Abstract:

 The task that aims to detect spam reviews for the automobile domain was divided into four sub-tasks: supporting review detection, irrelevant review detection, advertisement detection and fake review detection. Both rule-based methods and machine learning methods were used to identify spam reviews. Many aspects were considered in the rule-based method, such as automobile domain knowledge, words with polarity, and information of the author. The review content feature and author information were combined to train a model with a maxent classifier. Experimental results showed that machine learning method performs well for the domain whose property was obvious, with numerical feedback information and labeled training data.

Key words:  spam detection; advertisement detection; rule-based method; machine learnin