Method and system to filter out harassment from incoming social media data
View Patent ↗Social media users are subject to harassment when unwanted offending content reaches them. Social media companies are reluctant to police content. The Emakia system provides a solution at the point where incoming real-time data are received. The Emakia system proposes to use sets of Machine Learning classifiers to filter text, images, audio, and video. Classifiers are trained with labeled data. After training, the model is used to screen the incoming real-time data. On the user mobile device, only approved content is displayed. The unwanted data are still available if the user desires to access them. The system provides multiple classifiers and customized models to the individual user. When harassment content is detected a report is sent to an entity who can help support the receiver.
1. A method for filtering incoming social media data for a user on a mobile device, comprising:
transferring data from at least one social media platform to at least one user device, and at least one Emakia server;
collecting social media data with programs using search API calls;
creating a list of harassing terms into three sets: a first set with hardcore terms, a second set with words evincing a milder harassing tone, a third set with terms that have double meaning with one of the meanings being harassing;
defining harassing content, on at least one Emakia server, as having at least one word from the set with hardcore terms, or at least one word from the second set and at least one word from the third set;
labeling social media data content in any language including emojis as neutral or harassing;
training and evaluating classifier models with the labeled data in any language and any type of data;
running Enaëlle, an application, on the user mobile device;
detecting the language of the data content;
uploading the classifier model associated with detected language onto the user device;
filtering, on at least one Emakia server, incoming social media content with the classifier model, increasing a size of a labeled data set by collecting more labeled data from different sources;
increasing detection accuracy, on at least one Emakia server, and retraining the classifier model with an adaptive filter by catching content unknown to the classifier model with the three harassing sets;
defining harassing content allows the Emakia system to put in place a validation system that, validates the training set labeling during the classifiers' training and validates the model accuracy with the adaptive filter;
separating neutral content from unwanted content, unwanted content being harassing content or fake news;
displaying the neutral content in one tab of the mobile device in Enaëlle while harassing content is still accessible with a second tab of the mobile device in Enaëlle;
customizing the classifier model that has labeled harassing content and neutral content, by the user sliding the harassing content and the neutral content on a screen of the mobile device in Enaëlle and moving the harassing content and the neutral content to a different category, said customizing sent to the at least one Emakia server;
reporting unwanted content and recommending action to address conduct presented in a report, wherein the user can elect for professional intervention.