Detecting hate speech from tweets for sentiment analysis

摘要

In the era of the popularity of Social Networking Service (SNS), people became increasingly inseparable from mobile phones and computers. People want to get information and real-time updates from social media, and they want to know how many Internet citizens have comments and opinions on many dynamic news. The interaction among users on social networking platforms is usually positive, advisory and motivating and influential. However, sometimes people will also reveal objectionable content, such as hate speech, abusive and bullying or discriminatory words. According to multiple methods, we will find out which method has the best accuracy of detecting hate speech from tweets. Many papers using types of data for experimentation. The major innovation of this article is that we used different ratios of data to compare with multiple methods at the same time. As a result, good performance is obtained by using machine learning when data is small. The good results can be obtained by using deep learning when we use more data for our experiments. Using BiRNN can get the best results, compared with other methods we used. Even if this method is superior to other models, we have to consider the type of data set in the future.