A novel model on improving Chinese dialogue summarization with multi-perspective information enhancement

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Dialogue summarization faces challenges with topic shifts and multi-role interactions.
• Existing methods struggle with accurate topic tracking and error propagation.
• Limited exploration of multi-dimensional role interaction information in current models.

Purpose of the Study:

• To improve Chinese dialogue summarization by enhancing multi-perspective information.
• To address limitations in topic tracking and role interaction in existing summarization models.
• To develop a novel model that effectively synthesitsizes diverse interaction information.

Main Methods:

• A novel topic-guided tracker to implicitly learn topic information via topic loss.
• An adaptive attention fusion method combining contextual and role attention distributions.
• Integration of discourse context features with hybrid attention for summary generation.

Main Results:

• The proposed model significantly outperforms strong baselines on CSDS and MC datasets.
• Demonstrated effectiveness in incorporating multi-perspective information for improved summary quality.
• Experimental results confirm superior performance in tracking topics and role interactions.

Conclusions:

• The novel model effectively enhances Chinese dialogue summarization through multi-perspective information.
• The topic-guided tracker and adaptive attention fusion are key to improved performance.
• This approach offers a significant advancement in generating high-quality dialogue summaries.