Al application in Detecting inappropriate and misleading content on E-commerce Platforms

Abstract

  Objective: This study aimed to design and develop an advanced system for identifying and classifying fake Arabic reviews of e-commerce applications using machine learning techniques.   

Methodology: The current study adopted a theoretical approach by conducting a comprehensive analytical review of relevant previous studies to identify the most relevant studies that addressed the topic of the study. The proposed system evaluates the components sequentially through data processing, feature engineering, and model training. With the help of the natural language processing techniques, the Arabic text is optimized through text segmentation, common word removal, and rooting.  

Results: Feature extraction and model training were achieved using TF-IDF, which records word frequency, and Word2Vec, which is used for concept mapping. Supervised machine learning algorithms, such as LR, Naive Bayes, and Linear SVM, were among the algorithms developed and evaluated using a range of metrics: accuracy, fine-tuning, recall, and F1. The trained model's tools were integrated into a live streaming application using Streamlit. Application users could upload and compare a collection of documents, and the analysis of individual reviews would help assess the model's performance.  

 Conclusion: the primary impact of the presented machine learning model was to increase customer confidence in online shopping through the automation of review analysis. This model also served as a springboard for machine learning and artificial intelligence applications in e-commerce.