Recsperts - Recommender Systems Experts

In episode number 14 of Recsperts we talk to Daniel Svonava, CEO and Co-Founder of Superlinked, delivering user modeling infrastructure. In his former role he was a senior software engineer and tech lead at YouTube working on ad performance prediction and pricing.

We discuss the crucial role of user modeling for recommendations and discovery. Daniel presents two examples from YouTube’s ad performance forecasting to demonstrate the bandwidth of use cases for user modeling. We also discuss sources of information that fuel user models and additional personlization tasks that benefit from it like user onboarding. We learn that the tight combination of user modeling with (near) real-time updates is key to a sound personalized user experience.

Daniel also shares with us how Superlinked provides personalization as a service beyond ecommerce-centricity. Offering personalized recommendations of items and people across various industries and use cases is what sets Superlinked apart. In the end, we also touch on the major general challenge of the RecSys community which is rebranding in order to establish a more positive image of the field.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Chapters:

• (03:35) - Introduction Daniel Svonava

• Daniel Svonava on LinkedIn
• Daniel Svonava on Twitter
• Superlinked - User Modeling Infrastructure
• The 2023 MAD (Machine Learning, Artificial Intelligence, Data Science) Landscape
• Eric Ries: The Lean Startup
• Rob Fitzpatrick: The Mom Test

Papers:

• Liu et al. (2022): Monolith: Real Time Recommendation System With Collisionless Embedding Table
• RSPapers Collection

General Links:

• Follow me on Twitter: https://twitter.com/MarcelKurovski
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/

This episode of Recsperts features Justin Basilico who is director of research and engineering at Netflix. Justin leads the team that is in charge of creating a personalized homepage. We learn more about the evolution of the Netflix recommender system from rating prediction to using deep learning, contextual multi-armed bandits and reinforcement learning to perform personalized page construction. Deep content understanding drives the creation of useful groupings of videos to be shown in a personalized homepage.

Justin and I discuss the misalignment of metrics as just one out of many elements that is making personalization still “super hard”. We hear more about the journey of deep learning for recommender systems where real usefulness comes from taking advantage of the variety of data besides pure user-item interactions, i.e. histories, content, and context. We also briefly touch on RecSysOps for detecting, predicting, diagnosing and resolving issues in a large-scale recommender systems and how it helps to alleviate item cold-start.

In the end of this episode, we talk about the company culture at Netflix. Key elements are freedom and responsibility as well as providing context instead of exerting control. We hear that being really comfortable with feedback is important for high-performance people and teams.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Chapters:

• (03:13) - Introduction Justin Basilico

• Justin Basilico on Linkedin
• Justin Basilico on Twitter
• Netflix Research Publications
• The Netflix Tech Blog
• CONSEQUENCES+REVEAL Workshop at RecSys 2022
• Learning a Personalized Homepage (Alvino et al., 2015)
• Recent Trends in Personalization at Netflix (Basilico, 2021)
• RecSysOps: Best Practices for Operating a Large-Scale Recommender System (Saberian et al., 2022)
• Netflix Fourth Quarter 2022 Earnings Interview
• No Rules Rules - Netflix and the Culture of Reinvention (Hastings et al., 2020)
• Job Posting for Netflix' Recommendation Team

Papers:

• Steck et al. (2021): Deep Learning for Recommender Systems: A Netflix Case Study
• Steck et al. (2021): Negative Interactions for Improved Collaborative Filtering: Don't go Deeper, go Higher
• More et al. (2019): Recap: Designing a more Efficient Estimator for Off-policy Evaluation in Bandits with Large Action Spaces
• Bhattacharya et al. (2022): Augmenting Netflix Search with In-Session Adapted Recommendations

General Links:

• Follow me on Twitter: https://twitter.com/MarcelKurovski
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/

In this episode of Recsperts we talk to Rishabh Mehrotra, the Director of Machine Learning at ShareChat, about users and creators in multi-stakeholder recommender systems. We learn more about users intents and needs, which brings us to the important matter of user satisfaction (and dissatisfaction). To draw conclusions about user satisfaction we have to perceive real-time user interaction data conditioned on user intents. We learn that relevance does not imply satisfaction as well as that diversity and discovery are two very different concepts.

