Ohe-Rwnet: A Novel Neural Network-Based Vectorization Method for Fake Review Detection

Abstract

Advancements in internet technologies have enabled many activities to be carried out in online environments. In recent years, with the increase in online shopping, online reviews have played a significant role in influencing users' product preferences. Although such reviews sometimes provide accurate and objective information, in some cases, they can be misleading. So, this has made fake review detection an important research area today. Although various studies have been conducted in the literature on this subject, the growing prevalence of fake reviews highlights the need for more effective and innovative approaches. In this study, a novel vectorization method called OHE-RWNet is proposed for fake review detection. The proposed approach models semantic relationships between words using neural networks and converts words into machine-processable numerical representations by considering the target concept of spam in context. The OHE-RWNet approach is compared with widely used vectorization techniques, including One-Hot Encoding (OHE), Bag-of-Words (BoW), Term Frequency-Inverse Document Frequency (TF-IDF), Word2Vec, Global Vectors for Word Representation (GloVe), and Bidirectional Encoder Representations from Transformers (BERT) on seven different datasets. Experimental results demonstrate that the proposed method exhibits superior and competitive performance compared to the techniques considered.