Source-linked AI summary

Towards Topic-Guided Conversational Recommender System

Kun Zhou, Yuanhang Zhou, Wayne Xin Zhao, Xiaoke Wang, Ji-Rong Wen

arXiv:2010.04125v2cs.CLcs.HCcs.IR

TL;DR

Existing CRS datasets provide limited proactive guidance from ordinary conversation into recommendation. The paper introduces TG-ReDial, a real-data, semi-automatic dataset organized by topic threads, defines a three-subtask topic-guided recommendation task, and reports effective performance across those subtasks.

  • Problem

    Existing CRS datasets mainly assume immediate user requests and do not adequately study explicit semantic transitions into recommendation.

  • Method

    The paper constructs TG-ReDial from real user data with topic threads and presents a solution using multiple data signals for item recommendation, topic prediction, and response generation.

  • Results

    Extensive experiments demonstrate the proposed approach’s effectiveness on item recommendation, topic prediction, and response generation.

  • Takeaways & Limitations

    TG-ReDial provides a dataset and testbed for conversational recommendation with natural topic transitions and historical user interaction data.

  • Takeaways & Limitations

    The potential of TG-ReDial has not been fully explored, and further work is needed on additional tasks and more effective approaches.

Abstract

from arXiv · show

Conversational recommender systems (CRS) aim to recommend high-quality items to users through interactive conversations. To develop an effective CRS, the support of high-quality datasets is essential. Existing CRS datasets mainly focus on immediate requests from users, while lack proactive guidance to the recommendation scenario. In this paper, we contribute a new CRS dataset named \textbf{TG-ReDial} (\textbf{Re}commendation through \textbf{T}opic-\textbf{G}uided \textbf{Dial}og). Our dataset has two major features. First, it incorporates topic threads to enforce natural semantic transitions towards the recommendation scenario. Second, it is created in a semi-automatic way, hence human annotation is more reasonable and controllable. Based on TG-ReDial, we present the task of topic-guided conversational recommendation, and propose an effective approach to this task. Extensive experiments have demonstrated the effectiveness of our approach on three sub-tasks, namely topic prediction, item recommendation and response generation. TG-ReDial is available at https://github.com/RUCAIBox/TG-ReDial.

1 Introduction

Existing CRS datasets often assume immediate user requests and do not adequately model proactive semantic transitions into recommendation. TG-ReDial addresses this gap with a semi-automatically constructed, topic-guided dataset and a three-subtask recommendation framework.

  • Existing CRS datasets largely assume clear, immediate requests and underrepresent proactive transitions from non-recommendation conversation to recommendation.
  • TG-ReDial contains 10,000 two-party movie-domain dialogues between seekers and recommenders.
  • Topic threads guide each conversation from an initial non-recommendation topic toward recommendation through evolving topics and chit-chat.
  • Real user data supports recommended movies, topic threads, recommendation reasons, and user identities, while annotators revise, polish, or rewrite conversation data.
  • The paper releases TG-ReDial and proposes a Transformer-based solution that leverages multiple data signals across the three subtasks.
  • Topic-guided conversational recommendation comprises item recommendation, topic prediction, and response generation.

2 Related Work

Prior conversational systems span task-oriented and chit-chat settings, while CRS research has produced datasets with varying degrees of synthetic construction and human annotation. TG-ReDial differs from DuRecDial by modeling content evolution with topic threads and using real user-related data.

  • Conversation systems include task-oriented systems for specific goals and chit-chat systems for general-purpose dialogue.
  • Existing CRS datasets include synthetic template-based resources and human-annotated datasets such as ReDial, GoReDial, and DuRecDial.
  • DuRecDial uses goal sequences spanning multiple task types, whereas TG-ReDial uses topic threads to characterize evolving content flow.
  • TG-ReDial is designed to be easier to integrate with open-domain dialogue because its topic threads describe content evolution rather than task switching.
  • Unlike DuRecDial, TG-ReDial derives user-related data from real users rather than relying on annotators to generate it.

3 Dataset Construction

TG-ReDial is built from real movie-watching and user data, topic-linked recommendations, and semi-automatic human annotation. The resulting dataset combines guided dialogue construction with user profiles, interaction histories, and privacy protections.

