Advanced multiple document summarization via iterative recursive transformer networks and multimodal transformer.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Machine Learning

Background:

• Digital information growth demands efficient multiple document summarization.
• Existing methods face challenges in coherence, multimodal data integration, and learning strategies.

Purpose of the Study:

• To develop novel neural architectures and methodologies for enhanced multiple document summarization.
• To improve summary coherence, information integration, and performance across diverse data types and domains.

Main Methods:

• Recursive transformer networks (ReTran) for enhanced textual dependency comprehension.
• Multimodal transformers with cross-modal attention for integrating text, images, and metadata.
• Actor-critic reinforcement learning and meta-learning for optimized training and zero-shot summarization.
• Knowledge-enhanced transformers for improved semantic coherence.

Main Results:

• ReTran achieved 5-10% ROUGE score improvement.
• Cross-modal summarization showed 8-12% enhancement in quality metrics.
• Actor-critic RL surpassed Q-learning by 5-8%.
• Meta-learning and knowledge-enhanced transformers improved performance by 6-12%.

Conclusions:

• The proposed methods significantly advance multiple document summarization capabilities.
• Novel architectures produce more informative, coherent summaries across diverse modalities.
• This work sets a new benchmark for future research and applications in automated summarization.