Large-scale search systems evolve faster than human quality assurance scales, especially for long-tail intents and multilingual queries. LLM-as-a-judge approaches are a scalable alternative for evaluating the relevance of search engine result pages (SERPs), but judgments based solely on semantic similarity or world knowledge can drift from actual user preferences, particularly for ambiguous queries. We introduce a behavior-grounded LLM judge that augments each SERP item with a lightweight, auditable behavioral prior in the form of a Query--Relevance--Impressions (QRI) card. Each card summarizes how users have historically interacted with similar queries and results, providing compact empirical evidence that the judge can cite to resolve ambiguity and make more consistent relevance judgments, while still relying on semantic reasoning. In a large-scale music search evaluation at Spotify, using relevance estimates derived from historical user interactions across 6,000 recomposed SERPs, the behavior-grounded judge achieves stronger alignment with user preferences, improving Spearman rank correlation by approximately +5% overall and yielding a +91% relative improvement on disagreement cases. On a multilingual human-judged dataset spanning five languages, grounding further increases correlation with human relevance judgments by +15%. Importantly, when evaluated against outcomes from a live A/B test, the grounded judge shows consistently higher alignment with the observed winning model. While absolute alignment remains moderate, these findings demonstrate that lightweight behavioral grounding can improve the reliability and practical usefulness of LLM-based evaluation in real-world search systems.

Evaluating personalized recommendations remains a central challenge, especially in long-form audio domains like podcasts, where traditional offline metrics suffer from exposure bias and online methods such as A/B testing are costly and operationally constrained. In this paper, we propose a novel framework that leverages Large Language Models (LLMs) as offline judges to assess the quality of podcast recommendations in a scalable and interpretable manner. Our two-stage profile-aware approach first constructs natural-language user profiles distilled from 90 days of listening history. These profiles summarize both topical interests and behavioral patterns, serving as compact, interpretable representations of user preferences. Rather than prompting the LLM with raw data, we use these profiles to provide high-level, semantically rich context-enabling the LLM to reason more effectively about alignment between a user's interests and recommended episodes. This reduces input complexity and improves interpretability. The LLM is then prompted to deliver fine-grained pointwise and pairwise judgments based on the profile-episode match. In a controlled study with 47 participants, our profile-aware judge matched human judgments with high fidelity and outperformed or matched a variant using raw listening histories. The framework enables efficient, profile-aware evaluation for iterative testing and model selection in recommender systems.

Listeners of long-form talk-audio content, such as podcast episodes, often find it challenging to understand the overall structure and locate relevant sections. A practical solution is to divide episodes into chapters--semantically coherent segments labeled with titles and timestamps. Since most episodes on our platform at Spotify currently lack creator-provided chapters, automating the creation of chapters is essential. Scaling the chapterization of podcast episodes presents unique challenges. First, episodes tend to be less structured than written texts, featuring spontaneous discussions with nuanced transitions. Second, the transcripts are usually lengthy, averaging about 16,000 tokens, which necessitates efficient processing that can preserve context. To address these challenges, we introduce PODTILE, a fine-tuned encoder-decoder transformer to segment conversational data. The model simultaneously generates chapter transitions and titles for the input transcript. To preserve context, each input text is augmented with global context, including the episode's title, description, and previous chapter titles. In our intrinsic evaluation, PODTILE achieved an 11% improvement in ROUGE score over the strongest baseline. Additionally, we provide insights into the practical benefits of auto-generated chapters for listeners navigating episode content. Our findings indicate that auto-generated chapters serve as a useful tool for engaging with less popular podcasts. Finally, we present empirical evidence that using chapter titles can enhance effectiveness of sparse retrieval in search tasks.

Generative retrieval for search and recommendation is a promising paradigm for retrieving items, offering an alternative to traditional methods that depend on external indexes and nearest-neighbor searches. Instead, generative models directly associate inputs with item IDs. Given the breakthroughs of Large Language Models (LLMs), these generative systems can play a crucial role in centralizing a variety of Information Retrieval (IR) tasks in a single model that performs tasks such as query understanding, retrieval, recommendation, explanation, re-ranking, and response generation. Despite the growing interest in such a unified generative approach for IR systems, the advantages of using a single, multi-task model over multiple specialized models are not well established in the literature. This paper investigates whether and when such a unified approach can outperform task-specific models in the IR tasks of search and recommendation, broadly co-existing in multiple industrial online platforms, such as Spotify, YouTube, and Netflix. Previous work shows that (1) the latent representations of items learned by generative recommenders are biased towards popularity, and (2) content-based and collaborative-filtering-based information can improve an item’s representations. Motivated by this, our study is guided by two hypotheses: [H1] the joint training regularizes the estimation of each item’s popularity, and [H2] the joint training regularizes the item’s latent representations, where search captures content-based aspects of an item and recommendation captures collaborative-filtering aspects. Our extensive experiments with both simulated and real-world data support both [H1] and [H2] as key contributors to the effectiveness improvements observed in the unified search and recommendation generative models over the single-task approaches.

