| In the Internet,short text is one of the most common forms of data.It is widely used in user comments,news title,weibo and personalized signatures.How to classify massive short text data to facilitate people’s information acquisition and query is becoming the focus of researchers.Due to short text is characteristic by short length and sparse features,it is difficult for traditional text classification algorithms to extract short text features.At present,feature expansion of short text is still the main way to solve the problem of short text feature sparseness.Therefore,this dissertation studies the problem of feature expansion in short text classification from two aspects: the content of short text itself and the introduction of external knowledge.Based on the content of short text,this dissertation proposes a short text classification algorithm based on dependency syntax knowledge.The syntactic structure of short text contains rich semantic information.Therefore,this dissertation uses the syntactic information of short text to expand the features of short text.Firstly,the dependency syntactic structure of the sentence is obtained by dependency parsing.Then,based on the feature that the lower word is dominated by the upper word in the dependency syntax theory,this dissertation uses the information of the upper word to explain the meaning of the lower word,so as to improve the ability of describing the lexical semantics.From the perspective of introducing external knowledge,this dissertation proposes a short classification algorithm based on keyword feature expansion.In order to solve the problem of sparse short text features,this dissertation introduces Chinese Wikipedia corpus to expand the keyword features in short text.Firstly,the Text Rank algorithm is used to extract keywords from short text.Then,the Chinese Wikipedia corpus is used to train the word vector.The similar words of keyword are obtained according to the distance of word vector and part-of-speech constraints.And the similar words are used to supplement the keyword features.Finally,this dissertation proposes a SYN-CNN model to make full use of the lexical information of the text and the similar words features of keyword.The model uses convolutional neural network to extract vocabulary features and keyword expansion features respectively.The extracted results are concatenated as the feature vector of short text,and the extended feature of keywords is effectively introduced into the classification model.Aiming at these two feature expansion algorithms proposed above,detailed comparative experiment is carried out on the Chinese news-title dataset.The experiment results show that these two feature expansion methods can effectively improve the description ability of short text features,and then improve the classification performance of short text. |
