Roman Urdu Fake Reviews Detection Using Stacked LSTM Architecture

作     者：Hayat, Umer Saeed, Ali Vardag, Muhammad Humayon Khan Ullah, Muhammad Farhat Iqbal, Nadeem 

作者机构：Department of Software Engineering University of Lahore Lahore Pakistan Department of Software Engineering University of Central Punjab Lahore Pakistan Department of Computer Science and IT University of Lahore Lahore Pakistan 

出 版 物：《SN Computer Science》 (SN COMPUT. SCI.)

年 卷 期：2022年第3卷第6期

页      面：470页

主　　题：Deep learning approaches Roman Urdu fake reviews detection Supervised learning approaches Word embeddings 

摘      要：Fake reviews detection is a considerable challenge to the different e-commerce and online business settings. This task aims to develop such systems that could ensure the veracity of reviews. The research community has made a range of attempts to deal with this issue. But unluckily, these attempts were geared to only small set of languages like English, Arabic, and some others. In the subcontinent, Roman Urdu is being used on the web. It has not been explored thoroughly for this task, however. On the other hand, over the last few years, deep learning methods have proved very successful for the diverse Natural Language Processing tasks. But, deep learning methods have not been explored for the Roman Urdu fake review detection task. To address this gap, this study has rendered a two-fold contribution (1) Construction of a novel Roman Urdu Fake Reviews Detection Corpus (RU-FRDC) which composes 5150 annotated reviews and (2) Comparison of various deep learning architectures including Simple RNN, LSTM, GRU, Bi-LSTM, and Bi-GRU. The evaluation has been carried out using widely used evaluation measures, i.e., Precision, Recall, F1, and ACC-ROC. The highest results were achieved using the stacked LSTM model (ACC - ROC = 0.943 and F1= 0.88). © 2022, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.