| With the popularity of the Internet,the amount of emerging network text resources has grown rapidly.This leads to the traditional manual classification method is difficult to classify the network digital resources reasonably and efficiently in time and effectively because of the low efficiency.Therefore,it is necessary to use automatic text classification technology to classify and organize them.There is a basic assumption in traditional automatic text classification that the training and test data belong to the same type of literature.However,as a new type of library,the digital library has books,periodicals,web pages and other types of text need to be sorted.Therefore,it is of significant for promoting the development of digital library to study automatic classification of multiple types of text digital resources.This paper proposes a classification method using Wikipedia to extend semantic feature to improve the classification effect of multiple types of text.In view of the problem of feature mismatch caused by different literature types,this paper associate the original mismatched words through the third party corpus,so as to solve the problem that semantic correlation can not be calculated between different types of text in the case of feature word mismatch.On the one hand,it can enrich the semantic features of the text to be classified,match the characteristics of the classifier trained by different types of text,on the other hand,the problem of feature sparse in text classification can be solved.The research contents of this paper are as follows:(1)Based on the the explosive growth of text resource on the Internet,this paper deals with the problem of classification with exponentially increasing network text in the future digital library.Therefore,it is necessary to study the automatic classification technology of various types of documents.Then this paper puts forward the idea of solving the above problems by feature expansion,and demonstrates the feasibility and applicability of the proposed text classification method by discussing and analyzing the current research results and progress.(2)In this paper,we propose a method for text categorization based on feature extension,which is the key step to eliminate semantic differences between different types of documents.In order to extend the feature,we need to extract a part of feature words from the training text as feature extension candidate word set.Based on discussion of traditional methods of feature selection problems and illustration of its shortcomings,this paper puts forward the principle and method of improving it,and show that the new feature selection method can indeed solve the original deficiencies through example.Finally,the improved feature selection method is used to extract the candidate word set,and the validity of the method is proved by experiment.(3)To solve the problem of feature mismatch in automatic classification of different types of text,this paper presents a text classification algorithm based on feature extension.In this paper,Wikipedia’s semantic relevance calculation method is used to measure the degree of correlation between feature words.After feature expansion,LDA topic model is used to represent the text,however,the traditional LDA model can not model the weighted feature words.Therefore,the weighted LDA model is proposed in this paper.As the characteristic words are given weights,the accuracy of the weighted LDA model is also improved. |
