کنفرانس بین المللی فناوری اطلاعات و دانش

English

صفحه اصلی / چهاردهمین کنفرانس بین المللی فناوری اطلاعات و دانش

Sentiment Analysis of the Amazon Customers Using the BiGRU Neural Network Enhanced by Attention Mechanism

نویسندگان :

Sara Sinan Salman al-Abedi¹ Keyvan Mohebbi²

1- دانشگاه آزاد اسلامی واحد اصفهان (خوراسگان) 2- دانشگاه آزاد اسلامی واحد اصفهان (خوراسگان)

کلمات کلیدی :

Sentiment Analysis،Bidirectional Gated Recurrent Unit Neural Network،Deep Learning،Attention Mechanism

چکیده :

In order to effectively analyze the vast amount of information available on the Internet, including opinions, attitudes, and views, it is crucial to predict the emotions and behavior of users. While recurrent neural networks have shown promising results in natural language processing, they encounter challenges such as gradient vanishing in lengthy texts and limited text comprehension. Gated neural networks, on the other hand, excel in understanding long-term dependencies between word sequences. This research introduces a novel approach that leverages a bidirectional gated recurrent unit neural network with an attention mechanism to analyze the sentiments of Amazon users. The model begins by training the initial word input using Word2vec and transferring the appropriate weights for each embedding through transfer learning from GloVe. The model consists of three layers of bidirectional GRU with the output of the third layer feeding into the attention mechanism. This mechanism assigns weights to the words in the sentence, giving importance to crucial words. The final part of the model employs three fully connected layers. Evaluation results on the Amazon review dataset demonstrate that the proposed approach significantly improves precision, recall, F-score, and accuracy metrics, achieving rates of 96.30%, 91.10%, 93.62%, and 95.82% respectively.