Recsperts - Recommender Systems Experts

In episode 34 of Recsperts, I'm joined by Raghav Saboo, Staff Machine Learning Engineer at DoorDash and Tech Lead for Personalization and Search for New Verticals — groceries, convenience, retail, alcohol, pet supplies and more, beyond the original restaurant vertical. We discuss the particular challenges of personalized recommendation, ranking and search in quick commerce, recent trends in generative recommendations and the application of Semantic IDs to item ranking and query reformulation. Raghav's path into recommender systems started in chemical engineering before he moved into ML consulting, a Master's in Statistics, Machine Learning and Econometrics from Duke University, and building LLMs for new language launches on Amazon's Alexa AI, ahead of joining DoorDash.
We start with the marketplace itself: DoorDash connects consumers, merchants and couriers, and growing it means balancing the interests of all three so that the platform stays healthy for everyone on it. Raghav walks me through how his team frames the consumer side around three pillars — familiarity (surfacing what a consumer already trusts), affordability (matching price sensitivity and timely deals) and novelty (introducing new items and categories without adding friction). From there we get into how DoorDash uses LLMs to build "memory blocks," structured natural-language representations of a consumer organized around semantic domains like dietary preference, pet ownership or trusted brands, and how these feed LLM-generated collections that get resolved into real items through embedding-based retrieval.
We then turn to DoorDash's move to generative approaches, centered on Semantic IDs: hierarchical product identifiers learned through recursive clustering of item content embeddings, forming a taxonomy that captures attributes a human-built catalog structure might miss — as Raghav puts it, "within e-commerce, items really carry a lot of meaning." He walks me through two production use cases: replacing dozens of taxonomy-based dense features in the ranking model with Semantic ID n-gram aggregations while improving online metrics, and using Semantic IDs for query reformulation in search, letting the system traverse a learned hierarchy to refine or diversify a query. This connects to DoorDash's own paper on the topic and to a broader conversation about why search, recommendation and agentic ordering — DoorDash's own "Ask DoorDash" — are converging on a shared substrate of Semantic IDs and consumer memory, while today's app surfaces still need to grow more flexible for that convergence to feel seamless.
We close with a preview of the RecSys 2026 tutorial "Recommender Systems in Delivery Platforms: Challenges, Solutions and Learnings," which Raghav is co-presenting with Wolt's Paavo Camps and myself, and his advice for navigating a field that reinvents itself every quarter: be honest about whether that pace suits you, use AI agents to filter what's worth your attention, and build the judgment to recognize dead ends early.
Enjoy this enriching episode of RECSPERTS – Recommender Systems Experts.
Don't forget to follow the podcast and please leave a review.

(00:00) - Introduction

(02:19) - RecSys 2026 Tutorial Preview

(03:23) - About Raghav Saboo

(10:47) - Working on Amazon Alexa AI

(14:13) - About DoorDash

(16:31) - Operating Model of a Multi-Sided Marketplace

(20:45) - Affordability, Familiarity and Novelty

(34:01) - Advantages of LLM-based Consumer and Item Profiles

(46:53) - Generative Recommendations

(59:21) - Semantic IDs for Item Ranking and Query Reformulation

(01:19:49) - Agentic Shopping vs. Conversational RecSys

(01:29:39) - Tutorial on Recommender Systems in Delivery Platforms

(01:34:05) - Closing Remarks

Click here to view the episode transcript.

Links from the Episode:
Raghav Saboo on LinkedIn
Raghav Saboo's Substack
Using LLMs to Infer Grocery Preferences from Restaurant Orders
Building Ask DoorDash (Part 2): Intelligence
Bridging Affordability, Familiarity, and Novelty (KDD 2025)
Building a Unified Consumer Memory for Personalization at Scale
Offline LLMs, Online Personalization: Generating Carousels at DoorDash
RecSys 2026 Tutorials
Workshop on Unified Search and Recommendation (USRW) 2026
Papers:
Xu et al. (2026): One Hierarchy, Two Systems - Semantic Product IDs for Discovery-Surface Ranking and Search-Page Query Reformulation (RecSys 2026, USRW Workshop)
Xi et al. (2026): Mine and Refine - Optimizing Graded Relevance in E-commerce Semantic Search Retrieval (CIKM 2026, Applied Research Track)
Chen et al. (2026): Joint Optimization of Relevance and Engagement in Multi-Task Ranking for E-Commerce with Efficient LLM Supervision (SIGIR Industry Track)
Sinha et al. (2025): Mind the Gap - Bridging Behavioral Silos with LLMs in Multi-Vertical Recommendations (RecSys 2025 GenAI Workshop Talk)
Das et al. (2024): Applications of LLMs in E-Commerce Search and Product Knowledge Graph - The DoorDash Case Study (WSDM 2024)
General Links:
Follow me on LinkedIn
Follow me on X
Send me your comments, questions and suggestions to marcel.kurovski@gmail.com
Recsperts Website

