A framework for structured semantic representation capable of active sensing and interpretable inference: A cancer...

Area of Science:

• Computational linguistics
• Bioinformatics
• Natural Language Processing

Background:

• Accurate semantic representation is crucial for machine comprehension of biomedical literature.
• Existing methods struggle with inter-sentence relations, contextual information, and merging expert-curated semantics.
• Challenges include lack of contextual information and difficulty in interpretable semantic inference.

Purpose of the Study:

• To present a novel Framework for Structured Semantic Representation (FSSR) addressing limitations in current semantic modeling.
• To enhance machine comprehension of biomedical text by accurately modeling inter-sentence relations and context.
• To facilitate interpretable semantic inference and the merging of diverse semantic knowledge.

Main Methods:

• FSSR employs a double-layer Construct structure combining Paradigm and Instance for word/sentence semantics.
• Six rule types define semantic relations between sentence Constructs; a Computational Model represents actions.
• A graph-based approach represents semantics, with nodes as Constructs/Paradigms and edges as rules.

Main Results:

• The FSSR framework enables interpretable inference and active acquisition of new information.
• A case study successfully modeled semantics of a cancer prognostic article, reproducing its results and charts.
• A website visualizes the FSSR inference process, demonstrating its practical application.

Conclusions:

• FSSR offers a robust solution for structured semantic representation in biomedical literature.
• The framework improves contextual understanding and facilitates interpretable semantic inference.
• FSSR shows promise for advancing machine comprehension and knowledge integration in specialized domains.