Result Diversification in Search and Recommendation: A Survey

TL;DR

This survey addresses the need for diversity in retrieval by presenting a unified taxonomy that spans both search and recommendation. It systematically categorizes metrics into relevance-oblivious and relevance-aware families, detailing distance-based, coverage-based, and social-welfare measures, as well as IA- and D-family metrics for relevance-aware scenarios. It also reviews a broad set of offline (pre-, in-, post-processing) and online (bandits, reinforcement learning) approaches to enhance diversity, with representative methods such as MMR, DPP, and IA-/D-family formulations. The paper highlights open questions including time-aware diversity, direct optimization of non-differentiable metrics, explainability, multi-stakeholder trade-offs, and multi-modal diversity, outlining a concrete agenda for future research and deployment in real-world systems.

Abstract

Diversifying return results is an important research topic in retrieval systems in order to satisfy both the various interests of customers and the equal market exposure of providers. There has been growing attention on diversity-aware research during recent years, accompanied by a proliferation of literature on methods to promote diversity in search and recommendation. However, diversity-aware studies in retrieval systems lack a systematic organization and are rather fragmented. In this survey, we are the first to propose a unified taxonomy for classifying the metrics and approaches of diversification in both search and recommendation, which are two of the most extensively researched fields of retrieval systems. We begin the survey with a brief discussion of why diversity is important in retrieval systems, followed by a summary of the various diversity concerns in search and recommendation, highlighting their relationship and differences. For the survey's main body, we present a unified taxonomy of diversification metrics and approaches in retrieval systems, from both the search and recommendation perspectives. In the later part of the survey, we discuss the open research questions of diversity-aware research in search and recommendation in an effort to inspire future innovations and encourage the implementation of diversity in real-world systems.

Figures (6)

• Figure 1: The statistics of publications related to diversity in search and recommendation with the publication year and venue.
• Figure 2: Diversity in search and recommendation. Pink boxes indicate that they were originally proposed and generally used in search, while blue boxes indicate that they are generally used in the recommendation.
• Figure 3: A toy example to show that the individual-level diversity and system-level diversity in the recommendation are different concerns with little overlap. In this illustration, different shapes refer to different categories. Consider a top-3 recommendation and assume that there is an extremely large number of users and items in the system (the same as the real-world scenarios). In case 1, the system recommends the same 3 categories of items to each user; in case 2, the system always recommends a unique category of items to each unique user. Therefore, in case 1, the individual-level diversity is high and the system-level diversity is low, while in case 2, the individual-level diversity is low and the system-level diversity is high.
• Figure 4: Diversity metrics in search and recommendation. We unify the metrics in one unified classification since all metrics can be theoretically used in both fields. Metrics in pink boxes indicate that they were originally proposed and generally used in search, while those in blue boxes indicate that they are generally used in recommendation.
• Figure 5: Diversity approaches in search and recommendation, from both offline and online perspectives. Approaches in blue boxes indicate that they are generally used in recommendation, the pink box indicates it is generally used in search, while those in yellow boxes are equally widely used in search and recommendation.