Research On The Automatic Generation Of Customer Reviews Report Facing Network

As e-commerce is becoming more and more popular, the number of customer reviews that a product receives grows rapidly. For a popular product, the number of reviews can be in thousands or more. This makes it difficult for a potential customer to read them and make an informed decision on whether to purchase the product or not. It also makes it difficult for the manufacturer of the product to keep track and to manage customer opinions. In this research, we aim to mine all the customer reviews of a product and generate a report about it. This report is different from traditional text summarization. Because we are only concerned with the attributes of the product on which the customers have expressed their opinions.For extracting product attributes and customer reviews mining, this paper builds some resources. These resources include intensifier dictionary, negative words dictionary, first person pronoun dictionary, modal particles dictionary, structure words dictionary, feeling words dictionary and opinion words dictionary. These resources are common resources, rather than resources built for specific areas. When building the opinion words resources, this paper presents a method to build attributes and swing opinion words pairs based on template scoring, which is proved effective to analyze the opinion of the sentences with uncertain opinion words.In recent years, researchers have proposed a number of product-attributes extracted approaches. There are template-based approach and association rules approach, this paper analyzes the shortcomings of these two methods, and proposes a product-attributes extracted method based on templates which are automatically generated. First we extract nouns and noun phrases. At the same time we extract the verb and gerund structure by rules. They all form the candidate product attributes. Frequency information and stop-word dictionary are then combined to filter the candidate product attributes. And then we filter the candidate product attributes based on the template scoring. Finally, we classify the product attributes to master-slave attributes and single attributes. At this stage, we try to discovery new words and identify whether the new words are product attributes. According to the characteristics of different types of opinion sentences, we use appropriate opinion mining technology to deal with problems of semantic polarity. For sentences with different number of opinion words and product attribute words, we use different methods. There are two main methods to analysis sentiment:based on the opinion words matching and based on machine learning. Because based on machine learning methods require a large number of labeled corpuses and have a relatively poor portability. So based on the first method, this paper presents a new sentiment analysis method that based on attribute-opinion pairs to deal with the sentences which have attribute words and opinion words, and proposes a method for sentiment analysis based on opinion-sentence templates to deal with the sentences that have no option words. During dealing with comment sentences that have negative words, this paper proposes a negative shift algorithm.In the stage of report generation, this paper presents a method that based on hierarchical product attributes. We finally generate the report by the product attributes’ sentiment polarity and their slave attributes’ sentiment polarity.