A naïve bayes approach to classifying topics in suicide notes

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Machine Learning

Background:

• Automatic classification of emotions in suicide notes is crucial for mental health research.
• The 2011 i2b2 Challenge focused on sentiment classification within suicide notes.

Purpose of the Study:

• To develop and evaluate a system for automatically classifying sentences in suicide notes into 15 emotion-related topics.
• To compare the system's performance against other participants in the i2b2 Challenge.

Main Methods:

• A hybrid approach combining a Naïve Bayes classifier with rule-based pattern matching.
• Feature extraction based on lexico-semantic properties of words and regular expressions for topic patterns.
• Training data comprised 600 manually annotated suicide notes.

Main Results:

• The system achieved a micro-averaged F-measure of 53% (55% precision, 52% recall).
• This performance was significantly higher than the average F-measure of 48.75% achieved by 26 competing systems.
• Evaluation used a gold standard of 300 manually annotated suicide notes.

Conclusions:

• The proposed system demonstrates effective automatic sentiment classification for suicide notes.
• The combination of machine learning and rule-based methods is a viable strategy for this task.
• The system's performance indicates its potential utility in analyzing sensitive textual data.