Emotion Detection in Suicide Notes using Maximum Entropy Classification.

Area of Science:

• Computational linguistics
• Natural Language Processing (NLP)
• Psychological analysis

Background:

• Sentiment analysis in clinical texts is challenging.
• Identifying emotions in suicide notes requires nuanced NLP techniques.
• Previous research has explored automated text analysis for mental health.

Purpose of the Study:

• To develop and evaluate an ensemble of supervised maximum entropy classifiers for detecting sentiments in suicide notes.
• To assess the impact of lexical and syntactic features on emotion classification accuracy.
• To participate in the 2011 i2b2 NLP Shared Task, Track 2.

Main Methods:

• Extracted lexical and syntactic features from a training set of annotated suicide notes.
• Trained separate supervised maximum entropy classifiers for fifteen pre-specified emotions.
• Utilized an ensemble approach for improved classification performance.
• Evaluated classifier performance on unseen test data.

Main Results:

• The ensemble classifiers demonstrated accurate detection and identification of sentiments.
• Classifier precision and recall correlated positively with the frequency of emotions in the training data.
• The best system achieved an F(1) score of 0.534 on the test dataset.

Conclusions:

• Supervised maximum entropy classifiers are effective for sentiment analysis in suicide notes.
• Feature selection and training data size significantly influence classifier performance.
• This approach contributes to automated methods for analyzing sensitive textual data in mental health research.