The General Services Administration (GSA) gathers public input on prospective federal regulations. This feedback plays a crucial role in determining if the regulations proceed to approval or require additional deliberation. The process of reviewing comments is carried out manually, with an average of 3,000 comments received for each proposed regulation. These comments are carefully examined to identify any fraudulent submissions, which could potentially cause delays in the decision-making process for a given regulation.
REI has investigated deep learning techniques and a graph database approach to detect fraudulent comments and deliver advanced analysis for each proposed regulation’s feedback. This sophisticated analysis consists of three predictive components:
- Comment Sentiment Identification: Evaluating the ratio of negative, positive, and neutral sentiments.
- Fake Content Detection: Identifying bot-generated comments.
- Similarity Assessment: Recognizing the most similar and identical comments.
After undergoing a comprehensive analysis process that integrates all three predictive models, each comment is assigned a score determined by its sentiment, the likelihood of being fake, and its similarity to other comments. This refined model, which has been rigorously tested on thousands of new texts, boasts an impressive accuracy rate of approximately 93%, demonstrating its effectiveness in detecting fraudulent comments and providing valuable insights into public feedback on proposed regulations.
The implementation of automated advanced analysis for comment evaluation can significantly decrease manual labor and offer in-depth insights into the thousands of comments received. By effectively identifying fraudulent submissions, GSA is able to further improve the transparency of the bipartisan regulation process, ensuring that genuine public feedback plays a pivotal role in shaping federal regulations. This streamlined approach not only saves time and resources, but also bolsters the integrity and effectiveness of the decision-making process.
- Machine Learning
- Graph Database Application
- Fraud Detection
- Sentiment Analysis
- Natural Language Processing