Centrality-aware Product Retrieval and Ranking

TL;DR

A User-intent Centrality Optimization (UCO) approach for existing models, which optimizes for the user intent in semantic product search, and proposes a dual-loss based optimization to handle hard negatives, i.e., product titles that are semantically relevant but do not reflect the user’s intent.

Abstract

This paper addresses the challenge of improving user experience on e-commerce platforms by enhancing product ranking relevant to users' search queries. Ambiguity and complexity of user queries often lead to a mismatch between the user's intent and retrieved product titles or documents. Recent approaches have proposed the use of Transformer-based models, which need millions of annotated query-title pairs during the pre-training stage, and this data often does not take user intent into account. To tackle this, we curate samples from existing datasets at eBay, manually annotated with buyer-centric relevance scores and centrality scores, which reflect how well the product title matches the users' intent. We introduce a User-intent Centrality Optimization (UCO) approach for existing models, which optimises for the user intent in semantic product search. To that end, we propose a dual-loss based optimisation to handle hard negatives, i.e., product titles that are semantically relevant but do not reflect the user's intent. Our contributions include curating challenging evaluation sets and implementing UCO, resulting in significant product ranking efficiency improvements observed for different evaluation metrics. Our work aims to ensure that the most buyer-centric titles for a query are ranked higher, thereby, enhancing the user experience on e-commerce platforms.

Figures (6)

• Figure 1: Central Title: Thomas sabo charms with 18k Rose gold pearl
• Figure 2: Non-central title: Thomas Sabo charm club bracelet with detachable dragonfly charm
• Figure 3: The figure shows how the loss function algorithm works with hard negatives. The algorithm targets those non-central titles (red) that are inside the margin.
• Figure 4: Examples of query-title pairs from the CQ-common-str split. Both, positive and negative product titles have high semantic correlation to the user query, however only the positive product title exhibits a central idea/intent.
• Figure 5: Qualitative comparison of the proposed UCO on a sample from the CQ-common-str test set, when using the eBERT (siam) as the encoder backbone. We showcase the top-3 retrieved product titles for both encoders.