Deep Learning With Anaphora Resolution for the Detection of Tweeters With Depression: Algorithm Development and...

Area of Science:

• Computational linguistics
• Mental health informatics
• Social media analytics

Background:

• Mental health disorders represent a significant global public health issue.
• Social media platforms offer rich data for psychological cue extraction, but present challenges like distinguishing self-reference from third-party statements.
• Existing natural language processing (NLP) methods often analyze text and user classification separately, hindering integrated sentiment analysis.

Purpose of the Study:

• To develop a predictive model for detecting depression in users based on Twitter posts.
• To identify textual content associated with mental health topics.
• To address anaphoric resolution challenges in social media text analysis.

Main Methods:

• A dataset of 3682 Twitter users (1983 with self-declared depression) was collected.
• Two multiple instance learning models were developed: one incorporating an anaphoric resolution encoder.
• Model performance was evaluated against established machine and deep learning models.

Main Results:

• The anaphoric resolution model achieved maximum accuracy of 92%, F1 score of 92%, and area under the curve of 90%.
• This model demonstrated superior performance compared to alternative predictive models, including classical and deep learning approaches.
• The model successfully highlighted posts relevant to the author's mental health.

Conclusions:

• The developed anaphoric resolution model shows significant promise in detecting depression from social media data.
• It outperforms existing predictive models, offering a more effective approach to mental health analysis on platforms like Twitter.
• The model provides valuable insights into the textual content related to tweeter mental health.