التنقيب عن الاراء العربية باستخدام شجرة القرار المتوازية

With the popularity of online shopping it is increasingly becoming important for manufacturers and service providers to ask customers to review their product and associated service. Similarly, the number of customer reviews that a product receives grows rapidly and can be in hundreds or even thousands. This makes it difficult for a potential customer to decide whether or not to buy the product. It is also difficult for the manufacturer of the product to keep track and manage customer opinions. Hence the importance stemmed opinion mining which is an emerging area of research, that summarizes the customer reviews of a product or service and express whether the opinions are positive or negative. Various methods have been proposed as classifiers for opinion mining such as Naïve Bayesian, k-Nearest Neighbor techniques, and Support vector machine, the main drawback of these methods is classifying opinion without giving us the reasons about why the instance opinion is classified to certain class. Therefore, in our work, we investigate opinion mining of Arabic text at the document level, by applying decision trees classification method to have clear, understandable rules. In addition, we apply parallel decision trees classifiers to have efficient results. We applied parallel decision trees on two Arabic corpus BHA and OCA of text. To generate text representations, we apply some preprocessing operators such as Tokenize , filters Arabic stopwords, Stem Arabic, filters tokens based on their length, and filters tokens based on their content to exclude English words. In case of applying parallel decision tree family on OCA, we get the best results of accuracy (93.83%) , f-measure (93.22) and consumed time 42 Sec at thread 4, which is greater than sequential that have accuracy (92.59%) and f-measure (92.58), and consumed time 68 Sec. In case of applying parallel decision tree family on BHA we get the best results of accuracy (90.63%) , f-measure (82.29)and consumed time 219 Sec at thread 4, these results are different from sequential that have accuracy (90.70%) and f-measure (90.94), and consumed time 417 Sec