Deep Learning Model for Classifying Spam Review Over Social Media

The user evaluation of service-oriented websites plays a crucial role in consumers’ decision-making process. However, the prevalence of fake reviews, driven by commercial interests, undermines the reliability of these evaluations. Therefore, detecting and managing fake reviews is essential for ensuring the proper functioning of websites and maintaining a clean online environment. This research study proposes a novel approach for detecting fake reviews by leveraging a Graph Convolutional Network (GCN) that integrates semantic similarity. The method incorporates word-level and document-level information to construct a graph neural network for text classification. Edges between words are established based on the Pointwise Mutual Information (PMI) index and semantic similarity measured through word embeddings. Additionally, edges between words and comments are created using TF-IDF feature values. The transfer feature of the graph neural network is then utilized to aggregate and extract crucial features from the constructed graph, enabling the classification of fake reviews. Experimental results on public datasets demonstrate the effectiveness of the proposed method. Compared to traditional approaches such as CNN, LSTM, and Text-GCN, the proposed method achieves substantial improvements in accuracy, with respective increases of 7%, 4.8%, and 1.3%. These findings highlight the significance of the proposed approach in enhancing the detection of fake reviews and improving the overall reliability of user evaluations on service-oriented websites.