Sentiment Analysis Model Using Word2vec, Bi-LSTM and Attention Mechanism

Within the interdisciplinary field of data science, this paper proposes a sentiment classification model applied to text, specifically tweets, using neural networks. To do that, after gathering and pre-processing the data, firstly, word2vec model is used to convert tweets into word vectors. Secondly, Bi-LSTM neural network is used to capture the semantic meaning of the text. Thirdly, attention mechanism is added to extract the most relevant words. Fourthly, a sigmoid function is used to achieve the emotion classification. Thus, after crossing each layer of the neural network, the model finally returns the result of the binary classification to detect the emotion of the social media text data. The presented model gives an accuracy of 84.12%, surpassing other existing models for this purpose.