Mining Twitter Data to Improve Detection of Schizophrenia

Area of Science:

• Computational psychiatry
• Natural Language Processing
• Machine Learning in Healthcare

Background:

• Schizophrenia affects 1% of the US population, with individuals being 4x more likely to die by suicide.
• Identifying individuals with schizophrenia, especially those not seeking treatment, remains a significant challenge.
• Microblogging platforms offer a unique avenue for observing user-expressed thoughts and emotions.

Purpose of the Study:

• To leverage Twitter data and machine learning to detect individuals with schizophrenia.
• To develop a tool for identifying at-risk individuals who may not be seeking treatment.
• To facilitate early intervention and improve diagnostic processes for schizophrenia.

Main Methods:

• Utilized a large corpus of Twitter posts for analysis.
• Engineered features from tweets, including emoticon usage, posting time, and specific dictionary terms.
• Trained, built, and validated multiple machine learning models, including a Support Vector Machine (SVM).

Main Results:

• The Support Vector Machine model demonstrated superior performance.
• Achieved 92% precision and 71% recall on the test dataset.
• Developed a web application for dynamic display of cohort summary statistics.

Conclusions:

• Machine learning analysis of Twitter data can effectively aid in detecting individuals with schizophrenia.
• The developed tool can support outreach to undiagnosed individuals and enhance physician diagnoses.
• This technology has the potential to destigmatize schizophrenia and improve patient outcomes.