Temporal User Profiling with LLMs: Balancing Short-Term and Long-Term Preferences for Recommendations

Accurately modeling user preferences is crucial for improving the performance of content-based recommender systems. Existing approaches often rely on simplistic user profiling methods, such as averaging or concatenating item embeddings, which fail to capture the nuanced nature of user preference dynamics, particularly the interactions between long-term and short-term preferences. In this work, we propose LLM-driven Temporal User Profiling (LLM-TUP), a novel method for user profiling that explicitly models short-term and long-term preferences by leveraging interaction timestamps and generating natural language representations of user histories using a large language model (LLM). These representations are encoded into high-dimensional embeddings using a pre-trained BERT model, and an attention mechanism is applied to dynamically fuse the short-term and long-term embeddings into a comprehensive user profile. Experimental results on real-world datasets demonstrate that LLM-TUP achieves substantial improvements over several baselines, underscoring the effectiveness of our temporally aware user-profiling approach and the use of semantically rich user profiles, generated by LLMs, for personalized content-based recommendation.

Using LLMs to Capture Users' Temporal Context for Recommendation

Effective recommender systems demand dynamic user understanding, especially in complex, evolving environments. Traditional user profiling often fails to capture the nuanced, temporal contextual factors of user preferences, such as transient short-term interests and enduring long-term tastes. This paper presents an assessment of Large Language Models (LLMs) for generating semantically rich, time-aware user profiles. We do not propose a novel end-to-end recommendation architecture; instead, the core contribution is a systematic investigation into the degree of LLM effectiveness in capturing the dynamics of user context by disentangling short-term and long-term preferences. This approach, framing temporal preferences as dynamic user contexts for recommendations, adaptively fuses these distinct contextual components into comprehensive user embeddings. The evaluation across Movies&TV and Video Games domains suggests that while LLM-generated profiles offer semantic depth and temporal structure, their effectiveness for context-aware recommendations is notably contingent on the richness of user interaction histories. Significant gains are observed in dense domains (e.g., Movies&TV), whereas improvements are less pronounced in sparse environments (e.g., Video Games). This work highlights LLMs' nuanced potential in enhancing user profiling for adaptive, context-aware recommendations, emphasizing the critical role of dataset characteristics for practical applicability.

Towards Explainable Temporal User Profiling with LLMs

Accurately modeling user preferences is vital not only for improving recommendation performance but also for enhancing transparency in recommender systems. Conventional user profiling methods, such as averaging item embeddings, often overlook the evolving, nuanced nature of user interests, particularly the interplay between short-term and long-term preferences. In this work, we leverage large language models (LLMs) to generate natural language summaries of users' interaction histories, distinguishing recent behaviors from more persistent tendencies. Our framework not only models temporal user preferences but also produces natural language profiles that can be used to explain recommendations in an interpretable manner. These textual profiles are encoded via a pre-trained model, and an attention mechanism dynamically fuses the short-term and long-term embeddings into a comprehensive user representation. Beyond boosting recommendation accuracy over multiple baselines, our approach naturally supports explainability: the interpretable text summaries and attention weights can be exposed to end users, offering insights into why specific items are suggested. Experiments on real-world datasets underscore both the performance gains and the promise of generating clearer, more transparent justifications for content-based recommendations.

Mitigating Exposure Bias in Online Learning to Rank Recommendation: A Novel Reward Model for Cascading Bandits

Exposure bias is a well-known issue in recommender systems where items and suppliers are not equally represented in the recommendation results. This bias becomes particularly problematic over time as a few items are repeatedly over-represented in recommendation lists, leading to a feedback loop that further amplifies this bias. Although extensive research has addressed this issue in model-based or neighborhood-based recommendation algorithms, less attention has been paid to online recommendation models, such as those based on top-K contextual bandits, where recommendation models are dynamically updated with ongoing user feedback. In this paper, we study exposure bias in a class of well-known contextual bandit algorithms known as Linear Cascading Bandits. We analyze these algorithms in their ability to handle exposure bias and provide a fair representation of items in the recommendation results. Our analysis reveals that these algorithms fail to mitigate exposure bias in the long run during the course of ongoing user interactions. We propose an Exposure-Aware reward model that updates the model parameters based on two factors: 1) implicit user feedback and 2) the position of the item in the recommendation list. The proposed model mitigates exposure bias by controlling the utility assigned to the items based on their exposure in the recommendation list. Our experiments with two real-world datasets show that our proposed reward model improves the exposure fairness of the linear cascading bandits over time while maintaining the recommendation accuracy. It also outperforms the current baselines. Finally, we prove a high probability upper regret bound for our proposed model, providing theoretical guarantees for its performance.

Beyond Static Calibration: The Impact of User Preference Dynamics on Calibrated Recommendation

Calibration in recommender systems is an important performance criterion that ensures consistency between the distribution of user preference categories and that of recommendations generated by the system. Standard methods for mitigating miscalibration typically assume that user preference profiles are static, and they measure calibration relative to the full history of user's interactions, including possibly outdated and stale preference categories. We conjecture that this approach can lead to recommendations that, while appearing calibrated, in fact, distort users' true preferences. In this paper, we conduct a preliminary investigation of recommendation calibration at a more granular level, taking into account evolving user preferences. By analyzing differently sized training time windows from the most recent interactions to the oldest, we identify the most relevant segment of user's preferences that optimizes the calibration metric. We perform an exploratory analysis with datasets from different domains with distinctive user-interaction characteristics. We demonstrate how the evolving nature of user preferences affects recommendation calibration, and how this effect is manifested differently depending on the characteristics of the data in a given domain. Datasets, codes, and more detailed experimental results are available at: https://github.com/nicolelin13/DynamicCalibrationUMAP.

