Calculating semantic relatedness for biomedical use in a knowledge-poor environment

Area of Science:

• Biomedical Informatics
• Computational Linguistics
• Natural Language Processing

Background:

• Computing semantic relatedness for biomedical concepts is vital for knowledge extraction.
• Existing methods often require specialized, resource-intensive knowledge bases.
• This limits their applicability in diverse real-world scenarios.

Purpose of the Study:

• To present a novel, resource-efficient method for calculating semantic relatedness in knowledge-poor environments.
• To develop a generic and flexible approach for semantic relatedness computation.
• To demonstrate comparable or superior performance to state-of-the-art methods.

Main Methods:

• Compares vocabulary of best-fit documents associated with input texts.
• Utilizes a pre-processing step for efficient keyterm extraction on a per-document basis.
• Performs computations in a compact vector space defined by informative extraction results.

Main Results:

• Evaluated on five benchmarks, achieving correlation coefficients with average human answers.
• Demonstrated utility on Gene-Disease and Disease-Disease data pairs for analysis.
• Experimental results show performance comparable to, and often surpassing, state-of-the-art methods.

Conclusions:

• The proposed method enhances flexibility without compromising quality.
• It serves as a viable alternative to constructing expensive, specialized knowledge resources.
• The approach is suitable for knowledge-poor environments and diverse applications.