MAPS: Motivation-Aware Personalized Search via LLM-Driven Consultation Alignment

TL;DR

MAPS proposes motivation-aware personalized search by leveraging consultations to infer user intent and align it with product features. It introduces three components: ID-text representation fusion via LLMs with a Mixture of Attention Experts, a mapping-based general alignment to connect tokens with items, and a sequence-based personalized alignment that links motivation signals to current queries and item history. The model optimizes a dual objective combining contrastive general alignment and motivation-aware personalized alignment, achieving superior ranking and retrieval on real and synthetic datasets. This approach enhances e-commerce search by grounding recommendations in explicit motivation signals and cross-source semantics, with practical implications for more context-aware, user-centric search systems.

Abstract

Personalized product search aims to retrieve and rank items that match users' preferences and search intent. Despite their effectiveness, existing approaches typically assume that users' query fully captures their real motivation. However, our analysis of a real-world e-commerce platform reveals that users often engage in relevant consultations before searching, indicating they refine intents through consultations based on motivation and need. The implied motivation in consultations is a key enhancing factor for personalized search. This unexplored area comes with new challenges including aligning contextual motivations with concise queries, bridging the category-text gap, and filtering noise within sequence history. To address these, we propose a Motivation-Aware Personalized Search (MAPS) method. It embeds queries and consultations into a unified semantic space via LLMs, utilizes a Mixture of Attention Experts (MoAE) to prioritize critical semantics, and introduces dual alignment: (1) contrastive learning aligns consultations, reviews, and product features; (2) bidirectional attention integrates motivation-aware embeddings with user preferences. Extensive experiments on real and synthetic data show MAPS outperforms existing methods in both retrieval and ranking tasks.

Figures (6)

• Figure 1: Illustration of the user making multiple search attempts to find the best option.
• Figure 2: Examples of consultations with the corresponding search queries and the proportion of search sessions with related consultations (classified as "Lenient", "Moderate", or "Strict" under predefined NLP rules, detailed in App. \ref{['app:intro:relate:cons']}) in a real e-commerce platform equipped with AI consultation services.
• Figure 3: Overview of MAPS. ① denotes ID-text representation fusion with LLM. ② denotes the general alignment. ③ denotes the personalized alignment.
• Figure 4: Ranking performance on Amazon with different threshold $t$ in Eq. \ref{['eq:threshold']}. The default one is 2.
• Figure 5: Examples of consultations on the Amazon dataset.