Semantic similarity measure in biomedical domain leverage web search engine

Area of Science:

• Natural Language Processing
• Information Retrieval
• Biomedical Informatics

Background:

• Semantic similarity measurement is crucial for Information Retrieval (IR) and Natural Language Processing (NLP).
• Existing semantic similarity measures face challenges in accuracy and robustness, particularly in specialized domains like biomedicine.
• Previous research has explored various methods, but a gap remains for effective, scalable solutions.

Purpose of the Study:

• To propose and evaluate a novel page-count-based semantic similarity measure.
• To apply this measure within the biomedical domain, addressing specific challenges in this field.
• To integrate diverse similarity scores using machine learning for enhanced performance.

Main Methods:

• Utilizing web search engine page counts for querying individual terms (P, Q) and their conjunction (P AND Q).
• Developing a novel approach combining page counts with lexico-syntactic patterns for semantic similarity computation.
• Integrating multiple similarity scores via support vector machines (SVM) to improve measure robustness.

Main Results:

• Achieved a correlation coefficient of 0.798 on the A. Hliaoutakis dataset.
• Obtained a correlation coefficient of 0.705 with physician scores on the T. Pedersen dataset.
• Reached a correlation coefficient of 0.496 with expert scores on the T. Pedersen et al. dataset.

Conclusions:

• The proposed page-count-based semantic similarity measure demonstrates effectiveness, particularly when integrated with lexico-syntactic patterns and machine learning.
• The approach shows promise for improving semantic similarity calculations in biomedical information retrieval and NLP tasks.
• Experimental results validate the robustness and applicability of the developed measure across different datasets and scoring criteria.