Combating health misinformation with fusion-based credible retrieval techniques

Area of Science:

• Information Science
• Health Informatics
• Computer Science

Background:

• Combating health misinformation is crucial for public health.
• Existing health information retrieval systems struggle to consistently deliver credible content.
• Data fusion techniques can enhance retrieval but require effective subset selection.

Purpose of the Study:

• To improve the retrieval of credible health information by enhancing data fusion performance.
• To evaluate clustering-based subset selection methods for optimizing data fusion in health information retrieval.
• To reduce health misinformation by increasing access to trustworthy information.

Main Methods:

• Evaluated five clustering methods (K-means variants, Agglomerative Hierarchical (AH), BIRCH, Chameleon) for selecting optimal subsets of information retrieval systems.
• Conducted experiments on two TREC health-related datasets.
• Applied selected subsets in data fusion to boost retrieval quality and credibility.

Main Results:

• Agglomerative Hierarchical (AH) and BIRCH clustering outperformed other methods in identifying effective information retrieval (IR) subsets.
• AH-based fusion of up to 20 systems yielded a 60% gain in Mean Average Precision (MAP).
• AH-based fusion showed over a 30% increase in NDCG_UCC (a credibility-focused metric) compared to the best single system.

Conclusions:

• Clustering-based fusion strategies significantly enhance the retrieval of trustworthy health content.
• These methods are effective in combating health misinformation by improving access to reliable information.
• Incorporating advanced data fusion techniques is recommended for health information retrieval systems.