On Mitigating Data Sparsity in Conversational Recommender Systems

TL;DR

This paper introduces DACRS, a novel conversational recommender system that addresses data sparsity by augmenting dialogues, leveraging knowledge graphs for entity modeling, and fusing dialogue with entity embeddings to better capture user preferences.

Contribution

DACRS is the first model to integrate dialogue augmentation, knowledge-guided entity modeling, and dialogue-entity matching for improved CRS performance under data sparsity.

Findings

DACRS achieves state-of-the-art results on two public datasets.

The dialogue augmentation improves model generalization.

Knowledge-guided entity modeling enhances entity embedding expressiveness.

Abstract

Conversational recommender systems (CRSs) capture user preference through textual information in dialogues. However, they suffer from data sparsity on two fronts: the dialogue space is vast and linguistically diverse, while the item space exhibits long-tail and sparse distributions. Existing methods struggle with (1) generalizing to varied dialogue expressions due to underutilization of rich textual cues, and (2) learning informative item representations under severe sparsity. To address these problems, we propose a CRS model named DACRS. It consists of three modules, namely Dialogue Augmentation, Knowledge-Guided Entity Modeling, and Dialogue-Entity Matching. In the Dialogue Augmentation module, we apply a two-stage augmentation pipeline to augment the dialogue context to enrich the data and improve generalizability. In the Knowledge-Guided Entity Modeling, we propose a knowledge graph (KG) based entity substitution and an entity similarity constraint to enhance the expressiveness of entity embeddings. In the Dialogue-Entity Matching module, we fuse the dialogue embedding with the mentioned entity embeddings through a dialogue-guided attention aggregation to acquire user embeddings that contain both the explicit and implicit user preferences. Extensive experiments on two public datasets demonstrate the state-of-the-art performance of DACRS.