Understanding user intents in online health forums

Area of Science:

• Health Informatics
• Natural Language Processing
• Computational Linguistics

Background:

• Online health forums are valuable resources for patients seeking medical information and peer support.
• The unstructured nature of forum content hinders efficient information retrieval and user engagement.
• Identifying user intents is crucial for personalizing information delivery and improving forum usability.

Purpose of the Study:

• To develop a taxonomy of user intents specific to online health forums.
• To propose novel pattern-based features for intent classification.
• To evaluate the effectiveness of a multiclass support vector machine (SVM) classifier for identifying user intents.

Main Methods:

• A taxonomy of user intents was derived from online health forum data.
• Novel pattern-based features were engineered for intent classification.
• A multiclass support vector machine (SVM) classifier was trained and tested on a manually labeled dataset of 1192 HealthBoards posts.
• A hierarchical SVM classifier incorporating both pattern and word features was also evaluated.

Main Results:

• The SVM classifier utilizing pattern-based features achieved a maximum precision of 75% in identifying user intents.
• The hierarchical SVM classifier combining pattern and word features outperformed an SVM using only word features.
• Classification performance remained comparable when training and testing on posts from different forum topics.

Conclusions:

• Pattern-based features are highly effective for identifying user intents in online health forum posts.
• Hierarchical classification models integrating diverse features enhance intent recognition accuracy.
• The developed approach shows promise for improving information retrieval and user experience in online health communities.