A semantic enhancement-based multimodal network model for extracting information from evidence lists.

Area of Science:

• Legal Technology
• Artificial Intelligence
• Information Extraction

Background:

• Courts need accurate information extraction from diverse legal evidence for informed decision-making.
• Manual evidence screening is time-consuming, error-prone, and inadequate for large datasets.
• Existing methods struggle with the complexity and heterogeneity of legal evidence.

Purpose of the Study:

• To develop an automated model for accurate critical information extraction from legal evidence lists.
• To enhance the semantic understanding and fusion of multimodal information within legal documents.
• To improve the efficiency and accuracy of legal evidence analysis for judicial processes.

Main Methods:

• Construction of an entity semantic graph based on entity category differences.
• Multimodal feature extraction and guided fusion using entity semantic graphs.
• Application of an improved multimodal self-attention mechanism for feature interaction.
• Utilization of a hybrid loss function (Taylor polynomials and supervised contrast learning) to minimize information loss.

Main Results:

• The proposed semantic enhancement-based multimodal network model (SEBM) demonstrated superior performance.
• SEBM outperformed existing high-performing models in extracting critical information from authentic Chinese evidence lists.
• The model effectively handles diverse case types and extensive entity details across multiple law firms.

Conclusions:

• SEBM offers a robust and accurate solution for automated information extraction in the legal domain.
• The model's approach to multimodal feature fusion and semantic enhancement significantly improves extraction accuracy.
• This technology has the potential to streamline legal judgment and knowledge-driven decision-making.