Classification Bullying Tweet Using Convolutional Neural Network with Word2vec
Ricko, Priyo Sidik Sasongko · 2021
Twitter was a social media for interacting with friends, conveying thoughts, feelings, ideas, or just a medium for seeking entertainment or information. Twitter gave users the freedom to wrote anything there. That freedom may cause a lot of cyberbullying activities on twitter. One of the effects commonly felt by victims of bullying on social media was a feeling of insecurity that could even lead to depress. This studied classifies bullying tweets used the convolutional neural network method with the Word2vec model on Indonesian-language tweets. Word2vec pre-training provides better accuracy because word2vec considers the position and context of words in a sentence. The Convolutional Neural Network method was one of the deep learned algorithms that could provide more accurate and faster results by analyzing even large amounts of data. The data testing process had an accuracy valued of 93,83%. The Convolutional Neural Network model used in data testing was done by combining the max pooling results from bigram, trigram, and fourgram, because it could got smaller and more complex features also the amount of information extracted became more by used a learned rate of 0,001 and a dropout of 0,4.