Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BARCOR: Towards A Unified Framework for Conversational Recommendation Systems

Ting-Chun Wang, Shang-Yu Su, Yun-Nung Chen

TL;DR

This work addresses the fragmentation between recommendation and response generation in conversational recommendation systems by introducing BARCOR, a unified framework built on BART. BARCOR jointly optimizes a graph-encoded CORG knowledge graph with a bidirectional recommender and an autoregressive response generator, enabling end-to-end learning. The authors also construct CORG, a lightweight, movie-domain KG sourced from Wikidata, and augment training with descriptive entities to bridge semantic gaps. Experiments on the ReDial dataset show state-of-the-art performance in both automatic and human evaluations, along with increased training stability and clear ablation-driven insights into the contributions of each component. Overall, BARCOR demonstrates that a tightly integrated, KG-informed Transformer architecture can outperform modular CRS pipelines in both accuracy and naturalness of conversations.

Abstract

Recommendation systems focus on helping users find items of interest in the situations of information overload, where users' preferences are typically estimated by the past observed behaviors. In contrast, conversational recommendation systems (CRS) aim to understand users' preferences via interactions in conversation flows. CRS is a complex problem that consists of two main tasks: (1) recommendation and (2) response generation. Previous work often tried to solve the problem in a modular manner, where recommenders and response generators are separate neural models. Such modular architectures often come with a complicated and unintuitive connection between the modules, leading to inefficient learning and other issues. In this work, we propose a unified framework based on BART for conversational recommendation, which tackles two tasks in a single model. Furthermore, we also design and collect a lightweight knowledge graph for CRS in the movie domain. The experimental results show that the proposed methods achieve the state-of-the-art performance in terms of both automatic and human evaluation.

BARCOR: Towards A Unified Framework for Conversational Recommendation Systems

TL;DR

This work addresses the fragmentation between recommendation and response generation in conversational recommendation systems by introducing BARCOR, a unified framework built on BART. BARCOR jointly optimizes a graph-encoded CORG knowledge graph with a bidirectional recommender and an autoregressive response generator, enabling end-to-end learning. The authors also construct CORG, a lightweight, movie-domain KG sourced from Wikidata, and augment training with descriptive entities to bridge semantic gaps. Experiments on the ReDial dataset show state-of-the-art performance in both automatic and human evaluations, along with increased training stability and clear ablation-driven insights into the contributions of each component. Overall, BARCOR demonstrates that a tightly integrated, KG-informed Transformer architecture can outperform modular CRS pipelines in both accuracy and naturalness of conversations.

Abstract

Recommendation systems focus on helping users find items of interest in the situations of information overload, where users' preferences are typically estimated by the past observed behaviors. In contrast, conversational recommendation systems (CRS) aim to understand users' preferences via interactions in conversation flows. CRS is a complex problem that consists of two main tasks: (1) recommendation and (2) response generation. Previous work often tried to solve the problem in a modular manner, where recommenders and response generators are separate neural models. Such modular architectures often come with a complicated and unintuitive connection between the modules, leading to inefficient learning and other issues. In this work, we propose a unified framework based on BART for conversational recommendation, which tackles two tasks in a single model. Furthermore, we also design and collect a lightweight knowledge graph for CRS in the movie domain. The experimental results show that the proposed methods achieve the state-of-the-art performance in terms of both automatic and human evaluation.
Paper Structure (29 sections, 4 equations, 4 figures, 6 tables)

This paper contains 29 sections, 4 equations, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset
  4. Preliminaries
  5. Problem Formulation
  6. CORG (COnversational Recommender Graphs)
  7. Data Source
  8. Information Collection
  9. Data Processing
  10. BARCOR
  11. Graph Encoder
  12. BART as Conversational Recommender
  13. Bidirectional Recommender
  14. Data Augmentation
  15. Auto-Regressive Response Generator
  16. ...and 14 more sections

Figures (4)

  • Figure 1: The proposed framework is composed of three components: (1) knowledge graphs for providing external knowledge, (2) a bidirectional encoder as the recommender, and (3) an auto-regressive decoder as the response generator.
  • Figure 2: A sample subgraph of CORG. CORG has 5 types of node entities and 5 types of relations, the statistics of types and relations are shown in Table \ref{['tab:corg-stats']}.
  • Figure 3: Recommendation performance of BARCOR and the baselines on the validation set at different training epochs.
  • Figure 4: Ablation study: Recommendation performance on the validation set at different training epochs.