An adaptive term proximity based rocchio's model for clinical decision support retrieval

Area of Science:

• Medical Informatics
• Information Retrieval
• Natural Language Processing

Background:

• Connecting electronic health records (EHR) with biomedical literature is crucial for clinical decision-making.
• Pseudo Relevance Feedback (PRF) is effective for handling clinical jargon in EHR but often overlooks term importance and co-occurrence.
• Existing PRF models fail to consider both term importance and co-occurrence relationships in feedback documents.

Purpose of the Study:

• To develop an improved information retrieval model for clinical settings by integrating EHR data with biomedical literature.
• To propose a novel approach that considers both term importance and proximity in query expansion.
• To enhance the effectiveness of retrieving clinical support documents for medical professionals.

Main Methods:

• Incorporated the original HAL model into Rocchio's model to create a HAL-based Rocchio's model (HRoc) for query expansion.
• Introduced a new concept of term proximity feedback weight to better leverage co-occurrence information.
• Designed three normalization methods and an adaptive parameter for optimizing proximity information integration and window size selection.

Main Results:

• The HRoc and HRoc_AP models outperformed advanced models like PRoc2 and TF-PRF on the 2016 TREC Clinical Support medicine dataset.
• Achieved an 8.5% and 12.24% increase in Mean Average Precision (MAP) compared to PRoc2 and TF-PRF, respectively.
• Demonstrated a 7.86% and 9.88% increase in F1 score compared to PRoc2 and TF-PRF, respectively.

Conclusions:

• The HRoc model effectively enhances the precision and recall rate of information retrieval for clinical support documents.
• The HRoc_AP model, with its self-adaptive parameter, reduces hyper-parameters while maintaining performance, improving efficiency and applicability.
• The proposed models assist clinicians in effectively retrieving relevant clinical support information, aiding decision-making.