With the growth in the usage of social media, it has become increasingly common for people to hide behind a mask and abuse others. We have attempted to detect such tweets and comments that are malicious in intent, which either targets an individual or a group. Our best classifier for identifying offensive tweets for SubTask A (Classifying offensive vs. nonoffensive) has an accuracy of 83.14% and a f1score of 0.7565 on the actual test data. For SubTask B, to identify if an offensive tweet is targeted (If targeted towards an individual or a group), the classifier performs with an accuracy of 89.17% and f1-score of 0.5885. The paper talks about how we generated linguistic and semantic features to build an ensemble machine learning model. By training with more extracts from different sources (Facebook, and more tweets), the paper shows how the accuracy changes with additional training data.

No abstract

As for other forms of AI, speech recognition has recently been examined with respect to performance disparities across different user cohorts. One approach to achieve fairness in speech recognition is to (1) identify speaker cohorts that suffer from subpar performance and (2) apply fairness mitigation measures targeting the cohorts discovered. In this paper, we report on initial findings with both discovery and mitigation of performance disparities using data from a product-scale AI assistant speech recognition system. We compare cohort discovery based on geographic and demographic information to a more scalable method that groups speakers without human labels, using speaker embedding technology. For fairness mitigation, we find that oversampling of underrepresented cohorts, as well as modeling speaker cohort membership by additional input variables, reduces the gap between top-and bottom-performing cohorts, without deteriorating overall recognition accuracy.

No abstract