A Study of the Morpho-Semantic Relationship in Medline

Area of Science:

• Computational Biology
• Bioinformatics
• Natural Language Processing

Background:

• Traditional English morphological analysis focuses on inflections and derivations.
• Recent advancements involve automatic corpus analysis for rule derivation.
• The biological literature presents unique challenges and opportunities for morphological analysis.

Purpose of the Study:

• To adapt and apply morphological analysis techniques to the biological literature.
• To develop a novel computational approach for assessing semantic relatedness between tokens.
• To leverage character string similarity for predicting token relatedness.

Main Methods:

• Utilized a large-scale corpus of over 84 million sentences from the MEDLINE database.
• Analyzed over 2.3 million unique token types within the MEDLINE corpus.
• Developed an algorithm to assign probabilities of semantic relatedness based on character string similarity for token pairs.

Main Results:

• Identified over 36 million token type pairs with a semantic relatedness probability of at least 0.7.
• The method demonstrated high accuracy in predicting semantic relatedness based on string similarity.
• Manual evaluations confirmed the good quality of the generated predictions.

Conclusions:

• The proposed computational method offers an effective way to uncover semantic relationships in biological texts.
• Character string similarity is a viable feature for predicting semantic relatedness in specialized scientific literature.
• This approach has significant implications for knowledge discovery and information retrieval in bioinformatics.