Relevance meets Diversity: A User-Centric Framework for Knowledge Exploration through Recommendations

Abstract: Providing recommendations that are both relevant and diverse is a key consideration of modern recommender systems. Optimizing both of these measures presents a fundamental trade-off, as higher diversity typically comes at the cost of relevance, resulting in lower user engagement. Existing recommendation algorithms try to resolve this trade-off by combining the two measures, relevance and diversity, into one aim and then seeking recommendations that optimize the combined objective, for a given number of items to recommend. Traditional approaches, however, do not consider the user interaction with the recommended items. In this paper, we put the user at the central stage, and build on the interplay between relevance, diversity, and user behavior. In contrast to applications where the goal is solely to maximize engagement, we focus on scenarios aiming at maximizing the total amount of knowledge encountered by the user. We use diversity as a surrogate of the amount of knowledge obtained by the user while interacting with the system, and we seek to maximize diversity. We propose a probabilistic user-behavior model in which users keep interacting with the recommender system as long as they receive relevant recommendations, but they may stop if the relevance of the recommended items drops. Thus, for a recommender system to achieve a high-diversity measure, it will need to produce recommendations that are both relevant and diverse. Finally, we propose a novel recommendation strategy that combines relevance and diversity by a copula function. We conduct an extensive evaluation of the proposed methodology over multiple datasets, and we show that our strategy outperforms several state-of-the-art competitors. Our implementation is publicly available at https://github.com/EricaCoppolillo/EXPLORE.

• The paper introduces a user-centric framework that balances recommendation relevance and content diversity by incorporating user behavior and tailored diversity metrics.
• It employs a probabilistic model using the Weibull distribution to simulate user patience, ensuring sustained engagement over time.
• The proposed copula-based strategy outperforms traditional methods by achieving a superior balance between quality and diversity in recommendations.

Relevance Meets Diversity: A User-Centric Framework for Knowledge Exploration through Recommendations

This essay provides an overview of the paper "Relevance meets Diversity: A User-Centric Framework for Knowledge Exploration through Recommendations" by Erica Coppolillo, Giuseppe Manco, and Aristides Gionis. The paper addresses the challenging and often conflicting goals of optimizing relevance and diversity in recommender systems, putting forward a user-centric framework to balance both metrics by incorporating user behavior.

Context and Motivation

Recommender systems serve a crucial role in navigating the vast expanse of information available on various platforms. The primary objective of these systems has traditionally been to maximize relevance — the degree to which recommended items align with user preferences. However, frequent recommendations of similar items can create homogeneous content consumption patterns, often referred to as "rabbit holes," impacting users' overall knowledge and satisfaction negatively. Hence, integrating diversity into recommendation algorithms is essential for fostering broader knowledge exploration and long-term user engagement.

Contributions

The paper's contributions can be grouped as follows:

1. User-Centric Exploration Model: The authors propose a novel framework focusing on the interplay between relevance, diversity, and user behavior. Unlike traditional methods which consider only a static set of interactions, this user-centric approach models the recommender system interaction as an ongoing knowledge exploration process. Users continue interacting as long as the recommendations remain engaging, and they may quit if relevance drops significantly.
2. Measurement of Diversity: Two diversity measures are introduced — coverage-based, which considers the range of categories covered by the recommended items, and distance-based, which evaluates diversity based on pairwise Jaccard distances between items. These metrics are adapted for recommendation tasks, accurately capturing the level of knowledge expansion provided.
3. Probabilistic User-Behavior Model: The framework integrates a probabilistic model that accounts for user patience and tolerance over time, with reliance on the Weibull distribution to simulate the likelihood of quitting based on the relevance scores and temporal factors.
4. Recommendation Strategy Using Copula Function: A novel recommendation strategy combining relevance and diversity through the Clayton copula function is proposed. This ensures an optimal blend of both metrics, maximizing overall user knowledge while maintaining engagement.

Methodology

User Model and Problem Formulation

The user model is the cornerstone of the proposed framework. Here, the interaction process is dynamic, governed by a relevance-scoring function. Users are presented with a list of recommendations at each step and decide whether to continue interacting based on the relevance and diversity of the items. The quitting probability is modeled as a function of the item's utility and the user's patience, accounting for the natural decline in users' interest over time.

Diversity Definitions

1. Coverage-Based Diversity: This metric evaluates how well the set of recommended items covers the range of categories. Larger sets naturally score better, provided they cover a wider range.
2. Distance-Based Diversity: This metric uses Jaccard distances to assess pairwise diversity among items in the recommendation list. The normalized average distance serves as the diversity score.

Recommendation Algorithm

The proposed algorithm evaluates each item's relevance and marginal diversity (the incremental diversity an item adds to the set) and then combines these scores using the Clayton copula function. This approach ensures that both high relevance and substantial diversity are given due importance, balancing exploration and engagement.

Experimental Results

The methodology is tested across five benchmark datasets: Movielens-1M, Coat, KuaiRec-2.0, Netflix-Prize, and Yahoo-R2. The experimental setup involved splitting user interactions into training and test sets and evaluating key metrics such as Recall, Precision, and Hit-Ratio. Various competing algorithms, including traditional relevance-based methods, MMR, and DPP, were tested as baselines.

Numerical Results and Analysis:

• Quality-Diversity Trade-Off: The proposed strategy achieved superior performance on all datasets, particularly excelling in achieving a better balance of relevance and diversity compared to baseline methods.
• Diversity Scores: The scores for both coverage and distance-based diversity measures were consistently higher for the proposed methods. Incremental improvements were observed over state-of-the-art techniques, signifying the effectiveness of the copula-based blending strategy.
• Impact of Relevance Consideration: An ablation study verified the necessity of integrating relevance with diversity. Strategies considering only diversity showed reduced user engagement, underscoring the strategic importance of balancing the two measures for effective recommendations.

Implications and Future Work

The findings from this paper hold significant practical and theoretical implications. Practically, the proposed framework could transform recommendation strategies in various domains like content streaming, e-commerce, and news recommendation by ensuring users are presented with diverse and engaging content. Theoretically, it paves the way for exploring more integrative and sophisticated models that consider multifaceted user behavior aspects and diverse metrics.

Future Research Directions:

• Refinement of User Models: More granular user behavior models, accounting for dynamic adjustments and context-aware engagement, could offer further improvements.
• Exploration of Additional Metrics: Incorporating other metrics such as serendipity, novelty, and fairness into the copula-based strategy could deliver more balanced and user-centric recommendations.
• Addressing Relevance Score Inaccuracies: Adapting user behavior models to handle inaccuracies in relevance scores by integrating stochastic factors.

In summary, this paper proposes a robust and user-centric approach to balancing relevance and diversity in recommender systems, demonstrating significant improvements over existing methodologies. The framework stands to impact both academic research and practical implementations in recommender systems.