| An Ensemble Deep Model for Deceptive Opinion Detection Based on Opinion Text For English and Persian Languages |
| کد مقاله : 1112-CYSP2024 (R1) |
| نویسندگان |
|
Mahmoud Ali-Arab1، Kazim Fouladi-Ghaleh *2 1دانش‌آموخته کارشناسی ارشد مهندسی فناوری اطلاعات، دانشکده مهندسی، دانشکدگان فارابی، دانشگاه تهران 2عضو هیئت علمی گروه مهندسی کامپیوتر دانشکده مهندسی دانشکدگان فارابی دانشگاه تهران |
| چکیده مقاله |
| Spam reviews, written primarily to promote or demote a product or brand, mislead people for making purchases and make decisions difficult for customers. Much research has been done to detect spam reviews, and different methods have been developed, but these methods often use metadata to detect spam review, and because of the use of metadata, singleton reviews (reviews whose author has submitted only one comment) are dropped from the dataset because these types of reviews do not give much information to the model. In addition, in existing methods, comment text is considered any other text in text classification issues, while comment text contains many features that can be extracted and used to detect spam reviews. In this research, a hybrid model using 4 BiLSTM networks is presented, trained on the comment's text and the comments' polarity. Due to the lack of polarity of opinions in different datasets, a sentiment analysis model has been used that extracts the polarity of opinions from the comments text and adds it to the dataset. Since the model depends only on the comment's text and does not use metadata, there will be no problem in detecting singleton spam reviews using this model. The proposed model is evaluated for English and Persian languages. The performance of the proposed model is comparable for both Persian and English. For English, the accuracy was 89.4% on the OpSpam dataset and 87.7% on the HDR dataset. Also, 87.7% accuracy was obtained for the Persian language on the DigiKala dataset. |
| کلیدواژه ها |
| Deceptive Opinion Detection, Deep Learning, Ensemble Learning, Persian, English |
| وضعیت: پذیرفته شده برای ارائه شفاهی |