Key Technology Research On Network Public Opinion Analysis

With the rapid development of the Web technologies, internet public opinion has been a important part of public opinion.Compare with traditional public opinion, internet public opinion has a large amount of data, emergency event and a wide affecting of features. For the mass of public opinions, way of management by human obviously can’t have a good understanding of the development of public opinions, and it can’t deal with emergency events very well in time.As internet public opinion has these kinds of features, it is important to find a way to analysis the internet public opinion automatically, and get the development tendency of internet public opinion to help relevant departments to deal with.This paper analyzes public opinion in emergencies events text classification and analyzes the classification of the emotional tendencies. This paper use the text classification technology of data mining to classify emergency events from a large amount of text.As most of emergencies reports are text, so the nature of emergencies event warning is text classification.The main research and creative point are as follow:(1) This paper improves text classification process based on machine learning,and the location information feature words paragraphs position weighted algorithm,and based on the length of the word length feature words and synonyms weighted algorithm combined weighting algorithm. Traditional feature weight calculation algorithm does’t taking into account the location information of feature words and semantic information. Title can be a good representative of the contents of the article,feature words from title are more representative of feature words in the body of aticle,and it is better to represent the type of article.so it is important to adjust the weight of feature words from difference part of article,the longer feature words get greater weight.There are many synonyms in a article, traditional feature weight algorithm is based on statistics, it will treate two synonymous words as two features, the proposed merger synonym weighting algorithm can solve this problem.(2) This paper improves the comment text sentiment orientation classification algorithm based on machine learning,in the process of calculation the weight of feature words,we adjust the weight of feature words which has a sentiment orientation,and do phrases semantics analysis,according to the adverbs, negative word and conjunction around the sentiment words,we create a formula to calculate the strength of sentiment and the orientation of sentiment.we use this to improve the sentiment classification algorithm which based on machine learning.Experimental results show that the algorithm based on the position and length of feature words weighting algorithm, and synonyms combined weighting algorithm improve the accuracy of text classification about2-3%.The sentiment word weighted algorithm which base on semantic analysis performs better than other sentiment orientation classification algorithm in big training dataset.