Research On Mining Techniques Of Product Review Based On Sentiment Analysis

Reviews on the Internet implied the views and preferences of the user,such as demand and emotional information.These information can be used in product recommendation,new product development,and sales model design.Because the reviews of Chinese natural language expressive are ambiguous,uncertain and expressive diversity,especially the non-normative nature of the network review in the form of expression,there are many difficulties and challenges in review mining.This paper will propose a method of review mining based on domain ontology.The basic idea is to construct a knowledge model which contains the features,attributes,emotional words and semantic relations between them.We can accurately extract the features described in the review and the corresponding emotional words,and solve the problem that the same emotional word in the description of different features may express different emotional tendencies.The specific work is as follows:(1)Constructing a domain ontology for review mining:The domain ontology regards the features of the product and the emotional words as the conceptual nodes of the ontology,and describes the semantic relations between them.Through this domain ontology can identify the product features in the review and improve the accuracy of the sentiment analysis.(2)Proposing a similarity calculation method based on word vector:The similarity between the concepts is obtained through the word vector.First,we train the review corpus by word2vec tool,and the words appear in the corpus are expressed as low-dimensional vectors in the vector space.By calculating the distance of the distance of the word vector,the similarity of the lexical semantics is obtained.(3)Proposing semantic analysis method based on the domain ontology and emotional dictionary:On the basis of the domain ontology,we use the emotional dictionary to calculate the emotional intensity of the features in reviews,and consider the influence of the summary words,turning words and opinion words on the overall emotional tendencies,we propose a new method of review classification.The experimental results show that the method can improve accuracy rate,recall rate and F1 value.(4)Based on the domain ontology implement the hotel review mining system:Based on the above research results,we implement the review mining system from users needs.The system gives out the hotel's overall review information and the various features of emotional information.Users can quickly get the desired information,So as to help them make the right consumption decision.