I wanted to pick a bakery based on reviews, but ended up writing an anomaly analyzer with embeddings and graphs

The author of the article encountered difficulties finding reliable reviews on Yandex Maps, which led to the creation of ReviewScope. This tool is designed to automatically detect anomalies, repetitive texts, and suspicious patterns in user reviews. The solution is built upon vector embeddings, matching graphs, and weighted rating algorithms. The publication details the development process, including technical challenges, errors encountered, and the results of testing the system on a dataset of 21,000 real reviews. The author shares their experience in applying data analysis methods to filter content and improve the credibility of user ratings, turning a mundane task into a comprehensive engineering project.