MgRec: Multi-granularity intent interaction based sequential recommendation method

doi:10.1016/j.neucom.2026.134267

MgRec: Multi-granularity intent interaction based sequential recommendation method

2026-06-10

0

PRE

AI

DOI：10.1016/j.neucom.2026.134267

Original

Original request for help

Save

Abstract

En 中文

Traditional sequential recommendation methods focus on user behavior but often ignore the underlying latent intents. This paper presents Multi-granularity Intent Interaction based Sequential Recommendation Method, a novel multi-granularity framework designed to explicitly model these intents. At a fine-grained level, a Transformer captures specific user preferences from interaction sequences. Concurrently, at a coarse-grained level, it leverages a Knowledge Graph (KG) to explore broader item relationships and a Neural Topic Model (NTM) to distill these into abstract intents. By dynamically fusing these two granularities, MgRec achieves state-of-the-art (SOTA) performance, generating more accurate recommendations while effectively alleviating the popularity bias.

MgRec: Multi-granularity intent interaction based sequential recommendation method

Abstract

Journal

Organization