arXiv subjects

Gustavo Penha

Publications and source records attributed to Gustavo Penha.

At least 19 recordsLinked to original sources

arXiv2026

The Disconnect Between Better Descriptive Reasoning Trace Quality and Recommendation Effectiveness

Recent work has focused on improving explicit natural-language descriptive reasoning traces for generative recommendation. This includes systems that augment semantic ID (SID) prediction with chain-of-thought reasoning. However, because SIDs are opaque learned identifiers rather than natural language, they require costly alignment before an LLM can reason over them. This provides a controlled experimental setting in which both item representation (Title vs. SID) and semantic grounding (minimal vs. extensive SID alignment) can be varied independently. We therefore present the first controlled comparison of descriptive reasoning trace quality across semantic IDs and natural-language titles in a 2 x 2 factorial study on three Amazon product domains using a shared Qwen3-1.7B backbone. We find that introducing explicit descriptive reasoning traces reduces traditional offline recommendation effectiveness under standard SFT and RL training, even though natural language titles produce substantially more grounded and interpretable traces. Extensive SID alignment improves descriptive trace quality but not traditional offline recommendation effectiveness, while a richer reward signal partially recovers performance. Overall, our results show that improving descriptive reasoning trace quality is not, by itself, sufficient to consistently improve traditional offline recommendation effectiveness under the training objectives and evaluation protocols studied here.

cs.IR↗

arXiv2026

As It Was: Aligning LLM Search Evaluation with Historical User Preferences

Large-scale search systems evolve faster than human quality assurance can scale, especially for long-tail intents and multilingual queries. LLM-as-a-judge approaches provide 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 and 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.

cs.IR↗

arXiv2026

Deploying Semantic ID-based Generative Retrieval for Large-Scale Podcast Discovery at Spotify

Podcast listening is often grounded in a set of favorite shows, while listener intent can evolve over time. This combination of stable preferences and changing intent motivates recommendation approaches that support both familiarity and exploration. Traditional recommender systems typically emphasize long-term interaction patterns, and are less explicitly designed to incorporate rich contextual signals or flexible, intent-aware discovery objectives. In this setting, models that can jointly reason over semantics, context, and user state offer a promising direction. Large Language Models (LLMs) provide strong semantic reasoning and contextual conditioning for discovery-oriented recommendation, but deploying them in production introduces challenges in catalog grounding, user-level personalization, and latency-critical serving. We address these challenges with GLIDE, a production-scale generative recommender for podcast discovery at Spotify. GLIDE formulates recommendation as an instruction-following task over a discretized catalog using Semantic IDs, enabling grounded generation over a large inventory. The model conditions on recent listening history and lightweight user context, while injecting long-term user embeddings as soft prompts to capture stable preferences under strict inference constraints. We evaluate GLIDE using offline retrieval metrics, human judgments, and LLM-based evaluation, and validate its impact through large-scale online A/B testing. Across experiments involving millions of users, GLIDE increases non-habitual podcast streaming on Spotify home surface by up to 5.4% and new-show discovery by up to 14.3%, while meeting production cost and latency constraints.

cs.IR↗

arXiv2026

Stream: Scaling up Mechanistic Interpretability to Long Context in LLMs via Sparse Attention

As Large Language Models (LLMs) scale to million-token contexts, traditional Mechanistic Interpretability techniques for analyzing attention scale quadratically with context length, demanding terabytes of memory beyond 100,000 tokens. We introduce Sparse Tracing, a novel technique that leverages dynamic sparse attention to efficiently analyze long context attention patterns. We present Stream, a compilable hierarchical pruning algorithm that estimates per-head sparse attention masks in near-linear time $O(T \log T)$ and linear space $O(T)$, enabling one-pass interpretability at scale. Stream performs a binary-search-style refinement to retain only the top-$k$ key blocks per query while preserving the model's next-token behavior. We apply Stream to long chain-of-thought reasoning traces and identify thought anchors while pruning 97-99\% of token interactions. On the RULER benchmark, Stream preserves critical retrieval paths while discarding 90-96\% of interactions and exposes layer-wise routes from the needle to output. Our method offers a practical drop-in tool for analyzing attention patterns and tracing information flow without terabytes of caches. By making long context interpretability feasible on consumer GPUs, Sparse Tracing helps democratize chain-of-thought monitoring. Code is available at https://anonymous.4open.science/r/stream-03B8/.

