be presented. I assume that you are engaging also for the four weeks, so going all in on the main conference tutorials and workshops, is it right? It is true. So anything that you could recommend for the conference or something, what are your top picks for the conference? Maybe talking about tutorials and workshops and then maybe let's have a short look at the accepted contributions at the papers actually, whether there is something specifically of interest for you. So what are you going to pick or to visit? So on Monday, not on Monday, on Monday, I'm going to go to the consequences workshop. There are a lot of interesting workshops, so it's it's a difficult choice. But I mean this this is the only one that has featured a Nobel laureate in the past. Yes, that is true. The thing about recommended systems is that that a lot of the research being done has to be done offline. And so consequences is one of those workshops where they're really talking about how is the you handle recommended systems as a learning problem and as an online problem or and I think that's knowledge that is still not very accessible for the places. So I think consequences is definitely one that I'm going to. And I think that I would find very interesting is it's also to go through the tutorial, the economics of recommended systems. And without actually knowing what the tutorial is about. But one of the issues that you often have with recommended systems and industry is that that you go you go to Rexus for example and then you see these peak models that they have come up with and this is the state of yarn. And then it turns out that the the economy of having one of those running and the improvements that they actually do. It's expensive to run a model like that and train it and what you earn from it is very little. So it's important to also consider how does it those kind of factors when you create a recommended system. The tutorial is deep recommendations and recommendations using graphs, also sounds interesting just because it has so many of those interesting words put in one sense. And then yeah I think I could find all the tutorials interesting. The human sense of Rexus, everything is about the humans actually without them the recommended systems don't really do anything. So yeah, cars, contexts that were recommended systems which I also believe is some of the future most important points that you should look into. So anyway, I think everything is interesting. Yeah, but unfortunately you need to prioritize and make a choice because most of these workshops will take place at the very same time. I mean you might be a bit lucky with the tutorial. So deep recommendation using graphs is taking place in the morning and then the afternoon you do have the economics of recommended systems. But I mean at the same time you do have all those great workshops and especially as a consequences covers the whole day. So maybe it will be rather some switching rooms and checking out certain parts of those workshops or tutorials possibly. But yeah, you're right. There's so much stuff going on. Yeah, 21 workshops. I'm not even sure if we ever had so many workshops at Rexus taking place and it feels like it was growing from year to year. You are not only attending or watching other people's work or getting as would say new inspiration. You're also giving a talk yourself, is it right? Yeah, so come on, advertise a bit. Yeah, I'm giving a lightning talk on Friday at the El Rexus workshop. The idea with this workshop is to look at the alternative, unexpected and critical ideas in recommended systems. And so my little talk, it's more like a rant actually, I think. Because. Oh, that's the greatest once, yeah. I had the possibility to work with a Danish company who's making user experiences that helps people understand what kind of products they will buy in different types of domains. And the domain that we have been looking into with them is to buy six toys. And to create a good recommender system, we thought that since a large language model can do everything, we can use those ones to create embeddings where we can create a similarity model that would actually find similar products. But it turns out that it doesn't actually understand what a six toy is. And there are a lot of different features and that it needs to understand that it just doesn't. And I think that that trait with the fact that it's a small Scandinavian language makes it very clear that it can't do everything as people really wanted to do. And so my point in my lightning talk, there are three points. One is that you should be very careful trying to use the embeddings of an element outside of what is, I'll say, common knowledge areas. You should be very careful when you use it on languages that are not English. It's really great in many things and it also seems like it's understanding Danish a lot. But you should just be very careful because often when you train language models on languages like the Scandinavian languages, it's become some kind of soup of all of them mixed together. And which means that a lot of the Danish words that are similar to Swedish words but mean different things becomes confusing for it. And the last point I want to make in my four minute or two minute talk is that people should stop expecting that it can do miracles in those kind of domains because it creates unrealistic expectations on people on the Rex's engineers or the engineers that has to work with them. So that's my contribution this year. I had co-authored other things but sadly we were not accepted. Okay. But we will have the lightning talk so you surely don't want to miss out on that. You said it's on Friday as part of the old or old Rex's standing for alternative Rex's if I'm not mistaken. It's also a workshop that is new that is taking place for the very first time if I'm not mistaken. Great. So it's taking place on Friday afternoon. I'm just just pulled up the page of it. So quarter past two to quarter to six. Also at poorly technical debati. So much maybe to the program regarding workshops and tutorials. It's a favorite paper that you have in mind or that you are very eagerly looking forward to see being presented at Rex's. And I'm any again back to the offline online aspects of things. The one that I just have seen so far is the multi objective recommendations via multivariate policy learning by Orinia, June and from shared sets that I'm interested in hearing what they're going to talk about because as soon as you put multi into anything, then it becomes confusing. So Oh yeah. And I mean, even the part before that to ask your stakeholders or people how they might be able to quantify what they actually want to put that into an algorithm to optimize for something. Yes. Okay. Yeah, I guess he was pretty productive. So there are many papers that he was main author or that he co-authored. And this is I guess one of that bunch. Anything else or any other paper or something that that you're excited about. There is the keynote talk from Spotify that I'm also looking forward to here. Oh, by Muneo on Thursday. Yes, exactly. I think that she has been an important person in the Rex's environment. So I'm very curious to is the first time I hear her talk. So I'm looking forward to that. I think the key notes this year is I'm looking forward to hear all of them, but especially