Multi-level And Fine-grained Sentiment Mining Of Network Review Based On Domain Ontology

As online shopping continues into the lives of the general public,online comment has become one of the most important means for people to communicate and obtain product information.It is an urgent need for consumers and enterprises to quickly and effectively get rid of the emotional attitude of the people to the products from the large-scale network reviews.The traditional fine grained emotion analysis is mostly based on single attribute or attribute class,ignoring the hierarchical structure and attribute dimension weight between attributes,which can not meet the needs of today's users.In order to study the hierarchical relationship between the attributes of the product and the emotional attitude of consumers to the local and overall attributes of the product,optimize the result of emotional analysis,this paper constructs the product domain ontology,uses the ontology to extract the product attributes and constructs the product attribute hierarchy model,the implicit attribute is identified by calculating the weight of the collocation between emotional words and attribute words.The domain emotion dictionaries are constructed and the emotional tendencies of the product attributes at all levels are calculated.From the three levels of the product overall,the attribute class and the single attribute to realized the fine grained emotion analysis of the product.The specific research contents are as follows:First of all,research on the construction of product domain ontology,product attribute extraction and hierarchy division.Based on the ontology theory and construction method,based on the online reviews,existing ontology and product instructions to collected the related concepts and terminology in the field of product,the logical structure between semantic relations and concepts is combed,and the final domain ontology is constructed.Based on the ontology model,the explicit attribute of the product is extracted,and the implicit attribute is extracted by the improved TF-IDF algorithm.Finally,the hierarchical product attribute model is built according to the concept relationship in the ontology.Secondly,based on the general emotion dictionary,through the network reviewsto extract the special vocabulary,with the network popular words,the final static emotion dictionary is formed.By analyzing the context of word collocation in product,a dynamic emotion dictionary is constructed,which solves the problem of computing emotional tendency of dynamic emotion words.Besides,it not only constructs negative words,degree adverb dictionaries,but also adds pictures,symbols and other special emotional elements dictionaries,and the emotional intensity of each factor in the sentiment dictionary is marked.Finally,in view of the deficiency in traditional sentiment analysis,we discuss the importance and emotional weight of attribute dimension.According to the emotional weight,based on the emotional dictionary,the emotional computing method of product from the whole,attribute class to every specific attribute is designed.Empirical Study by selecting Mobile Phones as an Example.By adopting more than 10000 online comments of XiaoMi 6 and HUAWEI mate10 two mobile phones on the JD mall,the empirical study carried out a specific analysis and expansion of the construction of the ontology,product attribute extraction,the multi-level fine-grained emotion calculation and etc.And through the two groups of comparison experiments: the traditional SVM affective analysis method and the feature class affective analysis method,from three indexes of recall,precision and F value to evaluate.Compared with the contrast experiments,it was shown that the algorithm in this paper was obviously improved in the accuracy of emotional classification.Based on the above research results,the research results and conclusions are summarized,and the shortcomings and future prospects of this paper are analyzed.