[Paper Review] Improving Multi-turn Emotional Support Dialogue Generation with Lookahead Strategy Planning

Yi Cheng, Wenge Liu+6 more|arXiv (Cornell University)|Oct 9, 2022

Speech and dialogue systems4 citations

TL;DR

This paper proposes MultiESC, a multi-turn emotional support dialogue system that uses A*-inspired lookahead heuristics to plan support strategies with long-term impact, while dynamically modeling user state through subtle emotion expressions and emotion causes. It significantly outperforms strong baselines in both response quality and strategy planning, demonstrating the effectiveness of foresighted strategy selection in emotional support conversations.

ABSTRACT

Providing Emotional Support (ES) to soothe people in emotional distress is an essential capability in social interactions. Most existing researches on building ES conversation systems only considered single-turn interactions with users, which was over-simplified. In comparison, multi-turn ES conversation systems can provide ES more effectively, but face several new technical challenges, including: (1) how to adopt appropriate support strategies to achieve the long-term dialogue goal of comforting the user's emotion; (2) how to dynamically model the user's state. In this paper, we propose a novel system MultiESC to address these issues. For strategy planning, drawing inspiration from the A* search algorithm, we propose lookahead heuristics to estimate the future user feedback after using particular strategies, which helps to select strategies that can lead to the best long-term effects. For user state modeling, MultiESC focuses on capturing users' subtle emotional expressions and understanding their emotion causes. Extensive experiments show that MultiESC significantly outperforms competitive baselines in both dialogue generation and strategy planning. Our codes are available at https://github.com/lwgkzl/MultiESC.

Motivation & Objective

• To address the limitations of single-turn emotional support dialogue systems by enabling multi-turn interactions with strategic planning.
• To improve long-term emotional support effectiveness by forecasting future user feedback through lookahead heuristics.
• To dynamically model user state by capturing subtle emotional expressions and identifying emotion causes.
• To develop a system that aligns with psychological support procedures, such as inquiry before reassurance.
• To demonstrate that foresighted strategy planning enhances both dialogue quality and support efficacy.

Proposed method

• Adopting A*-inspired lookahead heuristics to estimate the expected future user feedback after applying a sequence of support strategies.
• Implementing a strategy sequence generator to predict the probability distribution over future strategy sequences.
• Using a user feedback predictor to estimate the emotional response to a given strategy sequence, enabling long-horizon planning.
• Incorporating the NRC VAD lexicon to model subtle emotional intensity changes in user expressions.
• Identifying emotion causes (e.g., social rejection) to deepen situational understanding and improve relevance of responses.
• Integrating these components into a unified framework, MultiESC, to jointly optimize strategy selection and response generation.

Experimental results

Research questions

• RQ1 Can lookahead strategy planning improve long-term emotional support effectiveness in multi-turn dialogue systems?
• RQ2 How can a dialogue system dynamically model user state beyond coarse-grained emotion detection?
• RQ3 Does incorporating emotion cause identification enhance the relevance and empathy of generated responses?
• RQ4 Can A*-like heuristic search improve strategy selection compared to reactive or greedy approaches?
• RQ5 To what extent does foresighted planning lead to more effective and human-aligned emotional support responses?

Key findings

• MultiESC significantly outperforms strong baselines, including BlenderBot-Joint and MoEL, in both automatic and human evaluation metrics.
• The model achieves a 12.3% absolute improvement in feedback score over the baseline 'w/o lookahead', demonstrating the value of long-horizon planning.
• The ablation study shows that removing lookahead heuristics leads to a notable drop in strategy planning quality, confirming its contribution.
• Case studies reveal that MultiESC proactively inquires about user experiences early in the conversation, aligning with psychological support procedures.
• The model generates more empathetic and contextually consistent responses than variants without strategy planning or lookahead.
• Performance saturates when considering more than six strategy sequences, indicating diminishing returns beyond a certain planning depth.