An important goal of online platforms is to enable content discovery, i.e. allow users to find a catalog entity they were not familiar with. A pre-requisite to discover an entity, e.g. a book, with a search engine is that the entity is retrievable, i.e. there are queries for which the system will surface such entity in the top results. However, machine-learned search engines have a high retrievability bias, where the majority of the queries return the same entities. This happens partly due to the predominance of narrow intent queries, where users create queries using the title of an already known entity, e.g. in book search “harry potter”. The amount of broad queries where users want to discover new entities, e.g. in music search “chill lyrical electronica with an atmospheric feeling to it”, and have a higher tolerance to what they might find, is small in comparison. We focus here on two factors that have a negative impact on the retrievability of the entities (I) the training data used for dense retrieval models and (II) the distribution of narrow and broad intent queries issued in the system. We propose CtrlQGen, a method that generates queries for a chosen underlying intent—narrow or broad. We can use CtrlQGen to improve factor (I) by generating training data for dense retrieval models comprised of diverse synthetic queries. CtrlQGen can also be used to deal with factor (II) by suggesting queries with broader intents to users. Our results on datasets from the domains of music, podcasts, and books reveal that we can significantly decrease the retrievability bias of a dense retrieval model when using CtrlQGen. First, by using the generated queries as training data for dense models we make 9% of the entities retrievable—go from zero to non-zero retrievability. Second, by suggesting broader queries to users, we can make 12% of the entities retrievable in the best case.

Explanations describe product recommendations in a human interpretable way in order to achieve a goal, e.g. persuade users to buy. Unlike web product search, where users have access to diverse information as to why the products might be suitable for their needs, in the voice product search domain the amount of information that can be disclosed is inherently limited. Users in general evaluate a maximum of two products and usually buy low consideration products when using the voice channel [3]. In order to enable decision making in voice product searches we propose here a framework for generating pointwise and pairwise review-based explanations that disclose further information about the products. The POINTWISE method selects a helpful sentence from the top review of the recommended product based on a BERT-based model and uses the extracted sentence to fill a response template. The PAIRWISE method first selects a diverse pair of products—in terms of their review-based representations—from the top-k ranked products for a query, then chooses a helpful review sentence for each product in the pair, and finally fills a template with the sentences. Besides further describing the product, the PAIRWISE method gives a reference point to the users and enables a comparison of the recommendations based on two diverse products for the same information need. Our crowd-sourced evaluation of explanations based on queries from a widely used e-commerce platform shows that the proposed pairwise explanations provide statistically significant improvements compared to the POINTWISE and BASELINE methods for two goals: Effectiveness, i.e. helping users to make good decisions, and Transparency, i.e. explaining how the system works. The gains of PAIRWISE over POINTWISE and BASELINE are consistent for different subsets of data based on the diversity of the selected pairs, average product price associated with the query and the query ambiguity.

The goal of the seventh edition of SCAI (https://scai.info) is to bring together and further grow a community of researchers and practitioners interested in conversational systems for information access. The previous iterations of the workshop already demonstrated the breadth and multidisciplinarity inherent in the design and development of conversational search agents. The proposed shift from traditional web search to search interfaces enabled via human-like dialogue leads to a number of challenges, and although such challenges have received more attention in the recent years, there are many pending research questions that should be addressed by the information retrieval community and can largely benefit from a collaboration with other research fields, such as natural language processing, machine learning, human-computer interaction and dialogue systems. This workshop is intended as a platform enabling a continuous discussion of the major research challenges that surround the design of search-oriented conversational systems. This year, participants have the opportunity to meet in person and have more in-depth interactive discussions with a full-day onsite workshop.

