Novel framework for dialogue summarization based on factual-statement fusion and dialogue segmentation

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• The increasing volume of dialogue data necessitates effective summarization techniques.
• Abstractive dialogue summarization aims to generate concise summaries that capture the essence of conversations.
• Existing methods often struggle with factual consistency and capturing the full context of dialogues.

Purpose of the Study:

• To propose a novel sequence-to-sequence framework, DS-SS (Dialogue Summarization with Factual-Statement Fusion and Dialogue Segmentation), for abstractive dialogue summarization.
• To enhance dialogue encoding by integrating factual statements and segmenting dialogues into topic-coherent units.
• To improve the factual consistency and informativeness of generated dialogue summaries.

Main Methods:

• Developed a novel sequence-to-sequence framework (DS-SS).
• Implemented factual statement extraction and fusion into the dialogue encoding process.
• Integrated a dialogue segmenter to divide conversations into topic-coherent segments.
• Conducted experiments on SAMSum and DialogSum datasets.

Main Results:

• The DS-SS framework demonstrated superior performance compared to strong baselines.
• Both automatic evaluation metrics and human evaluations confirmed the framework's effectiveness.
• Generated summaries exhibited enhanced factual consistency and informativeness.

Conclusions:

• The proposed DS-SS framework effectively addresses limitations in abstractive dialogue summarization.
• Fusing factual statements and dialogue segmentation are key innovations for improved summarization.
• The framework shows significant potential for real-world dialogue summarization applications.