  • Dataset setting: The two-party setting begins with chit-chat, proactively moves toward a target topic, and then recommends a movie using the seeker’s interests.
  • Collection process: Three movies sharing the “family” tag illustrate how film records, ConceptNet topic threads, and candidate sentences are combined during collection.
  • Real-world data: Conversations are associated with real Douban users, enabling watching records, user profiles, and review-derived recommendation reasons to support recommendation.
  • Movie and tag preparation: Movies are grouped into coherent watching subsequences and annotated with meaningful tags drawn from original metadata, reviews, and manual selection.
  • Creating topic threads: Topic threads connect an initial greeting to a recommended movie’s target tag by traversing ConceptNet and using shortest paths identified with depth-first search.
  • Human annotation: Annotators generate utterances from retrieved topic-relevant candidates and revise them for dialogue consistency, while recommendation reasons come from retrieved review sentences.
  • Dataset statistics: TG-ReDial contains 129,392 utterances from 1,482 users; dialogues average 7.9 topics and 19.0 words per utterance.
  • Privacy: Privacy protection restricts sampling to users with many watching records and anonymizes or modifies derived user data before excluding identifiable users.

4 Our Approach

The approach formulates topic-guided conversational recommendation as coordinated topic prediction, item recommendation, and response generation. It combines conversation, topic, profile, interaction, and pretrained language-model signals across specialized modules.

  • Task formulation: The task uses user profile, interaction history, prior utterances, and topics to predict the next topic, recommend an item, and generate a response.
  • Recommendation module: The recommendation module combines BERT representations of historical utterances with SASRec representations of the user interaction sequence.These signals form the user representation used for recommendation.
  • Recommendation module: The recommendation module ranks items using their learned embeddings and the user representation, selecting the item with the largest probability.
  • Dialog module: The dialog module generates responses for topic guidance or item recommendation using specific models, including GPT-2 for response generation.
  • Dialog module: The topic prediction model uses conversation-BERT, topic-BERT, and profile-BERT to encode historical utterances, topic history, and the user profile.The model ranks candidate topics and selects the one with the largest probability.

5 Experiments

Experiments evaluate the approach on item recommendation, topic prediction, and response generation using TG-ReDial and task-specific baselines and metrics. The proposed model outperforms the baselines across the reported tasks, with its gains attributed to combining relevant conversational, interaction, topic, and item signals.

  • Experimental Setup: TG-ReDial is split into training, validation, and test sets with an 8:1:1 ratio, and evaluation covers item recommendation, topic prediction, and response generation.The model generates replies or recommendations turn by turn from each conversation's first utterance.
  • Item Recommendation: Item recommendation uses NDCG@k and MRR@k with k = 10 and 50 to rank all possible items.
  • Item Recommendation: The proposed model significantly outperforms all item-recommendation baselines by combining historical utterances with interaction sequences.The comparison includes CRS, interaction-history, and text-based baselines; BERT outperforms TextCNN among the text-based models.
  • Topic Prediction: Topic prediction uses Hit@k with k = 1, 3, and 5, comparing heuristic, recurrent, BERT-based, and target-topic ablation baselines.The task predicts the next topic leading toward the recommendation target.
  • Topic Prediction: The proposed topic-prediction model outperforms all baselines, while removing the target topic significantly decreases performance.The model jointly encodes historical utterances, topics, and user profiles with different BERT models.
  • Response Generation: For response generation, the proposed model outperforms all baselines in most cases, especially for perplexity, Distinct, and human evaluation.Transformer, GPT-2, and the proposed model have similar BLEU scores, while predicted topics or items enhance generated text quality in the proposed approach.

6 Conclusion

The paper introduces TG-ReDial, a human-annotated dataset based on real-world user data, and a topic-guided conversational recommendation task with an effective solution. Experiments demonstrate effectiveness across three sub-tasks, while the dataset's broader potential remains underexplored.

  • TG-ReDial is a high-quality conversational-recommender dataset constructed through human annotation based on real-world user data.
  • The paper presents topic-guided conversational recommendation and a solution evaluated on item recommendation, topic prediction, and response generation.
  • The proposed approach demonstrates effectiveness on all three sub-tasks in extensive experiments.
  • TG-ReDial's potential has not been fully explored, including possible applications to personalized chit-chat, target-guided conversation, and sequential recommendation.The authors identify broader task exploration and more effective approaches as future work.
Loading 2010.04125v2…