SKIMMR: facilitating knowledge discovery in life sciences by machine-aided skim reading

Area of Science:

• Biomedical Informatics
• Information Retrieval
• Text Mining

Background:

• Biomedical research generates a continuous influx of articles, overwhelming researchers.
• Traditional reading methods are insufficient for keeping pace with this information explosion.
• A need exists for tools that facilitate rapid, yet informative, content assimilation.

Purpose of the Study:

• To develop a system that emulates human 'skim-reading' for biomedical literature.
• To assist researchers and clinicians in efficiently processing large volumes of scientific articles.
• To provide a dynamic, graph-based interface for exploring entity networks within and across documents.

Main Methods:

• Utilized shallow parsing, co-occurrence analysis, and semantic similarity for entity extraction.
• Constructed weighted binary statements (co-occurrence and similarity) using point-wise mutual information and cosine distance.
• Developed fuzzy indices and a graph-based interface for querying and browsing entity networks and relevant articles.
• Implemented automated experimental evaluation against curated datasets (PubMed, TREC, MeSH).

Main Results:

• Introduced SKIMMR, a web-based prototype generating interactive entity networks from documents.
• The system links network areas of interest to the most relevant source articles.
• Demonstrated practical applicability in Spinal Muscular Atrophy and Parkinson's Disease research.
• Experimental evaluation showed SKIMMR outperforms PubMed in focused browsing and context informativeness.

Conclusions:

• SKIMMR offers a novel methodology for machine-aided skim-reading of biomedical literature.
• The graph-based approach enhances the efficiency and informativeness of literature browsing.
• This tool can significantly aid biomedical professionals in managing and synthesizing research information.