Rishabh takes us even further on his industry research journey where we also touch on relevance, fairness and satisfaction and how to balance them towards a fair marketplace. He introduces us into the creator economy of ShareChat. We discuss the post lifecycle of items as well as the right mixture of content and behavioral signals for generating recommendations that strike a balance between revenue and retention.

In the end, we also conclude our interview with the benefits of end-to-end ownership and accountability in industrial RecSys work and how it makes people independent and effective. We receive some advice for how to grow and strive in tough job market times.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Chapters:

• (03:44) - Introduction Rishabh Mehrotra

• Rishabh Mehrotra on Linkedin
• Rishabh Mehrotra on Twitter
• Rishabh's Website

Papers:

• Mehrotra et al. (2017): Auditing Search Engines for Differential Satisfaction Across Demographics
• Mehrotra et al. (2018): Towards a Fair Marketplace: Counterfactual Evaluation of the trade-off between Relevance, Fairness & Satisfaction in Recommender Systems
• Mehrotra et al. (2019): Jointly Leveraging Intent and Interaction Signals to Predict User Satisfaction with Slate Recommendations
• Anderson et al. (2020): Algorithmic Effects on the Diversity of Consumption on Spotify
• Mehrotra et al. (2020): Bandit based Optimization of Multiple Objectives on a Music Streaming Platform
• Hansen et al. (2021): Shifting Consumption towards Diverse Content on Music Streaming Platforms
• Mehrotra (2021): Algorithmic Balancing of Familiarity, Similarity & Discovery in Music Recommendations
• Jeunen et al. (2022): Disentangling Causal Effects from Sets of Interventions in the Presence of Unobserved Confounders

General Links:

• Follow me on Twitter: https://twitter.com/LivesInAnalogia
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/

In this episode of Recsperts we talk to Flavian Vasile about the work of his team at Criteo AI Lab on personalized advertising. We learn about the different stakeholders like advertisers, publishers, and users and the role of recommender systems in this marketplace environment. We learn more about the pros and cons of click versus conversion optimization and transition to econ(omic) reco(mmendations), a new approach to model the effect of a recommendations system on the users' decision making process. Economic theory plays an important role for this conceptual shift towards better recommender systems.

In addition, we discuss generative recommenders as an approach to directly translate a user’s preference model into a textual and/or visual product recommendation. This can be used to spark product innovation and to potentially generate what users really want. Besides that, it also allows to provide recommendations from the existing item corpus.

In the end, we catch up on additional real-world challenges like two-tower models and diversity in recommendations.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Chapters:

• (02:37) - Introduction Flavian Vasile

• Flavian Vasile on LinkedIn
• Flavian Vasile on Twitter
• Modern Recommendation for Advanced Practitioners - Part I (2019)
• Modern Recommendation for Advanced Practitioners - Part II (2019)
• CONSEQUENCES+REVEAL Workshop at RecSys 2022: Causality, Counterfactuals, Sequential Decision-Making & Reinforcement Learning for Recommender Systems

Papers:

• Heymann et al. (2022): Welfare-Optimized Recommender Systems
• Samaran et al. (2021): What Users Want? WARHOL: A Generative Model for Recommendation
• Bonner et al (2018): Causal Embeddings for Recommendation
• Vasile et al. (2016): Meta-Prod2Vec: Product Embeddings Using Side-Information for Recommendation

General Links:

• Follow me on Twitter: https://twitter.com/LivesInAnalogia
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/

In episode number ten of Recsperts I welcome David Graus who is the Data Science Chapter Lead at Randstad Groep Nederland, a global leader in providing Human Resource services. We talk about the role of recommender systems in the HR domain which includes vacancy recommendations for candidates, but also generating talent recommendations for recruiters at Randstad. We also learn which biases might have an influence when using recommenders for decision support in the recruiting process as well as how Randstad mitigates them.

