Sentiment Analysis for Online Communication Platforms Using Bi -LSTM with Self Gated Attention

Le résumé fourni par la source

Background: emotional classification has become a significant tool in analysing the text and people's emotion in crisis or events, especially on social media platforms. Challenges: however, for handling long range dependency in short data that contains the huge amount of noisy content and correlation between the words is more difficult. Therefore, it is difficult to classify the text in the long text. Proposed Methodology: to overcome these challenges the proposed Bidirectional Long Short-Term Memory (Bi-LSTM) with Self Gated Attention (SGA) used for online communication platforms to classifies the sentiment analysis. Initially data was collected in sentiment 140, which is collected from twitter API. In the data preprocessing tokenization is used to transform the text into words using NLPK library. In particular stop word removal is used to remove the unnecessary words and also frequently used word in the text. Therefore, kullback–Leibler divergence used for the stop words, which chose the words randomly in the document and give rank to specific term. Subsequently, stemming is used to remove the prefix and suffix of the word to keep the static meaning and avoid confusion. Further, in feature extraction word2vec is used to transform the text into words using neural network. Furthermore, Bi-LSTM uses the three gates such as input gate, forget gate and output gate with self-gated to learn relationship between the text. Therefore, output gate generates the output using SoftMax classifier of sentiment analysis such as positive, negative or neutral comments. The proposed model outperformed the existing model such as HGAR includes accuracy (98.6%), precision (97.9%), recall (98.3%) and F1-score (98.1%) results are reported in test set.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.