عرفان بختياري رمضاني

Sentiment analysis is a subtask of natural language processing that can be used to identify and classify texts based on the feelings of the author of the text. Therefore, various techniques for sentiment analysis in the text have been developed to decide whether user comments or tweets are positive or negative. Among the various deep neural architectures used for emotion analysis, the combination of recurrent neural networks, convolutional neural networks, and attention mechanisms has attracted increasing attention. These models can process texts of arbitrary length and pay attention to important features. Convolutional neural networks reduced feature dimensions and improved the performance of models by extracting local features. The use of recurrent neural networks for feature extraction at the beginning of the model cannot extract local features well; thus, the final performance is weakened. To address these problems, An Attention-Based Bidirectional-LSTM-GRU and Temporal Inception-module Convolution Deep Model (ABTICDM) is proposed. Using two layers of GRU and two-way independent LSTM and standard and temporal convolutional neural networks, ABTICDM extracts both past and future contexts by considering information flow in both directions. Also, the attention mechanism is applied to the output of two-way layers and Temporal Inception-module Convolution (TIC) to identify important features. ABTICDM uses TIC, convolutional neural networks, and a combination of global integrations to extract local features better and reduce feature dimensions. The effectiveness of the proposed new ABTICDM framework is eva‎luated for sentiment polarity detection, which is the most common sentiment analysis task. Experiments were performed on five review datasets and three Tweet datasets. The comparison of ABTICDM with eight recent deep neural networks for sentiment analysis shows that ABTICDM can obtain advanced results in long comments and has also obtained advanced results in short tweet polarity classification.

پردازش زبان طبيعي , يادگيري عميق , تحليل احساسات , Bi-LSTM , Bi-GRU

Natural Language Processing , Deep Learning , Sentiment Analysis , Bi-LSTM , Bi-GRU

Erfan Bakhtiari Ramezani

Mohammad Reza Kangavari