Research On Product Attribute Recognition And Its Application In Online Reviews

The analysis of users' online reviews in the automotive industry can effectively acquire user preference information.This information can not only provide a basis for potential users before shopping,reduce purchase risk,but also help merchants to locate market demand and make reasonable market decisions.In recent years,online reviews have gradually become the focus of public attention,and the research on online review mining has become a hot research field in academic circles.At present,online reviews mining researches are mostly limited to coarse-grained sentiment analysis.Due to the limited scope of application of the overall emotional polarity demonstrated by product reviews,fine-grained online reviews text mining research is an urgent need for product review text mining.The product attributes in the online product reviews reflect the topic information expressed in the review text.In view of the characteristics of large amount of data,non-standard content structure and too many short sentences of the online reviews,it is difficult to extract text features in the research of fine-grained text opinion mining,and the accuracy of text processing is low.This paper studies the identification of product attributes in online product reviews and its application in the analysis of automotive user reviews.The main research content is as follows:(1)Product attribute extraction based on hierarchical feature conditional random field model.Aiming at the problems of sparse online short review text data,domain difference and lack of valid features,this paper studies the influencing factors of topic information extraction in online product reviews,and extracts effective feature sets.For the oral presentation of online reviews text,the use of non-standard words and the incompleteness of existing dictionaries,some of the attribute words in the text are segmented and cannot be accurately identified,the method of expressing the internal label of the word is proposed,and the segmented attribute words are regrouped to improve the recall rate of the model recognition.Aiming at the characteristics of Chinese grammatical structure in word-bag model,this paper proposes a hierarchical feature conditional random fields model,considers multiple attribute position parameter information at the same time,introduces complex features,and studies hierarchical processing methods of feature function sets to improve the precision and recall of model recognition.(2)Product attribute classification based on Word2 vec.In view of the characteristic that the product attribute vocabulary is single and cannot be divided,the traditional feature extraction method is not suitable for the word level text classification,unable to obtain effective feature space,the grammatical and semantic features possessed by the word vector model are used to represent the words.Aiming at the problem that semantic spatial similarity clustering cannot distinguish the lexical category effectively,a feature fusion method based on word vector model is proposed,which combines the support vector machine to realize the word level text classification.(3)Research on the application of product attribute recognition in the analysis of automotive user reviews.This paper study the architecture of the opinion analysis system under B/S mode,design and develop data acquisition module,data preprocessing module,and product attribute extraction and classification statistics analysis module,analyze and obtain the user's preference information,implement a user opinion analysis system in the automotive field,and verify the practicability and effectiveness of the proposed method.