In episode 33 of Recsperts, I speak with Joseph A. Konstan, Distinguished McKnight University Professor and Distinguished University Teaching Professor at the University of Minnesota, co-founder of GroupLens, and the very first General Chair of the ACM RecSys conference back in 2007. This year he returns in that role as General Co-Chair of the 20th RecSys in Minneapolis. We talk about what actually makes a recommendation useful, why the field is more than a machine learning application, and how the community around RecSys came into being.
We start with what Joe cares about most: usefulness. He recalls the supermarket thought experiment of printing "buy bananas and bread" on every shopping cart and explains why the early systems were valuable because they were wrong a lot. A recommender that took ten people like you and often got it wrong was also often surprising when it got it right, whereas today's systems are wrong far less and useless far more. As Joe puts it, "I don't care about prediction at all. I care about changing people's behavior." We discuss why optimizing for click-through in news reliably produces clickbait, why leave-one-out evaluation only makes sense if you assume the user already knew the right answer and had simply forgotten it, and why usefulness can never be read off a single metric but depends on the task, the context and the breadth of the user's intent. Sometimes the most useful thing is not the recommendation itself, but the stars, the reviews or the comparison table you put around it.
From there we turn to the community itself. Joe traces it back to the spring of 1996 and the Berkeley Collaborative Filtering Workshop organized by Hal Varian and Paul Resnick, through a decade of scattered workshops at SIGIR, CSCW and CHI, to the first RecSys in 2007: a room on the Minnesota campus, over 120 people, and a substantial delegation from industry including Amazon. We discuss why the conference has always been heavily international and always at the intersection of research and practice, what surprised him most in 20 years (that we are still here and thriving), and the vision he pushed against the pull of becoming just another application of machine learning: a highly constrained, highly contextualized, multi-measure and often multi-stakeholder problem spanning algorithms, interfaces, data and business. This year the 20th RecSys comes home to Minneapolis, with Joe as General Co-Chair alongside George Karypis and Gediminas Adomavicius.
We close on where things are heading. Joe's advice for newcomers is to immerse yourself in an application and find the real problems and opportunities there, rather than arriving enamored with a tool and treating every nail as something to hit. He picks up Karl Higley's point from the POPROX team that the research community keeps focusing on the model when the real object is the system: your algorithm is useless if it is not embedded in something that can deliver its results in a useful way. And he names the work he is most excited about — human decision-making and consumer psychology in the context of choice, the business school perspective that brings marketing and pricing into the picture, multi-sided marketplaces with their often invisible market maker, and the ethics of these systems as a practical lens on the long-term value our metrics still fail to capture.

Enjoy this enriching episode of RECSPERTS – Recommender Systems Experts.
Don’t forget to follow the podcast and please leave a review.

