An Interpretable Classification Framework for Information Extraction from Online Healthcare Forums.

Area of Science:

• Computational linguistics
• Medical informatics
• Natural Language Processing

Background:

• Online healthcare forums (OHFs) are vital for patients sharing experiences, offering insights into diseases and patient situations.
• Classifying sentences in OHFs aids understanding but unstructured text poses challenges for existing algorithms.
• Interpretability of complex models like deep neural networks is crucial for healthcare applications.

Purpose of the Study:

• To develop an effective and interpretable framework for classifying sentences within online healthcare forum posts.
• To categorize sentences into medication, symptom, and background classes for improved meaning extraction.
• To provide explicit decision rules for understanding classification outcomes in healthcare contexts.

Main Methods:

• Sentences are projected into an interpretable feature space using labeled sequential patterns, UMLS semantic types, and heuristic features.
• A forest-based model is employed for the categorization of online healthcare forum posts.
• An interpretation method is developed to extract explicit decision rules for gaining insights from text.

Main Results:

• The proposed framework effectively classifies online healthcare forum posts.
• The interpretable nature of the model allows for explicit extraction of decision rules.
• Experimental results on real-world data validate the computational framework's effectiveness.

Conclusions:

• The developed framework offers an effective and interpretable solution for classifying online healthcare forum posts.
• This approach enhances the understanding of patient-generated health information.
• The interpretability of the model is critical for its application and trust in healthcare settings.