Political Ideology Prediction from Bengali Text Using Word Embedding Models

摘要

The universal espousal of social media for political communication makes easy and extraordinary chances to keep an eye on the opinions of a large number of politically active individuals in real-time by providing an overall idea on the ideologies of those individuals regarding governmental issues. Nowadays, websites and android applications like Facebook, YouTube, and Instagram are the most embraced and popular means of information, communication, and entertainment to the people of Bangladesh. Hence, these social sites are a great source for collecting data related to the political views of the users from the perspective of this country. In this study, we have trained the data by an unsupervised machine learning and deep neural network model named word2vec to predict ideology from Bengali text. We have experimented with the word embedding model by utilizing CBOW and Skip-gram algorithms. The results of both of the algorithms were analyzed and compared with each other and as well as with the previous works related to this. Between them, CBOW provides higher accuracy which is 76.22% than the Skip-gram model in predicting political ideology.