In this episode we learn more about another domain where recommender systems can serve humans by effective decision support: Human Resources. Here, everything is about job recommendations, matching candidates with vacancies, but also exploiting knowledge about career path to propose learning opportunities and assist with career development. David Graus leads those efforts at Randstad and has previously worked in the news recommendation domain after obtaining his PhD from the University of Amsterdam.
We discuss the most recent contribution by Randstad on mitigating bias in candidate recommender systems by introducing fairness-oriented post- and preprocessing to a recommendation pipeline. We learn that one can maintain user satisfaction while improving fairness at the same time (demographic parity measuring gender balance in this case).

David and I also touch on his engagement in co-organizing the RecSys in HR workshops since RecSys 2021.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Links from the Episode:

• David Graus on LinkedIn
• David Graus on Twitter
• David's Website
• RecSys in HR 2022: Workshop on Recommender Systems for Human Recources
• Randstad Annual Report 2021
• Talk by David Graus at Anti-Discrimination Hackaton on "Algorithmic matching, bias, and bias mitigation"

Papers:

• Arafan et al. (2022): End-to-End Bias Mitigation in Candidate Recommender Systems with Fairness Gates
• Geyik et al. (2019): Fairness-Aware Ranking in Search & Recommendation Systems with Application to LinkedIn Talent Search

General Links:

• Follow me on Twitter: https://twitter.com/LivesInAnalogia
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/

• (02:23) - Introduction David Graus

In episode number nine of Recsperts we talk with the creators of RecPack which is a new Python package for recommender systems. We discuss how Froomle provides modularized personalization for customers in the news and e-commerce sectors. I talk to Lien Michiels and Robin Verachtert who are both industrial PhD students at the University of Antwerp and who work for Froomle. We also hear about their research on filter bubbles as well as model drift along with their RecSys 2022 contributions.

In this episode we introduce RecPack as a new recommender package that is easy to use and to extend and which allows for consistent experimentation. Lien and Robin share with us how RecPack evolved, its structure as well as the problems in research and practice they intend to solve with their open source contribution.
My guests also share many insights from their work at Froomle where they focus on modularized personalization with more than 60 recommendation scenarios and how they integrate these with their customers. We touch on topics like model drift and the need for frequent retraining as well as on the tradeoffs between accuracy, cost, and timeliness in production recommender systems.

In the end we also exchange about Lien's critical reception of using the term 'filter bubble', an operationalized definition of them as well as Robin's research on model degradation and training data selection.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Links from the Episode:

• Lien Michiels on LinkedIn
• Lien Michiels on Twitter
• Robin Verachtert on LinkedIn
• RecPack on GitLab
• RecPack Documentation
• FROOMLE
• PERSPECTIVES 2022: Perspectives on the Evaluation of Recommender Systems
• PERSPECTIVES 2022: Preview on "Towards a Broader Perspective in Recommender Evaluation" by Benedikt Loepp
• 5th FAccTRec Workshop: Responsible Recommendation

Papers:

• Verachtert et al. (2022): Are We Forgetting Something? Correctly Evaluate a Recommender System With an Optimal Training Window
• Leysen and Michiels et al. (2022): What Are Filter Bubbles Really? A Review of the Conceptual and Empirical Work
• Michiels and Verachtert et al. (2022): RecPack: An(other) Experimentation Toolkit for Top-N Recommendation using Implicit Feedback Data
• Dahlgren (2021): A critical review of filter bubbles and a comparison with selective exposure

General Links:

• Follow me on Twitter: https://twitter.com/LivesInAnalogia
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/

• (03:23) - Introduction Lien Michiels

In episode number eight of Recsperts we discuss music recommender systems, the meaning of artist fairness and perspectives on recommender evaluation. I talk to Christine Bauer, who is an assistant professor at the University of Utrecht and co-organizer of the PERSPECTIVES workshop. Her research deals with context-aware recommender systems as well as the role of fairness in the music domain. Christine published work at many conferences like CHI, CHIIR, ICIS, and WWW.

In this episode we talk about the specifics of recommenders in the music streaming domain. In particular, we discuss the interests of different stakeholders, like users, the platform, or artists. Christine Bauer presents insights from her research on fairness with respect to the representation of artists and their interests. We talk about gender imbalance and how recommender systems could serve as a tool to counteract existing imbalances instead of reinforcing them, for example with simulations and reranking. In addition, we talk about the lack of multi-method evaluation and how open datasets incline researchers to focus too much on offline evaluation. In contrast, Christine argues for more user studies and online evaluation.

