Mining Topics And Users For Organizations In Micro-Blogs

Micro-blog is a new means of information exchange and publishing. It provides users with fast, convenient and flexible services, and has gained worldwide popularity since it was created in2006. In this paper, we propose a system which tries to mine relevant information, including both topics and users for a given organization, which is of great significance for the enterprises and agencies to collect feedback and improve their product and service.The system will collect tweets and user network via API interface of the micro-blog service provider. The system includes a crawler that keeps collecting tweets by constructing queries with different keywords and monitoring the official accounts of the given organization. A classifier is trained to remove tweets that are not relevant enough. The system judges the relevance of a tweet on both its content-based and network-based properties. Moreover, the system keeps updating its classifier by adding new relevant documents to the training dataset, making sure the classifier can successfully maintain a set of features which can represent the organization’s current relevance. In the end, the relevant tweets are clustered into different groups according to their content. The paper also define the’UserScore’to measure users’ relevance to the organizations, in order to find users with either smaller distances to the organization’s official accounts or frequent participation in the discussion of the relevant topic. The system is as well trying to mining the group of users that display collective relevance to the given organization. The groups are expected to be highly relevant to organizations in the real word. In this paper, we use community detection algorithms to find user groups that are connected with the official accounts. According to experiments, the relevance of groups has strong correlation with the number of highly relevant users.