her. So one of the things that people should consider when they go to Rex's is that you need to be selected because if you want to learn everything that is on the program, I downloaded the proceedings the other day. And it's what is it a thousand five hundred pages of research articles. And so if how many research articles can you learn in one day? This is that if you if you look at the actual paper page number, then that's insane. So try and find things in a sense that go to everything but try to be selective when you're really trying to turn on your brain and listen efficiently because I think that if you try to listen to everything then you just not remember anything afterwards. Yeah, that's a very, very good recommendation. I think maybe also take the chance at least consider it not to force yourself attending every session, but also take the opportunity to reach out to people and have discussions because as far as I know all of these sessions are going to be recorded and let's plan to release them to the Rex's YouTube space. So there's plenty of time to look at the recordings if you haven't attended them or if you feel like some strong FOMO kicking in. So this is definitely something that you can't have on record or can't go back to because this will only be taking place at Rex's. It's an impressive, but there will also be opportunities to do so virtually. So I would say rather see the opportunity to talk to people instead of forcing yourself to follow every session and every talk. I think that's the most important. Of course you can tell that to your employers, but the most important thing that Rex is excited to meet people and discuss the problems you have. Compared to psychology sessions, the unique people that understands your problems. That's a great way of putting it. Yeah. So get out of your rabbit hole and I'm sure that no matter how complex the problem that you're working on right now, there are at least 50 other people at the conference that are working on exactly the same. And so if you can find those ones then without revealing any big business secrets or research studies, then you can still get inspiration and brainstorming based on the internet's important. Yeah. It's Saturday morning and I feel like I do have the need for another coffee or at least finish packing my luggage so that I could also get on my way to Barry. I mean, you are already pretty close, but you also aren't there yet. I'm not mistaken. You plan to arrive there by tonight. It's great. I mean, and it's already like you're not the first person who is arriving there. So there is the summer school taking place this week already. So a couple of folks are already on the ground. And most people are then expected to flock in by the beginning of next. Yeah. So plenty of opportunity to also for all of those who are then already listening. So if you're out to folks and maybe already make the best out of your Saturday or Sunday, if you're not too sleepy, like the both of us. It does. It does. And I mean, we are in Italy. So there should be plenty of good coffee and I've already also seen like due to the fact that there won't be any coffee provided the conference venue. I'm sure that the weather or the chamber of commerce, not 100% sure, but there is a kind of a game around this to explore the best coffee places nearby. So I'm going to be excited about that. And I'm pretty sure that we will find lots of good coffee there. I'm sure about that. I should say before we say goodbye that you introduce me as writing on the second edition of my book. It's not going to be a second edition. It's going to be a new book. And the working title right now is it's going to be called modern recommender systems. So the traditional part of of recommender systems where everything is split into rather filtering and content based filtering is kind of downplayed a little bit. And then I'm writing about the many new aspects also. It should still be accessible for new people to get up to speed on what's happening with recommender systems, but it should be the more modern aspects of it. It's not all about last language models, but they will of course have an important part in the book too. But that sounds like like a great opportunity. So in that sense, it will be a bit of introduction into recommender system. So that's still, as you said, accessible for beginners. So that's what I wanted to start it from a practical point of view. But then also taking into account like more recent developments and I would assume more like former industrial or application oriented point of view what has shown to be successful or so possibly a lot about also like two tower model sequence model that's all that stuff. Yes, there will be a chapter of sequence based models also. I'm not so much in love with those models as many people as a research problem sequence based models are great and in a few domains, but I see them applied in places where I think they create more confusion than actually provide quality recommendations. Okay, fair point. So it will be a full in new book and called modern recommender systems. I have to admit the first moment you said that I was actually hearing board in recommender systems. So you could have also made a book about more than recommender systems, but it will be modern recommender systems. Sorry, more if you're hearing this. So, but the main question given all the works that you have on your plate right now, when do you think we can expect that book to appear? There are different points of views and that because many actually already send out the request for reviewers for the first three chapters. And they are close, but they're not ready yet. And so the idea is, I say, when the first three comes out as early release, then within a year, the rest of the books should be finished. So the hope was actually that it had been out as early release before this Rexus. And then it would be in pre-production by the next Rexus. So I'm counting in Rexus's. Great, great. Okay. I guess for the moment, I'm the only person that is standing between you and a proper decent Italian coffee. So let's remove that piece out of that equation or that streak. And I would say thanks for being on Rexbirds again. On the list of these 25 episodes being my very first guest and now also my guest for this special episode. And I'm very much looking forward to meet you in person again, then on the ground in Barry for a coffee for wine and for some proper good chat and geeking about recommender systems. Thank you. I think in that process, I would say that Rexbirds is great podcast to listen to while you're traveling to Bali because travel times are a bit long for many people. And secondly, you should not forget that Italy also creates great beers by now. I have one of my fidgets from South Tyrol. Alright, but they don't consider themselves as Italian citizens though, but not as Italy. Yeah, that's a nice way of saying it. Yes, but I look forward to seeing you too. See you at Barry Kim. Bye. Thank you so much for listening to this episode of Rexbirds, recommender systems experts, the podcast that brings you the experts and recommender systems. If you enjoy 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 recommendations for an interesting expert you want to have a 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. [Music] [MUSIC]