Simple harassment detector bot for Facebook -- Developed at HackUPC Winter 2017
Within the code of conduct of MLH there is a section that is highlighted whenever there is a hackathon: the fight against harassment. Taking into account this problem, a program that could detect this conduct would be a great improve and help for moderators.
Due to these requirements, we have developed harambot, a monitoring application that detects online harassment towards other users. Although this technology can be implemented multiple platforms (twitter, slack ...) we have focused on using facebook.
The main objectives about this project are two:
- Detect harassment in online chats
- Filter and self-learning about potentially harmful messages and achieve more accuracy
The operation of the program is simple: first, all the comments of a facebook group are analyzed by the Go language and saved in a database. Then, through Everis API, the data are treated to classify them into harassment or potentially harassment, and are re-stored in the database. The results obtained are displayed on a graphical interface (GUI) where a moderator will finish manually filtering offensive comments.
Although the results have been quite good, we would like to implement an artificial intelligence to improve the results. We've tried to develop it but it didn't work very well, so we had to readjust the program's functionality. In this terms, the application can only identify partially if the comment is good or bad.
In conclusion, the program could be a great solution if improves. As a team we are very proud of our job because we are conformed by an informatics, telecom and chemist students and not everyone is familiar with the code programmed. It's been a great way to learn and improve our knowledge.