(00:00) - Introduction

(04:02) - About Joseph Konstan

(10:57) - Loving and Hating Machine Learning

(29:05) - What Makes Recommendations Useful

(35:11) - Three Decades of GroupLens

(40:29) - POPROX and Open Online Experimentation

(51:41) - From the Berkeley Workshop to the First RecSys in 2007

(01:11:00) - 20 Years of RecSys and the Vision for the Field

(01:17:38) - Further Challenges and Closing Remarks

Links from the Episode:Joseph Konstan on LinkedIn
Website of Joseph Konstan
GroupLens Research
POPROX: Platform for OPen Recommendation and Online eXperimentation
Net Perceptions
LensKit: Python Recommendation Toolkit
Recommender Systems Specialization on Coursera (Konstan & Ekstrand)
ACM TechTalk by Joseph Konstan: Recommender Systems - Beyond Machine Learning
ACM TechTalk by Joseph Konstan: Recommender Systems - The Power of Personalization
RecSys: The ACM Conference on Recommender Systems
TLDR AI Newsletter
Associated Press
Papers:
Resnick et al. (1994): GroupLens - An Open Architecture for Collaborative Filtering of Netnews
Hill et al. (1995): Recommending and Evaluating Choices in a Virtual Community of Use
Shardanand & Maes (1995): Social Information Filtering - Algorithms for Automating "Word of Mouth"
Resnick & Varian (1997): Recommender Systems (CACM Special Issue)
Sarwar et al. (2001): Item-Based Collaborative Filtering Recommendation Algorithms
Das et al. (2007): Google News Personalization - Scalable Online Collaborative Filtering
Su et al. (2023): Long-Term Value of Exploration - Measurements, Findings and Algorithms
Tran et al. (2024): Transformers Meet ACT-R - Repeat-Aware and Sequential Listening Session Recommendation (Deezer)
Ekstrand et al. (2024): Conducting Recommender Systems User Studies Using POPROX
Higley et al. (2025): What News Recommendation Research Did (But Mostly Didn't) Teach Us About Building A News Recommender
General Links:
Follow me on LinkedIn
Follow me on X
Send me your comments, questions and suggestions to marcel.kurovski@gmail.com
Recsperts Website

In episode 32 of Recsperts, I’m joined by my colleague Sasha Fedintsev, Staff Applied Scientist at Wolt (DoorDash), working across personalization and ads, to unpack the realities of building large-scale recommender systems in food, grocery, and retail delivery. Together, we discuss the specifics of personalization in the delivery domain, and the models and ideas that power Wolt’s recommender system across 30+ markets - where theory quickly meets messy, high-stakes practice.
We explore what makes this domain fundamentally different from traditional e-commerce: strong locality constraints, real-time context, and a heavy skew toward repurchasing behavior. Sasha explains how these factors break many textbook approaches - like standard collaborative filtering - and require creative adaptations such as clustering strategies and multi-stage ranking systems optimized for latency, all while respecting locality constraints.
We also discuss the evolution of recommendation approaches over time - from classical collaborative filtering with ALS, to Neural Collaborative Filtering with BPR, and ultimately to transformer-based models for user sequence modeling and next-purchase prediction powering today’s venue ranking systems.
We also touch on practical challenges such as evaluation in real-world systems, including A/B testing pitfalls and biases in logged data, as well as the complexity introduced by multi-surface experiences like discovery pages, vertical lists, and search. Beyond venues, we discuss why item-level recommendation is an order of magnitude harder - due to scale, context dependence, and availability constraints - and what this implies for future system design.
Throughout the episode, Sasha provides a candid view on the evolving role of a Staff Applied Scientist - bridging research and production, setting scientific standards, and driving cross-team impact.
Enjoy this enriching episode of RECSPERTS – Recommender Systems Experts.
Don’t forget to follow the podcast and please leave a review.