cs.CL↗

arXiv2025

From IR to RecSys: Evaluating LLM-based Judges in Cranfield-style Recommendation Collections

The Cranfield paradigm has long provided reliable, reproducible evaluation in ad hoc retrieval, and recent work has begun extending this framework to recommender systems. A recent development in IR is the use of Large Language Models (LLMs) as automatic relevance judges, showing promising agreement with human assessors. Whether this LLM-judge paradigm---studied predominantly on query--document pairs---transfers to the subjective, profile-driven nature of recommendation remains an open question. This paper bridges the IR and RecSys evaluation traditions by systematically investigating LLM-based judges within a Cranfield-style recommendation collection. Using the ML-32M-ext movie recommendation collection, we first demonstrate that traditional train--test splits yield substantially incomplete relevance labels and unreliable system rankings compared to Cranfield-style pooling. We then assess LLM-judge alignment with human labels, finding that richer item metadata and longer user histories improve agreement, although item-level agreement remains moderate overall. Rankings derived from LLM-judge labels achieve high agreement with human-based rankings (Kendall's tau up to 0.92 for nDCG@100 across 52 system configurations), comparable to values reported for TREC ad hoc retrieval collections. Crucially, LLM-judge recovers system rankings that are distorted under traditional evaluation---correctly identifying systems that are undervalued or overvalued by incomplete labels. An industrial case study in podcast recommendation further demonstrates the practical value of LLM-judge for model selection. Rather than positioning LLM-judges as a replacement for human or interaction-based evaluation, our results support their use as a promising complementary signal: item-level agreement with humans is moderate, yet system-level rankings---which aggregate judgments over many user--item pairs---remain stable.

cs.IR↗

arXiv2025

AudioBoost: Increasing Audiobook Retrievability in Spotify Search with Synthetic Query Generation

Spotify has recently introduced audiobooks as part of its catalog, complementing its music and podcast offering. Search is often the first entry point for users to access new items, and an important goal for Spotify is to support users in the exploration of the audiobook catalog. More specifically, we would like to enable users without a specific item in mind to broadly search by topic, genre, story tropes, decade, and discover audiobooks, authors and publishers they may like. To do this, we need to 1) inspire users to type more exploratory queries for audiobooks and 2) augment our retrieval systems to better deal with exploratory audiobook queries. This is challenging in a cold-start scenario, where we have a retrievabiliy bias due to the little amount of user interactions with audiobooks compared to previously available items such as music and podcast content. To address this, we propose AudioBoost, a system to boost audiobook retrievability in Spotify's Search via synthetic query generation. AudioBoost leverages Large Language Models (LLMs) to generate synthetic queries conditioned on audiobook metadata. The synthetic queries are indexed both in the Query AutoComplete (QAC) and in the Search Retrieval engine to improve query formulation and retrieval at the same time. We show through offline evaluation that synthetic queries increase retrievability and are of high quality. Moreover, results from an online A/B test show that AudioBoost leads to a +0.7% in audiobook impressions, +1.22% in audiobook clicks, and +1.82% in audiobook exploratory query completions.

cs.IR↗

arXiv2025

Semantic IDs for Joint Generative Search and Recommendation

Generative models powered by Large Language Models (LLMs) are emerging as a unified solution for powering both recommendation and search tasks. A key design choice in these models is how to represent items, traditionally through unique identifiers (IDs) and more recently with Semantic IDs composed of discrete codes, obtained from embeddings. While task-specific embedding models can improve performance for individual tasks, they may not generalize well in a joint setting. In this paper, we explore how to construct Semantic IDs that perform well both in search and recommendation when using a unified model. We compare a range of strategies to construct Semantic IDs, looking into task-specific and cross-tasks approaches, and also whether each task should have its own semantic ID tokens in a joint search and recommendation generative model. Our results show that using a bi-encoder model fine-tuned on both search and recommendation tasks to obtain item embeddings, followed by the construction of a unified Semantic ID space provides an effective trade-off, enabling strong performance in both tasks. We hope these findings spark follow-up work on generalisable, semantically grounded ID schemes and inform the next wave of unified generative recommender architectures.

cs.IR↗

arXiv2025

Evaluating Podcast Recommendations with Profile-Aware LLM-as-a-Judge

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.