We wrap up with some final remarks on context-aware recommender systems and the potential of sensor data for improving context-aware personalization.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Links from the Episode:

• Website of Christine Bauer
• Christine Bauer on LinkedIn
• Christine Bauer on Twitter
• PERSPECTIVES 2022: Perspectives on the Evaluation of Recommender Systems
• 5th FAccTRec Workshop: Responsible Recommendation

Papers:

• Ferraro et al. (2021): What is fair? Exploring the artists' perspective on the fairness of music streaming platforms
• Ferraro et. al (2021): Break the Loop: Gender Imbalance in Music Recommenders
• Jannach et al. (2020): Escaping the McNamara Fallacy: Towards More Impactful Recommender Systems Research
• Bauer et al. (2015): Designing a Music-controlled Running Application: a Sports Science and Psychological Perspective
• Dey et al. (2000): Towards a Better Understanding of Context and Context-Awareness

General Links:

• Follow me on Twitter: https://twitter.com/LivesInAnalogia
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/

• (03:18) - Introducing Christine Bauer

In episode number seven, we meet Jacopo Tagliabue and discuss behavioral testing for recommender systems and experiences from ecommerce. Before Jacopo became the director of artificial intelligence at Coveo, he had founded tooso, which was later acquired by Coveo. Jacopo holds a PhD in cognitive intelligence and made many contributions to conferences like SIGIR, WWW, or RecSys. In addition, he serves as adjunct professor at NYU.

In this episode we introduce behavioral testing for recommender systems and the corresponding framework RecList that was created by Jacopo and his co-authors. Behavioral testing goes beyond pure retrieval accuracy metrics and tries to uncover unintended behavior of recommender models. RecList is an adaption of CheckList that applies behavioral testing to NLP and which was proposed by Microsoft some time ago. RecList comes with an open-source framework with ready set datasets for different recommender use-cases like similar, sequence-based and complementary item recommendations. Furthermore, it offers some sample tests to make it easier for newcomers to get started with behavioral testing. We also briefly touch on the upcoming CIKM data challenge that is going to focus on the evaluation of recommender systems.

In the end of this episode Jacopo also shares his insights from years of building and using diverse ML Ops tools and talk about what he refers to as the "post-modern stack".

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Links from the Episode:

• Jacopo Tagliabue on LinkedIn
• GitHub: RecList
• CIKM RecEval Analyticup 2022 (sign up!)
• GitHub: You Don't Need a Bigger Boat - end-to-end (Metaflow-based) implementation of an intent prediction (and session recommendation) flow
• Coveo SIGIR eCOM 2021 Data Challenge Dataset
• Blogposts: The Post-Modern Stack - Joining the modern data stack with the modern ML stack
• TensorFlow Recommenders
• TorchRec
• NVIDIA Merlin
• Recommenders (by Microsoft)
• recbole

Papers:

• Chia et al. (2022): Beyond NDCG: behavioral testing of recommender systems with RecList
• Ribeiro et al. (2020): Beyond Accuracy: Behavioral Testing of NLP models with CheckList
• Bianchi et al. (2020): Fantastic Embeddings and How to Align Them: Zero-Shot Inference in a Multi-Shop Scenario

General Links:

• Follow me on Twitter: https://twitter.com/LivesInAnalogia
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/

In episode number six, we welcome Manel Slokom to the show and talk about purpose-aware privacy-preserving data for recommender systems. Manel is a 4th year PhD student at Delft University of Technology. For three years in a row she served as student volunteer at RecSys - before becoming student volunteer co-chair herself in 2021. Besides working on privacy and fairness, she also dedicates herself to simulation and in particular synthetic data for recommender systems - also co-organizing the 1st SimuRec Workshop as part of RecSys 2021.