(00:00) - Introduction

(05:10) - About Sasha Fedintsev

(15:26) - The Role of a Staff Applied Scientist

(25:50) - Challenges and Specifics of the Delivery Industry

(47:24) - Ranking and Recommendation Problems at Wolt

(51:31) - NCF with BPR for Wolt's First DNN Recommendation Model

(01:16:43) - User Sequence Transformers for Next Purchase Prediction

(01:26:51) - Explore vs. Exploit or New vs. Recurring Purchases

(01:31:29) - Ads Personalization at Wolt

(01:36:16) - Further Challenges in RecSys

(01:37:58) - A Final Note on Radical Longevity

(01:46:30) - Closing Remarks

Links from the Episode:Alexander "Sasha" Fedintsev on LinkedIn
Alexander on X
Wolt
Alexander Fedintsev at Wolt Tech Talks: Restaurant discovery with Wolt: Deep Neural Networks to power recommendations
H3 Geospatial Indexing System
Recommenders Repository
Tanja Reilly: The Staff Engineer's Path
Will Larson: Staff Engineer: Leadership beyond the management track
Coupon collector's problem
Alexander Fedintsev (2026): Longevity Bottlenecks: Part I — Dementia
Papers:
Rendle et al. (2009): BPR: Bayesian personalized ranking from implicit feedback
He et al. (2017): Neural Collaborative Filtering
Dacrema et al. (2019): Are we really making much progress? A worrying analysis of recent neural recommendation approaches
Rendle et al (2020): Neural Collaborative Filtering vs. Matrix Factorization Revisited
Hu et al. (2008): Collaborative Filtering for Implicit Feedback Datasets
Grbovic et al. (2015): E-commerce in Your Inbox: Product Recommendations at Scale
Quadrana et al. (2018): Sequence-Aware Recommender Systems
Su et al. (2024): Long-Term Value of Exploration: Measurements, Findings and Algorithms
Tran et al. (2024): Transformers Meet ACT-R: Repeat-Aware and Sequential Listening Session Recommendation
Lichtenberg et al. (2024): Ranking Across Different Content Types: The Robust Beauty of Multinomial Blending
General Links:
Follow me on LinkedIn
Follow me on X
Send me your comments, questions and suggestions to marcel.kurovski@gmail.com
Recsperts Website

In episode 31 of Recsperts, I sit down with Elisabeth Lex, Full Professor of Human-Computer Interfaces and Inclusive Technologies at Graz University of Technology and a leading researcher at the intersection of recommender systems, psychology, and human–computer interaction. Together, we explore how recommender systems can become truly human-centric by integrating cognitive, emotional, and personality-aware models into their design.

Elisabeth begins by addressing a common reductionism in the field: treating users primarily as data points rather than as humans with goals, emotions, memories, and cognitive boundaries. We revisit the origins of psychology-informed recommendation, including the Grundy system -the first recommender system, built nearly 50 years ago - which framed book recommendation through stereotype modeling. From there, we discuss how the community’s focus shifted toward solving recommendation mainly as an algorithmic optimization problem, often sidelining richer models of human decision-making.

We then map out the three major branches of psychology-informed RecSys - cognition-inspired, affect-aware, and personality-aware - and dive into practical examples. Elisabeth walks us through her work on modeling music re-listening behavior using cognitive architectures such as ACT-R (Adaptive Control of Thought–Rational) and shows how cognitive constructs like memory decay, attention, and familiarity can meaningfully augment standard approaches like collaborative filtering. We also explore how hybrid systems that combine cognitive models with collaborative filtering can yield not just higher accuracy but also more novelty, diversity, and clearer explanations.

Our conversation also turns to user-centric evaluation. Elisabeth argues that accuracy metrics alone cannot tell us whether a system is genuinely helpful. Instead, we must measure attitudes, perceptions, motivations, and emotional responses - while carefully accounting for cognitive biases, UI effects, and users’ lived experiences.

Towards the end, Elisabeth discusses emerging research directions such as hybrid AI (symbolic + sub-symbolic methods), the role of LLMs and agents, the risks of replacing human studies with automated evaluations, and the responsibility our community has to understand users beyond their clicks.

Enjoy this enriching episode of RECSPERTS – Recommender Systems Experts.
Don’t forget to follow the podcast and please leave a review.

