Link: https://gitlab.com/alexander_kinsora/misinfofinder
Paper link: Creating a Labeled Dataset for Medical Misinformation in Health Forums
Task:
- binary classification - post is misinformative (1) or not (0), according to label attribute
This project will provide storage and source control for code related to the development of MisInfoFinder - an algorithm for the identification of health-related misinformation in textual data sources at scale.
Note: The dataset after feature extraction was chosen for analysis. However, in repository (link) there is also dataset with raw data and even one bigger dataset in different branch (called nlpFeatures).
Note2: The samples of the dataset are not news articles, but comment posts in threads with question related to medicine.
- id - id of sample
- label - whether post was misinformative (1) or not (0)
- last_misinfo_dist_score - score reflecting the distance of the comment from the most recent comment judged misinformative in the thread
- running_misinfo_count - running count of the number of posts in the thread that had been classified as misinformative
- date - date
- comment_post_id - comment post id
- equivocation_terms - count of terms indicating equivocation ("could have", "thought", etc.)
- comment_length - length of the comment
- comment_text - text of the comment
- all_caps_word_count - count of fully capitalized words
- question_post_id - id of question post
- simple_capitalization_percent - percentage of capitalized characters in each post
- hearsay_terms - count of hearsay terms ("I heard", "they said", etc.)