Modeling Actions of PubMed Users with N-Gram Language Models

Area of Science:

• Information Science
• Medical Informatics
• Computer Science

Background:

• Transaction logs from online search engines offer insights into user information-seeking behavior.
• Log data is crucial for training user models to improve information retrieval systems.

Purpose of the Study:

• To analyze user activities within a specialized search engine, PubMed.
• To model user search behavior using n-gram language models.

Main Methods:

• Encoding user actions from PubMed transaction logs as string sequences.
• Applying n-gram language models to these sequences.
• Evaluating model performance using perplexity and sequence prediction tasks.

Main Results:

• Developed n-gram language models to represent PubMed user search sequences.
• Demonstrated the models' ability to capture user search patterns.
• Identified potential for improving the PubMed search experience.

Conclusions:

• N-gram language models effectively represent user search behavior in specialized domains like PubMed.
• Understanding search patterns can lead to significant improvements in information retrieval systems.
• This research provides a foundation for enhancing user search experiences in biomedical literature databases.