Fairness of Exposure in Dynamic Recommendation

Exposure bias is a well-known issue in recommender systems where the exposure is not fairly distributed among items in the recommendation results. This is especially problematic when bias is amplified over time as a few items (e.g., popular ones) are repeatedly over-represented in recommendation lists and users' interactions with those items will amplify bias towards those items over time resulting in a feedback loop. This issue has been extensively studied in the literature in static recommendation environment where a single round of recommendation result is processed to improve the exposure fairness. However, less work has been done on addressing exposure bias in a dynamic recommendation setting where the system is operating over time, the recommendation model and the input data are dynamically updated with ongoing user feedback on recommended items at each round. In this paper, we study exposure bias in a dynamic recommendation setting. Our goal is to show that existing bias mitigation methods that are designed to operate in a static recommendation setting are unable to satisfy fairness of exposure for items in long run. In particular, we empirically study one of these methods and show that repeatedly applying this method fails to fairly distribute exposure among items in long run. To address this limitation, we show how this method can be adapted to effectively operate in a dynamic recommendation setting and achieve exposure fairness for items in long run. Experiments on a real-world dataset confirm that our solution is superior in achieving long-term exposure fairness for the items while maintaining the recommendation accuracy.

Exposure-Aware Recommendation using Contextual Bandits

Exposure bias is a well-known issue in recommender systems where items and suppliers are not equally represented in the recommendation results. This is especially problematic when bias is amplified over time as a few items (e.g., popular ones) are repeatedly over-represented in recommendation lists and users' interactions with those items will amplify bias towards those items over time resulting in a feedback loop. This issue has been extensively studied in the literature on model-based or neighborhood-based recommendation algorithms, but less work has been done on online recommendation models, such as those based on top-K contextual bandits, where recommendation models are dynamically updated with ongoing user feedback. In this paper, we study exposure bias in a class of well-known contextual bandit algorithms known as Linear Cascading Bandits. We analyze these algorithms on their ability to handle exposure bias and provide a fair representation for items in the recommendation results. Our analysis reveals that these algorithms tend to amplify exposure disparity among items over time. In particular, we observe that these algorithms do not properly adapt to the feedback provided by the users and frequently recommend certain items even when those items are not selected by users. To mitigate this bias, we propose an Exposure-Aware (EA) reward model that updates the model parameters based on two factors: 1) user feedback (i.e., clicked or not), and 2) position of the item in the recommendation list. This way, the proposed model controls the utility assigned to items based on their exposure in the recommendation list. Extensive experiments on two real-world datasets using three contextual bandit algorithms show that the proposed reward model reduces exposure bias amplification in long run while maintaining the recommendation accuracy.

Behavioral Player Rating in Competitive Online Shooter Games

Competitive online games use rating systems for matchmaking; progression-based algorithms that estimate the skill level of players with interpretable ratings in terms of the outcome of the games they played. However, the overall experience of players is shaped by factors beyond the sole outcome of their games. In this paper, we engineer several features from in-game statistics to model players and create ratings that accurately represent their behavior and true performance level. We then compare the estimating power of our behavioral ratings against ratings created with three mainstream rating systems by predicting rank of players in four popular game modes from the competitive shooter genre. Our results show that the behavioral ratings present more accurate performance estimations while maintaining the interpretability of the created representations. Considering different aspects of the playing behavior of players and using behavioral ratings for matchmaking can lead to match-ups that are more aligned with players' goals and interests, consequently resulting in a more enjoyable gaming experience.

Using user's local context to support local news

American local newspapers have been experiencing a large loss of reader retention and business within the past 15 years due to the proliferation of online news sources. Local media companies are starting to shift from an advertising-supported business model to one based on subscriptions to mitigate this problem. With this subscription model, there is a need to increase user engagement and personalization, and recommender systems are one way for these news companies to accomplish this goal. However, using standard modeling approaches that focus on users' global preferences is not appropriate in this context because the local preferences of users exhibit some specific characteristics which do not necessarily match their long-term or global preferences in the news. Our research explores a localized session-based recommendation approach, using recommendations based on local news articles and articles pertaining to the different local news categories. Experiments performed on a news dataset from a local newspaper show that these local models, particularly certain categories of items, do indeed provide more accuracy and effectiveness for personalization which, in turn, may lead to more user engagement with local news content.

Player Modeling using Behavioral Signals in Competitive Online Games

Competitive online games use rating systems to match players with similar skills to ensure a satisfying experience for players. In this paper, we focus on the importance of addressing different aspects of playing behavior when modeling players for creating match-ups. To this end, we engineer several behavioral features from a dataset of over 75,000 battle royale matches and create player models based on the retrieved features. We then use the created models to predict ranks for different groups of players in the data. The predicted ranks are compared to those of three popular rating systems. Our results show the superiority of simple behavioral models over mainstream rating systems. Some behavioral features provided accurate predictions for all groups of players while others proved useful for certain groups of players. The results of this study highlight the necessity of considering different aspects of the player's behavior such as goals, strategy, and expertise when making assignments.