Review Spam Detection Based On Trustworthiness Transmission

With the development of the Internet, it has dramatically changed the way people express themselves online and interact with others. Online reviews play a crucial role in today's electronic commerce. Customers often to check reviews of products or stores before making the decision of what or where to buy. However, due to the pervasive spam reviews, customers can be misled to buy low quality products, while decent stores can be defamed by malicious reviews. Hence, in recent years, the intelligent detection of review spam has become a popular topic for the researchers of artificial intelligence.The thesis discusses the current status of the review spam detection and analyses the relative algorithms and techniques. Contrapose the problems such as traditional evaluation system based on artificial labels which need heavy workload and would be bad for the computer processing. The algorithm performance is measured by two index in the thesis. The method which compares the recommendation accuracy and the positive feedback ratio basing on original data sample and experiment results provide new perspective for the effect of detecting algorithm. Therefore, it can be regard as a complement of traditional evaluation system.This thesis uses trustworthiness score to measure reviews, reviewers and products' credible degree. By analyzing the key factors affect the trustworthiness of reviews, and extracting the length of the review text, attributes coverage proportion and time distribution features the scores of the initial confidence can be computed. In addition, we propose combine the statistical method of the word frequency with the topic model in extracting the attribute dictionary, and uses third-party tools which named word2 vec to construct extraction model.Inspired by review graph and fact discovery on the web, the interactional relationships among the trustworthiness of reviews, the trustworthiness of reviewers and the trustworthiness of products were discovered. However, the previous researchers only concentrate on one, and so they ignore the relationships among them. The thesis come up with an algorithm of review spam detection based on trustworthiness transmission which uses the graph model abstracted away from reviews, reviewers, and products, the initial review trustworthiness and the network making up from their relationship to constitute the model computing their trustworthiness score. Then it makes the review trustworthiness score more accurate and remove some reviews which are less than trustworthiness level threshold. Experiment shows that this algorithm has improved in the precision rate and recall rate.