VectorSage: enhancing PubMed article retrieval with advanced semantic search

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Information Retrieval

Background:

• The exponential growth of academic literature necessitates advanced search methods for efficient knowledge discovery.
• Traditional keyword-based search engines face challenges with linguistic variability and lexical ambiguity, leading to imprecise search results.
• Advances in natural language processing offer context-aware semantic similarity techniques to overcome limitations of basic text matching.

Purpose of the Study:

• To develop an advanced biomedical search system for retrieving PubMed abstracts.
• To improve the efficiency and precision of literature exploration and knowledge discovery.

Main Methods:

• Developed VectorSage, a biomedical search system utilizing a hybrid approach.
• Combined term relevance scoring with embedding-based semantic similarity for PubMed abstract retrieval.
• Implemented a global ranking mechanism to sort retrieved documents, balancing semantic relevance and keyword specificity.

Main Results:

• VectorSage provides an advanced system for retrieving PubMed abstracts.
• The hybrid approach enhances search relevance by integrating semantic similarity with term relevance.
• Global ranking improves the balance between semantic relevance and keyword specificity in search results.

Conclusions:

• VectorSage offers an effective solution for navigating the growing body of biomedical literature.
• The system facilitates efficient literature exploration and knowledge discovery through context-aware search.
• This approach addresses the limitations of traditional search engines in handling linguistic complexity.