An aspect-level sentiment analysis dataset for therapies on Twitter

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Social Media Analysis

Background:

• Sentiment analysis datasets are crucial for developing AI systems that understand user opinions.
• Existing datasets often lack granularity, particularly for specific domains like healthcare therapies.
• Understanding public sentiment towards therapies is vital for public health and treatment development.

Purpose of the Study:

• To create a novel aspect-level sentiment analysis dataset focused on various therapies (medication, behavioral, etc.).
• To enable the training of machine learning models for accurate sentiment detection regarding specific treatments.
• To provide a valuable resource for researchers and developers in NLP and AI.

Main Methods:

• Collected Twitter posts using therapy-related keywords.
• Manually annotated a subset of posts for sentiment (positive, negative, neutral) towards specific therapies.
• Developed annotation guidelines and calculated inter-annotator agreement using Cohen's Kappa score.

Main Results:

• The dataset comprises 5364 posts covering 32 distinct therapies.
• Annotated sentiments include 18.6% positive, 11.5% negative, and 69.9% neutral.
• Achieved a high inter-annotation agreement score of 0.82 using Cohen's Kappa.

Conclusions:

• The developed dataset is the first of its kind, specifically encoding sentiments towards individual therapies.
• It serves as a foundation for building advanced AI systems to analyze public opinion on treatments from social media.
• Facilitates research in NLP, sentiment analysis, and machine learning for healthcare applications.