Topic categorisation of statements in suicide notes with integrated rules and machine learning.

Area of Science:

• Computational linguistics
• Psychiatry
• Natural Language Processing

Background:

• Suicide notes contain vital information for mental health research.
• Automated analysis of suicide notes can aid in understanding risk factors and emotional states.

Purpose of the Study:

• To develop and evaluate an automated system for categorizing statements within suicide notes into 15 predefined topics.
• To assess the effectiveness of a hybrid approach combining rule-based and machine learning methods.

Main Methods:

• A hybrid approach integrating lexico-syntactic rules with machine learning models.
• Machine learning models utilized features such as named entities, lexical, lexico-semantic, and presentation characteristics.
• Evaluation performed on a dataset of 300 suicide notes from the i2b2 2011 challenge.

Main Results:

• The automated approach achieved an overall best micro F-measure of 53.36%.
• Rule-based methods yielded the highest precision (67.17%), while integrated methods offered the best recall (50.57%).
• Higher performance was observed for frequent topics like 'Love' and well-defined categories such as 'Thankfulness' (precision 68%-79%).

Conclusions:

• Automated text mining is a viable method for topic categorization in suicide notes.
• The hybrid approach shows promise, though certain topics remain challenging for current models.