(00:00) - Introduction

(03:15) - About Elisabeth Lex

(07:55) - Grundy, the first Recommender System

(09:03) - Bridging the Gap between Psychology and Modern RecSys

(17:21) - On how and when Elisabeth became a Researcher

(21:39) - Survey on Psychology-Informed RecSys

(39:29) - Personality-Aware Recommendation

(49:43) - Affect- and Emotion-Aware Recommendation

(01:01:37) - Cognition-Inspired Recommendation and the ACT-R Framework

(01:14:39) - Combining Collaborative Filtering and ACT-R for Explainability

(01:21:26) - Human-Centered Design

(01:26:15) - Further Challenges and Closing Remarks

Links from the Episode:Elisabeth Lex on LinkedIn
Website of Elisabeth
AI for Society Lab
First International Workshop on Recommender Systems for Sustainability and Social Good | co-located with RecSys 2024
Second International Workshop on Recommender Systems for Sustainability and Social Good | co-located with RecSys 2025
HyPer Workshop: Hybrid AI for Human-Centric Personalization
Tutorial on Psychology-Informed RecSys
ACT-R: Adaptive Control of Thought-Rational
POPROX: Platform for OPen Recommendation and Online eXperimentation
Papers:
Elaine Rich (1979): User Modeling via Stereotypes
Lex et al. (2021): Psychology-informed Recommender Systems
Reiter-Haas et al. (2021): Predicting Music Relistening Behavior Using the ACT-R Framework
Moscati et al. (2023): Integrating the ACT-R Framework with Collaborative Filtering for Explainable Sequential Music Recommendation
Tran et al. (2024): Transformers Meet ACT-R: Repeat-Aware and Sequential Listening Session Recommendation
General Links:
Follow me on LinkedIn
Follow me on X
Send me your comments, questions and suggestions to marcel.kurovski@gmail.com
Recsperts Website

In episode 30 of Recsperts, I speak with Annelien Smets, Professor at Vrije Universiteit Brussel and Senior Researcher at imec-SMIT, about the value, perception, and practical design of serendipity in recommender systems. Annelien introduces her framework for understanding serendipity through intention, experience, and affordances, and explains the paradox of artificial serendipity - why it cannot be engineered, but only designed for.

We start by unpacking the paradox of serendipity: while serendipity cannot be engineered or planned, systems and environments can be designed to increase the likelihood that serendipitous experiences occur. Annelien explains why randomness alone is not enough and why serendipity always emerges from an interplay between an unexpected encounter and a user’s ability to recognize its relevance and value.

A central part of our discussion focuses on Annelien’s recent framework that distinguishes between intended, experienced, and afforded serendipity. We explore why organizations first need to clarify why they want serendipity - whether as an ideal, a common good, a mediator to achieve other goals (such as long-term retention or long-tail exposure), or even as a product feature in itself. From there, we dive into how users actually experience serendipity, drawing on qualitative interview research that identifies three core components: encounters must feel fortuitous, refreshing, and enriching. These components can manifest in different “flavors,” such as taste broadening, taste deepening, or rediscovering forgotten interests.

We then move beyond algorithms to discuss affordances for serendipity - design principles that span content, user interfaces, and information access. Using examples from libraries, urban spaces, and digital platforms, Annelien shows why serendipity is a system-level property rather than a single metric or model tweak. We also discuss where serendipity can go wrong, including the Netflix “Surprise Me” feature, and why mismatched expectations can actually harm user experience.

To close, we reflect on open research questions, from measuring different types of serendipity to understanding how content types, business models, and platform economics shape what is possible. Annelien also challenges a common myth: serendipity does not automatically burst filter bubbles—and should not be treated as a silver bullet.

Enjoy this enriching episode of RECSPERTS – Recommender Systems Experts.
Don’t forget to follow the podcast and please leave a review.

(00:00) - Introduction

(03:57) - About Annelien Smets

(14:42) - Paradox and Definition of (Artificial) Serendipity

(27:04) - Intended Serendipity

(43:01) - Experienced Serendipity

(01:01:18) - Afforded Serendipity

(01:13:49) - Examples of Serendipity Going Wrong

(01:17:40) - Framework for Serendipity

(01:22:41) - Further Challenges and Closing Remarks

Links from the Episode:Annelien Smets on LinkedIn
Website of Annelien
LinkedIn Article by Annelien Smets (2025): Overcoming the Paradox of Artificial Serendipity
The Serendipity Society
Serendipity Engine
Papers:
Smets (2025): Intended, afforded, and experienced serendipity: overcoming the paradox of artificial serendipity
Smets et al. (2022): Serendipity in Recommender Systems Beyond the Algorithm: A Feature Repository and Experimental Design
Binst et al. (2025): What Is Serendipity? An Interview Study to Conceptualize Experienced Serendipity in Recommender Systems
Ziarani et al. (2021): Serendipity in Recommender Systems: A Systematic Literature Review
Chen et al. (2021): Values of User Exploration in Recommender Systems
Smets et al. (2025): Why Do Recommenders Recommend? Three Waves of Research Perspectives on Recommender Systems
Smets (2023): Designing for Serendipity, a Means or an End?
General Links:
Follow me on LinkedIn
Follow me on X
Send me your comments, questions and suggestions to marcel.kurovski@gmail.com
Recsperts Website