The Impact of applying Different Preprocessing Steps on Review Spam Detection

Online reviews become a valuable source of information that indicate the overall opinion about products and services, which affect customer's decision to purchase a product or service. Since not all online reviews and comments are truthful, it is important to detect fake and poison reviews. Many machine learning techniques could be applied to detect spam reviews by extracting a useful features from review's text using Natural Language Processing (NLP). Many types of features could be used in this manor such as linguistic features, Word Count, n-gram feature sets and number of pronouns. In order to extract such features, many types of preprocessing steps could be performed before applying the classification method, this steps may include POS tagging, n-gram term frequencies, stemming, stop word and punctuation marks filtering, etc. this preprocessing steps may affect the overall accuracy of the review spam detection task. In this research, we will investigate the effects of preprocessing steps on the accuracy of reviews spam detection. Different machine learning algorithms will be applied such as Support Victor Machine (SVM) and Naïve Bayes (NB), and a labeled dataset of Hotels reviews will be analyze and process. The efficiency will be evaluated according to many evaluation measures such as: precision, recall and accuracy. Peer-review under responsibility of the Conference Program Chairs.