|
Sentiment Analysis, Drug Industry, Hybrid Deep Learning, Convolutional And Recursive Architecture.
|
Shayan Tabatabaei   , Mehdi Barati , Maryam Bayati |
|
|
|
Abstract: (835 Views) |
Social media are very popular environments for their users to exchange opinions. The sharing of users' opinions in these environments along with the daily increase of users who join these media has created a huge source of unstructured data that contains a huge amount of hidden knowledge. The practical results obtained from the analysis and investigation of the hidden emotions in the unstructured texts of the users have caused financial and commercial organizations and governments and various organizations to pay more attention to this field. This issue and its importance led to the introduction of the field of sentiment analysis based on machine learning techniques. Traditional machine learning methods, although they have provided relatively good results in this field, but compared to deep learning and deep neural networks, the results are not very promising. Recently, most studies are based on deep neural networks individually or combined. In combined methods, the way of combining two types of deep architecture has a great impact on their results. In many methods, two deep networks are combined separately, and these models could not provide completely successful results. In this research, the combination of two deep convolutional neural networks and GRU recurrent neural network in a single architecture has been introduced in line with sentiment analysis in the field of pharmaceuticals and regarding drugs.The appropriate combination of the layers of the above two networks in a deep architecture as well as the use of the word embedding technique and the FastText pre-trained word embedding network have made the proposed architecture achieve more successful results compared to other combined architectures. The experimental results of the proposed hybrid deep architecture on Drug reviews data show that the proposed model has achieved 90/70% accuracy, which has improved the accuracy of sentiment analysis by about 2/2% compared to previous methods. The above results indicate that the proposed hybrid deep model, which combines two types of deep networks in one network, as well as the use of pre-trained word embedding network, has a successful performance in sentiment analysis in the pharmaceutical industry.
|
Article number: 3 |
| Keywords: Sentiment Analysis, Drug Industry, Hybrid Deep Learning, Convolutional And Recursive Architecture. |
|
|
Full-Text [PDF 1881 kb]
(303 Downloads)
|
Type of Study: Research |
Subject:
Special
Received: 2022/11/6 | Accepted: 2022/12/15 | Published: 2023/01/1
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
