A Designing And Implementation Of Network Content Audit System Based On Text Analysis

Nowadays, the network overflows with baneful information, such as pornography, reactionism and violence etc. As to the task of filtering them out, the topic_based classifiers' ability do not equal to their ambition. We believe the word's information could be classified to 3 classes: foreign to the topic, positive to the topic and negative to the topic. To accomplish the filtering mission perfectly, we put the text polarity_based filtering forward.The work we mainly did are as follows:Firstly, aiming at the high dimensionality and the Sparsity problem of the classification, we brought a novel feature selection algorithm combined the CHI with the IG forward. It is benefit for the characters with high frequency and characters that only appears in a few classes but not with so high frequency.Second, in order to filter the topic-free information out effectively and to draw up the users' true demands, we designed a topic filtering model based on the feedback policy and the SNoW algorithm. It has the function of filtering the irrespective information and classifying the network text to the right class.Third, we established a mechanism which could recognize the polarity of the text. It syncretizes rules base method and the Stat. based algorithm. Taking the identified phrases, BOW etc. as characters, the SVM classifier gain a better result.Finally, we developed a Network Content Audit System Based on Text Polarity Analysis which has the quality of high efficiency and high precision.In a word, the thesis expounds the designing of the Network Content Audit System Based on Text Polarity Analysis in detail, emphazing on the Combined Feature Selection Algorithm, the semantic mechanism of recognizing the text's polarity. The approach we put forward holds meanings to further research of text polarity.