cs.IR↗

arXiv2025

Adaptive Repetition for Mitigating Position Bias in LLM-Based Ranking

When using LLMs to rank items based on given criteria, or evaluate answers, the order of candidate items can influence the model's final decision. This sensitivity to item positioning in a LLM's prompt is known as position bias. Prior research shows that this bias exists even in large models, though its severity varies across models and tasks. In addition to position bias, LLMs also exhibit varying degrees of low repetition consistency, where repeating the LLM call with the same candidate ordering can lead to different rankings. To address both inconsistencies, a common approach is to prompt the model multiple times with different candidate orderings and aggregate the results via majority voting. However, this repetition strategy, significantly increases computational costs. Extending prior findings, we observe that both the direction -- favoring either the earlier or later candidate in the prompt -- and magnitude of position bias across instances vary substantially, even within a single dataset. This observation highlights the need for a per-instance mitigation strategy. To this end, we introduce a dynamic early-stopping method that adaptively determines the number of repetitions required for each instance. Evaluating our approach across three LLMs of varying sizes and on two tasks, namely re-ranking and alignment, we demonstrate that transitioning to a dynamic repetition strategy reduces the number of LLM calls by an average of 81%, while preserving the accuracy. Furthermore, we propose a confidence-based adaptation to our early-stopping method, reducing LLM calls by an average of 87% compared to static repetition, with only a slight accuracy trade-off relative to our original early-stopping method.

cs.LG↗

arXiv2025

Aligned Query Expansion: Efficient Query Expansion for Information Retrieval through LLM Alignment

With the breakthroughs in large language models (LLMs), query generation techniques that expand documents and queries with related terms are becoming increasingly popular in the information retrieval field. Such techniques have been shown to improve the effectiveness of traditional lexical retrieval methods by dealing with the vocabulary mismatch problem. Recent work has found that generating queries with a greedy decoding strategy can produce sub-optimal queries, including hallucinations, and proposed to filter out queries before expansion. This `generate-then-filter' approach is costly, as it requires generating multiple queries and applying a relevance model to all of them and does not teach the LLM which of the generated queries is more effective for expansion. To overcome such limitations, we propose Aligned Query Expansion (AQE), a novel approach to enhance query expansion for passage retrieval in open-domain question answering. AQE leverages recent techniques in LLM alignment to fine-tune models for generating query expansions that directly optimize the effectiveness of the retrieval task, eliminating the need for additional filtering steps. This alignment ensures that queries are more relevant, reducing computational costs while improving retrieval effectiveness. Empirical evaluations show that AQE outperforms baseline models for query expansion in both in-domain and out-of-domain settings, demonstrating significant improvements in retrieval effectiveness.

cs.IR↗

arXiv2025

Contextualizing Spotify's Audiobook List Recommendations with Descriptive Shelves

In this paper, we propose a pipeline to generate contextualized list recommendations with descriptive shelves in the domain of audiobooks. By creating several shelves for topics the user has an affinity to, e.g. Uplifting Women's Fiction, we can help them explore their recommendations according to their interests and at the same time recommend a diverse set of items. To do so, we use Large Language Models (LLMs) to enrich each item's metadata based on a taxonomy created for this domain. Then we create diverse descriptive shelves for each user. A/B tests show improvements in user engagement and audiobook discovery metrics, demonstrating benefits for users and content creators.

cs.IR↗

arXiv2025

Text2Tracks: Prompt-based Music Recommendation via Generative Retrieval

In recent years, Large Language Models (LLMs) have enabled users to provide highly specific music recommendation requests using natural language prompts (e.g. "Can you recommend some old classics for slow dancing?"). In this setup, the recommended tracks are predicted by the LLM in an autoregressive way, i.e. the LLM generates the track titles one token at a time. While intuitive, this approach has several limitation. First, it is based on a general purpose tokenization that is optimized for words rather than for track titles. Second, it necessitates an additional entity resolution layer that matches the track title to the actual track identifier. Third, the number of decoding steps scales linearly with the length of the track title, slowing down inference. In this paper, we propose to address the task of prompt-based music recommendation as a generative retrieval task. Within this setting, we introduce novel, effective, and efficient representations of track identifiers that significantly outperform commonly used strategies. We introduce Text2Tracks, a generative retrieval model that learns a mapping from a user's music recommendation prompt to the relevant track IDs directly. Through an offline evaluation on a dataset of playlists with language inputs, we find that (1) the strategy to create IDs for music tracks is the most important factor for the effectiveness of Text2Tracks and semantic IDs significantly outperform commonly used strategies that rely on song titles as identifiers (2) provided with the right choice of track identifiers, Text2Tracks outperforms sparse and dense retrieval solutions trained to retrieve tracks from language prompts.

cs.IR↗

arXiv2024

Bridging Search and Recommendation in Generative Retrieval: Does One Task Help the Other?

