Research On The Method Of Public Crisis Event Mining, Analysis And Evolution

In the age of the explosion in network information, the network public opinion influence on people’s daily life more and more. The public crisis events in the network of public opinion often become the focus of people’s attention; some events threaten the stability of the society and people’s life and property security. So, recognize the public crisis events in time and analysis these events has important meaning for grasping the development direction of public opinion. This paper researches on the public crisis event recognition in network news and event evolution in micro-blog.This paper researches the public crisis event recognition in news’title. Because the trigger word is the core identification of event, this paper uses the method based on trigger word to accomplish the event recognition task. At first, we use the extended trigger word method to improve the trigger word method which can improve the recall rate. But it leads to a problem that the accuracy rate decreases. And then, we use the method based on rule to reform the recognition method and we propose a method based on Apriori to mining the keywords of the rule which can save more time than finding the keywords artificially. Besides, this paper studies the method based on statistics. We propose a feature selection method which marks words in sentences in two steps and use this method construct the maximum entropy model. Then we prove that it can improve in a little scale by using the method than without using the method through experiment and it works well on the test set.On the research of evolution of public crisis event, we build an emotion dictionary by using the NTUSD and common expressions in micro-blog. Then we study the emotion distribution based on other researcher’s study and point that the distribution fit the Poisson distribution. What’s more, we propose a method called "Incremental Iterative Approximation" which can be used to compute the fitting parameter. At last, we predict the trend of hot events by computing the difference between positive emotion words and negative emotion words. When the negative words have more weight, we regard that the event will trend to a bad way.In additional, we design and accomplish a public crisis event recognition system to combine theory and reality, which can recognize public crisis event in web news timely.This paper has two innovation points:1. Propose a feature selection method which marks words in sentences in two steps which can improve the performance of the maximum entropy model. 2. Propose a method called "Incremental Iterative Approximation" which can be used to compute the fitting parameter for Poisson distribution.