A Field Sensor: computing the composition and intent of PubMed queries.

Area of Science:

• Biomedical Informatics
• Information Retrieval

Background:

• PubMed is a critical resource for biomedical information, processing millions of queries daily.
• Effective information retrieval relies on understanding user query composition and intent.

Purpose of the Study:

• To introduce Field Sensor, a domain-specific tool designed to analyze PubMed query structure and predict user intent.
• To identify the types of fields users search for within PubMed records (e.g., title, author, journal).

Main Methods:

• Field Sensor employs a multi-step process: syntactic chunking, rule-based tagging, and probabilistic field prediction.
• The tool infers fields for tokens or token sequences within a query.
• Evaluation utilized both human-annotated (10,000 queries) and machine-annotated (103,000 queries) datasets.

Main Results:

• Field Sensor achieved high accuracy: 93% on the human-annotated corpus and 91% on the machine-annotated corpus.
• Nearly half of all PubMed searches were identified as navigational (e.g., author, title searches).
• The other half of searches were classified as informational (e.g., topical searches).

Conclusions:

• Field Sensor accurately deciphers PubMed query composition and user intent.
• The tool has been integrated into PubMed to offer relevance-sorted results for informational queries.
• Understanding query composition is vital for optimizing the PubMed search experience.