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.

cs.IR↗

arXiv2024

PODTILE: Facilitating Podcast Episode Browsing with Auto-generated Chapters

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.

cs.IR↗

arXiv2023

Improving Content Retrievability in Search with Controllable Query Generation

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.

cs.IR↗

arXiv2023

Do the Findings of Document and Passage Retrieval Generalize to the Retrieval of Responses for Dialogues?

A number of learned sparse and dense retrieval approaches have recently been proposed and proven effective in tasks such as passage retrieval and document retrieval. In this paper we analyze with a replicability study if the lessons learned generalize to the retrieval of responses for dialogues, an important task for the increasingly popular field of conversational search. Unlike passage and document retrieval where documents are usually longer than queries, in response ranking for dialogues the queries (dialogue contexts) are often longer than the documents (responses). Additionally, dialogues have a particular structure, i.e. multiple utterances by different users. With these differences in mind, we here evaluate how generalizable the following major findings from previous works are: (F1) query expansion outperforms a no-expansion baseline; (F2) document expansion outperforms a no-expansion baseline; (F3) zero-shot dense retrieval underperforms sparse baselines; (F4) dense retrieval outperforms sparse baselines; (F5) hard negative sampling is better than random sampling for training dense models. Our experiments -- based on three different information-seeking dialogue datasets -- reveal that four out of five findings (F2-F5) generalize to our domain

cs.IR↗

arXiv2022

Sparse and Dense Approaches for the Full-rank Retrieval of Responses for Dialogues

Ranking responses for a given dialogue context is a popular benchmark in which the setup is to re-rank the ground-truth response over a limited set of $n$ responses, where $n$ is typically 10. The predominance of this setup in conversation response ranking has lead to a great deal of attention to building neural re-rankers, while the first-stage retrieval step has been overlooked. Since the correct answer is always available in the candidate list of $n$ responses, this artificial evaluation setup assumes that there is a first-stage retrieval step which is always able to rank the correct response in its top-$n$ list. In this paper we focus on the more realistic task of full-rank retrieval of responses, where $n$ can be up to millions of responses. We investigate both dialogue context and response expansion techniques for sparse retrieval, as well as zero-shot and fine-tuned dense retrieval approaches. Our findings based on three different information-seeking dialogue datasets reveal that a learned response expansion technique is a solid baseline for sparse retrieval. We find the best performing method overall to be dense retrieval with intermediate training, i.e. a step after the language model pre-training where sentence representations are learned, followed by fine-tuning on the target conversational data. We also investigate the intriguing phenomena that harder negatives sampling techniques lead to worse results for the fine-tuned dense retrieval models. The code and datasets are available at https://github.com/Guzpenha/transformer_rankers/tree/full_rank_retrieval_dialogues.

cs.IR↗

arXiv2022

Evaluating the Robustness of Retrieval Pipelines with Query Variation Generators

Heavily pre-trained transformers for language modelling, such as BERT, have shown to be remarkably effective for Information Retrieval (IR) tasks, typically applied to re-rank the results of a first-stage retrieval model. IR benchmarks evaluate the effectiveness of retrieval pipelines based on the premise that a single query is used to instantiate the underlying information need. However, previous research has shown that (I) queries generated by users for a fixed information need are extremely variable and, in particular, (II) neural models are brittle and often make mistakes when tested with modified inputs. Motivated by those observations we aim to answer the following question: how robust are retrieval pipelines with respect to different variations in queries that do not change the queries' semantics? In order to obtain queries that are representative of users' querying variability, we first created a taxonomy based on the manual annotation of transformations occurring in a dataset (UQV100) of user-created query variations. For each syntax-changing category of our taxonomy, we employed different automatic methods that when applied to a query generate a query variation. Our experimental results across two datasets for two IR tasks reveal that retrieval pipelines are not robust to these query variations, with effectiveness drops of $\approx20\%$ on average. The code and datasets are available at https://github.com/Guzpenha/query_variation_generators.

cs.IR↗