[Paper Review] Deep Conversational Recommender in Travel

Lizi Liao, Ryuichi Takanobu+4 more|arXiv (Cornell University)|Jun 25, 2019

Topic Modeling43 references33 citations

TL;DR

The paper introduces DCR, a deep conversational recommender for travel that combines a neural topic component with a GCN-based venue recommender and a pointer-based response integration to handle multi-subtask travel dialogs.

ABSTRACT

When traveling to a foreign country, we are often in dire need of an intelligent conversational agent to provide instant and informative responses to our various queries. However, to build such a travel agent is non-trivial. First of all, travel naturally involves several sub-tasks such as hotel reservation, restaurant recommendation and taxi booking etc, which invokes the need for global topic control. Secondly, the agent should consider various constraints like price or distance given by the user to recommend an appropriate venue. In this paper, we present a Deep Conversational Recommender (DCR) and apply to travel. It augments the sequence-to-sequence (seq2seq) models with a neural latent topic component to better guide response generation and make the training easier. To consider the various constraints for venue recommendation, we leverage a graph convolutional network (GCN) based approach to capture the relationships between different venues and the match between venue and dialog context. For response generation, we combine the topic-based component with the idea of pointer networks, which allows us to effectively incorporate recommendation results. We perform extensive evaluation on a multi-turn task-oriented dialog dataset in travel domain and the results show that our method achieves superior performance as compared to a wide range of baselines.

Motivation & Objective

• Motivate building an intelligent travel agent that handles multi-subtask dialogs (hotels, restaurants, taxis, etc.).
• Propose a hybrid model that combines global topic control with a GCN-based venue recommender to respect constraints and context.
• Develop an integration mechanism to incorporate recommendation results into response generation.
• Evaluate the model on a multi-turn travel dialog dataset and show improvements over baselines.

Proposed method

• Extend a seq2seq backbone with a neural latent topic component to capture global dialog semantics and enable within-topic generation.
• Use a two-layer graph convolutional network to learn venue representations and sponsor matching scores with dialog context.
• Integrate topic-driven generation and GCN-based recommendations via a pointer-network-inspired mechanism that uses a sentinel token to switch between word generation and venue insertion.
• Train the topic component with variational inference and a topic-conditioned decoding bias, plus a KL-divergence term.
• Train the GCN-based recommender to optimize venue matching to dialog contexts using a cross-entropy objective over dialog-venue pairs.
• Fine-tune the joint model end-to-end with a weighted combination of topic and GCN losses.

Experimental results

Research questions

• RQ1 RQ1: Can DCR generate appropriate, travel-domain responses and recommendations in multi-turn dialogs?
• RQ2 RQ2: Does the global topic control component improve response coherence and topic consistency?
• RQ3 RQ3: Does the GCN-based venue recommender effectively capture venue relations and constraints to improve recommendations?

Key findings

• DCR outperforms baselines on corpus-based metrics BLEU and Entity Accuracy in the travel domain.
• The topic component helps generate within-topic, coherent responses.
• The GCN-based recommender captures venue attributes and relationships, improving recommendation quality.
• The pointed integration mechanism effectively combines topic-driven generation with recommendation results.