| With the development and popularization of the intemet,web reviews have become an important way for people to obtain the product information,of which online book review is a kind of typical cultural product review.Compared with the traditional entity or service product reviews,book reviews have no fixed pattern or normalized structure in form,and they are more wide-ranging and abstract in content.Besides,they are also generally quite colloquial and freely,which make it a challenging and frontier research of the book,review text mining.This paper chose online Chinese reviews of science&technology books as the research object,aiming at exploring and implementing a text mining method of such reviews,and also presenting the results to readers with the form of text summarization to help them get the book information.The research work of this paper mainly consisted of three aspects as follows:(1)Content analysis of book reviews and construction of the text mining framework:for the purpose of confirming the main information in the online reviews of science&technology books,this paper firstly made a qualitative analysis based on a large amount of Chinese book reviews.We Divided the content information into seven categories and labeled each sentence through manual way.Then a quantitative analysis was done respectively from the perspective of sentence and discourse to find out the distribution of the information categories,and we confirmed the content of books,the subjective comment and the target readers as the core content of the seven kinds of information.Based on the content analysis,this paper chose these three kinds of core content as the main mining object and constructed a text mining framework towards them.(2)Implementation of the text mining framework:based on the traditional product review mining method,this paper designed a text mining process towards the core content of Chinese book reviews to implement the above-mentioned framework.The process mainly include three steps:?Text preprocessing of book reviews:this paper mainly introduced a way of semantic sentence segmentation based on the dependency relations and the way of text representation based on the vector space model.The semantic sentence segmentation way is an improvement of the common method utilizing punctuation,which can enhance the accuracy on some level and contribute to the task of text categorization and information extraction.The text representation part mainly introduce two ways of feature selection and the TF-IDF way of weight calculation.?Text content identification of book reviews:this paper adopt the text classification method to identify the three types of sentences containing the core content.We compared the document frequency method and information gain method in the basal experiment and chose the latter as the main way as it had a better performance.In order to improve the classification performance,we used the SMOTE method to balance the dataset and extended the short text features based on word embedding.In addition,this paper proposed a construction way of sentiment lexicon of Chinese book reviews utilizeing word embedding,based on which we divided the subjective comment sentences into positive ones and negative ones.?Content information extraction of book reviews:this step aimed at extracting the important information from three types of sentences at a fine-grained level,and we proposed a way of information extraction based on the dependency grammar analysis of sentences.We worked out different rules and algorithms according to characteristics of the three types of information and verified their feasibility by analyzing the precision and recall values of the results.(3)Application of the text mining results:this paper explored the application and display of the mining results,and designed a text summarization template integrate the information.In the end we evaluated the quality and usefulness of the summarizations using questionnaires.The assessment results showed that the text summarizations generated in the way of our research were in high quality,which on the other hand also verified that the text minig way of Chinese book reviews proposed in this paper had a certain feasibility and practicability. |
PDF Full Text Request