This episode is definitely worth a longer run. Manel and I discussed fairness and privacy in recommender systems and how ratings can leak signals about sensitive personal information. For example, classifiers may exploit ratings in order to effectively determine one's gender. She explains "Personalized Blurring", which is the approach she developed to personalize gender obfuscation in user rating data, as well as how this can contribute to more diverse recommendations.
In our discussion, we also touch "data-centric AI", a term recently formulated by Andrew Ng, and how adapting feedback data may yield underestimated effects on recommendations that can lead to "data-centric recommender systems". In addition, we dived into the differences between simulated and synthetic data which brought us to the SimuRec workshop that she co-organized as part of RecSys 2021.

Finally, Manel provides some recommendations for young researcher to become active RecSys community members and benefit from exchange: talk to people and volunteer at RecSys.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Links from the Episode:

• Manel on Twitter
• Manel on LinkedIn
• Manel at TU Delft (find more papers referenced there)
• SimuRec Workshop at RecSys 2021
• FAccTrec Workshop at RecSys 2021
• Andrew Ng: Unbiggen AI (from IEEE Spectrum)

Papers:

• Slokom et al. (2021): Towards user-oriented privacy for recommender system data: A personalization-based approach to gender obfuscation for user profiles
• Weinsberg et al. (2012): BlurMe: Inferring and Obfuscating User Gender Based on Ratings
• Ekstrand et al. (2018): All The Cool Kids, How Do They Fit In?: Popularity and Demographic Biases in Recommender Evaluation and Effectiveness
• Slokom et al. (2018): Comparing recommender systems using synthetic data
• Burke et al. (2018): Synthetic Attribute Data for Evaluating Consumer-side Fairness
• Burke et al. (2005): Identifying Attack Models for Secure Recommendation
• Narayanan et al. (2008): Robust De-anonymization of Large Sparse Datasets

General Links:

• Follow me on Twitter: https://twitter.com/LivesInAnalogia
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/

In episode five my guest is Zeno Gantner, who is a principal applied scientist at Zalando. Zeno obtained his PhD from the University of Hildesheim where he was investigating ML-based recommender systems. As a principal applied scientist he is responsible for strategy, mentoring and setting standards for different initiatives on fashion recommendations impacting over 48 million customers in Europe.

We discuss the ramifications and limitations of positive-only implicit feedback, touch on how reinforcement learning and more rating-like feedback can help as well as how to treat multiple feedback levels. In the main part, we turn our focus towards fashion recommendations and the “usual suspects” of typical e-commerce recommender systems.  We also discuss the goal of creating more fashion-specific recommendations and making users come back for inspiration. This involves a lot of domain-specific modeling and design of experiences to cater the needs for various user segments: from fashionistas to pragmatic customers. This also involves putting users into the “driver seat” of recommenders as well as understanding how to achieve long-term customer satisfaction.

Finally, we briefly touch on the topic of size and fit recommendations and finish with an outlook on the future developments leading to fashion recommendations becoming its own subfield within the recommender systems space.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Links from this Episode:

• Preferably reach out to Zeno Gantner via email (find his address mentioned by the end of the episode)
• Fashion DNA by Zalando Research (Paper)
• Fashion MNIST (image dataset)
• Workshop on Recommender Systems in Fashion 2021
• RecSys Challenge 2022 on Session-based Fashion Item Recommendation by Dressipi
• H&M Personalized Fashion Recommendation Challenge on Kaggle
• Spotify: A Product Story - Episode 4: Human vs Machine
• Dataset for trivago RecSys Challenge 2019
• RecSys 2020: Tutorial on Conversational Recommender Systems

Papers:

• Rendle et al. (2009): Bayesian Personalized Ranking from Implicit Feedback (2009)
• Loni et al. (2016): Bayesian Personalized Ranking with Multi-Channel User Feedback
• Sheikh et al. (2019): A Deep Learning System for Predicting Size and Fit in Fashion E-Commerce
• Wilhelm et al. (2018): Practical Diversified Recommendations on YouTube with Determinantal Point Processes

General Links:

• Follow me on Twitter: https://twitter.com/LivesInAnalogia
• Send me your comments, questions and suggestions to [email protected]
• Podcast Website: https://www.recsperts.com/