According to the Probability Ranking Principle (PRP), ranking documents in decreasing order of their probability of relevance leads to an optimal document ranking for ad-hoc retrieval.The PRP holds when two conditions are met: [C1] the models are well calibrated, and, [C2] the probabilities of relevance are reported with certainty.We know however that deep neural networks (DNNs) are often not well calibrated and have several sources of uncertainty, and thus [C1] and [C2] might not be satisfied by neural rankers.Given the success of neural Learning to Rank (L2R) approaches-and here, especially BERT-based approaches-we first analyze under which circumstances deterministic neural rankers are calibrated for conversational search problems.Then, motivated by our findings we use two techniques to model the uncertainty of neural rankers leading to the proposed stochastic rankers, which output a predictive distribution of relevance as opposed to point estimates.Our experimental results on the ad-hoc retrieval task of conversation response ranking 1 reveal that (i) BERTbased rankers are not robustly calibrated and that stochastic BERT-based rankers yield better calibration; and (ii) uncertainty estimation is beneficial for both risk-aware neural ranking, i.e. taking into account the uncertainty when ranking documents, and for predicting unanswerable conversational contexts.1 The source code and data are available at https://github.com/Guzpenha/transformer_ rankers/tree/uncertainty_estimation.2 Standard retrieval task where the user specifies his information need through a query which initiates a search by the 0.9 0.1 0.7

Heavily pre-trained transformer models such as BERT have recently shown to be remarkably powerful at language modelling, achieving impressive results on numerous downstream tasks. It has also been shown that they implicitly store factual knowledge in their parameters after pre-training. Understanding what the pre-training procedure of LMs actually learns is a crucial step for using and improving them for Conversational Recommender Systems (CRS). We first study how much off-the-shelf pre-trained BERT “knows” about recommendation items such as books, movies and music. In order to analyze the knowledge stored in BERT’s parameters, we use different probes (i.e., tasks to examine a trained model regarding certain properties) that require different types of knowledge to solve, namely content-based and collaborative-based. Content-based knowledge is knowledge that requires the model to match the titles of items with their content information, such as textual descriptions and genres. In contrast, collaborative-based knowledge requires the model to match items with similar ones, according to community interactions such as ratings. We resort to BERT’s Masked Language Modelling (MLM) head to probe its knowledge about the genre of items, with cloze style prompts. In addition, we employ BERT’s Next Sentence Prediction (NSP) head and representations’ similarity (SIM) to compare relevant and non-relevant search and recommendation query-document inputs to explore whether BERT can, without any fine-tuning, rank relevant items first. Finally, we study how BERT performs in a conversational recommendation downstream task. To this end, we fine-tune BERT to act as a retrieval-based CRS. Overall, our experiments show that: (i) BERT has knowledge stored in its parameters about the content of books, movies and music; (ii) it has more content-based knowledge than collaborative-based knowledge; and (iii) fails on conversational recommendation when faced with adversarial data.

Ensembling multiple recommender systems via stacking has shown to be effective at improving collaborative recommendation. Recent work extends stacking to use additional user performance predictors (e.g., the total number of ratings made by the user) to help determine how much each base recommender should contribute to the ensemble. Nonetheless, despite the cost of handcrafting discriminative predictors, which typically requires deep knowledge of the strengths and weaknesses of each recommender in the ensemble, only minor improvements have been observed. To overcome this limitation, instead of engineering complex features to predict the performance of different recommenders for a given user, we propose to directly estimate these performances by leveraging the user’s own historical ratings. Experiments on real-world datasets from multiple domains demonstrate that using performance estimates as additional features can significantly improve the accuracy of state-of-the-art ensemblers, achieving [email protected] improvements by an average of 23% over not using them.

The prominent success of music streaming services has brought increasingly complex challenges for music recommendation. In particular, in a streaming setting, songs are consumed sequentially within a listening session, which should cater not only for the user's historical preferences, but also for eventual preference drifts, triggered by a sudden change in the user's context. In this paper, we propose a novel online learning to rank approach for music recommendation aimed to continuously learn from the user's listening feedback. In contrast to existing online learning approaches for music recommendation, we leverage implicit feedback as the only signal of the user's preference. Moreover, to adapt rapidly to preference drifts over millions of songs, we represent each song in a lower dimensional feature space and explore multiple directions in this space as duels of candidate recommendation models. Our thorough evaluation using listening sessions from Last.fm demonstrates the effectiveness of our approach at learning faster and better compared to state-of-the-art online learning approaches.