ITE Law Enforcement Support through Detection Tools of Fake News, Hate Speech, and Insults in Digital Media

In the era of technological advancement, the relatively low literacy rate of the people in the digital world will result in many behaviors that have negative impacts, such as quickly spreading fake news and unhealthy communication. The government has taken steps to regulate digital information transactions through the ITE Law to overcome this. However, people's understanding of this matter still needs improvement. An automatic detection device for inappropriate speech in the digital media needs to be built to overcome this. An utterance entering this detector will be identified into one of 4 categories: normal speech, fake news, hate speech, and offensive. This detector is built by training a neural network model on a compilation of various digital speech data sources. We select several architectural models by comparing the results obtained from the same data. We also build a classification model with three categories considering that hateful and offensive utterances are often quite similar and difficult to distinguish. The model obtains good final results, with the best performance obtained by the LSTM model with three categories.