Research On Algorithms Of Mining Opinion Leaders Based On Micro-blog

At present, most of researches on microblogging opinion leaders are simplex instead of effectively compound research which includes users’ attribute-relationship,the network transmission and text information interaction as well as the study of relationship between opinion leaders and their emotion tendency. Most of the recent researches adopt static analysis which is not able to meet the dynamic change of opinion leaders over time.Aiming at the shortcomings of existing research, the thesis researches the related theories of mining opinion leader based on users’ attributes, rules of micro-blog dissemination and emotional tendency of micro-blog’s text. Furthermore the thesis also puts forward three sorts of algorithms. It includes:Firstly it presents an algorithm on users’ attributes character extraction on the basis of the improved TFN-AHP. It improved the traditional algorithm of TFN-AHP by constructing fuzzy precision matrix and using real number within the closed interval [0,1] as the scale value of the fuzzy judgment matrix to avoid errors which some users’ attributes were judged as 0 Arbitrarily in traditional TFN-AHP. More importantly, this algorithm has increased efficiency, the time complexity of this algorithm has declined from traditional4O(n) to2O(n) by using alterable accuracy iteration method to calculate eigenvectors of the fuzzy judgment matrix.Moreover, thesis extracted the eigenvectors of the users’ attributes based on this algorithm.Secondly the thesis proposes an algorithm based on the analysis of the users’ influence towards micro-blogging information dissemination. It puts forward a kind of thought that the cumulative number of microblogging forwarded and commented was used as the measure index of the capability of the micro-blogging information disseminating and thought that micro-blogging information dissemination was regarded as a kind of carrier of the influence of users by integrating large amounts of data to conduct research. This algorithm established the model of power law distribution with exponential truncation exponentiation distribution in the dissemination of micro-blogging. It achieved the objective of researching influence ofusers dynamically over time. It finds that the heat of micro-blog and users’ attribute determine their influence.Thirdly it comes up with an algorithm towards emotional tendency called POSTSPM which is based on the sequence pattern of part-of-speech tagging. This algorithm thought that micro-blog’s text is regarded as an ordered part of speech sequence composed of a combination of several words. It preserved the main part of speech sequence which determines the emotional tendency of micro-blog’s text by simplifying the words feature tagging sequence and adopted sliding window to match the parts of speech tagging sequence by using a matching pattern based on micro-blog emotional tendency and the How Net.The algorithm is able to determine the emotional tendency of the whole micro-blog’s text, and the accuracy is higher.At last, considering the dynamic changes of the opinion leaders of micro-blog,the thesis makes a mining algorithm that the opinion leaders of micro-blog could be found on the basis of specific topics based on the time window by three algorithms mentioned above. It made an operation that time period of the Topic of discussion was divided into several successive sub-windows and selected opinion leaders by calculating the rank of the influence of users in every sub-windows and the emotional tendency of users’ micro-blog’s text. The remarkable effects of those algorithms have been testified by experiments and result analysis.The algorithm theories has been able to applied to mining the opinion leaders of micro-blog.