Bingying Xu

Tracking the evolution of public concerns in social media

Microblogging is becoming a popular social media in recent years. Observations show that a large part of posts in microblogging were talking about public events occurred in the real world. Public concerns reflect interests and expectations of the mass for an event. Therefore, to understand and analyze of public concerns will help us to grasp an event, and predict its trend. This paper presents an evolution analysis method of public concerns for a special kind of post in microblogging, which can provides sufficient background information about an event by its attachments, e.g. a URL for details, a picture, or a video, etc. we called it expandable post. We use expandable posts to reconstruct the topic space. Their reposts are regarded as public concerns, and are located on the space. Thus, the task of tracking public concerns is transformed into tracking the movement of those reposts, and analyzing the relationships between them and their corresponding expandable posts on the topic space. The preliminary experiments on our dataset about H7N9 bird flu collected from Weibo, shows the effectiveness of our method.

Overlapping Community Detection on Dynamic Social Network

Overlapping is one of the common characteristics of the social network community structure. The existing overlapping community detection methods are rarely applied to dynamic network, this paper provides an overlapping community detection method on dynamic social network based on local fitness. Analysis the formation and evolution mechanism of social network communities, we not only consider the impact of increment of nodes and edges to the community structure, but also the impact of deletion of nodes and edges. Experiment on simulation network shows the effectiveness of the method.

Local Community Detection Using Seeds Expansion

The hidden knowledge in the information network has attracted a large number of researchers from different subjects such as sociology, physics and computer science. Community discovery has great significance for the analysis of information network structure, the understanding of its function, the discovery of its hidden patterns, and the predication of its behavior. In the practical life, people tend to analyze the information network with a heuristic method, that is, analyze the partial structure which meets the specific needs abstracted from the huge amounts of relational data. For this case, a method of community discovery based on seeds expansion is put forward in this paper. The node that should be paid special attention to in the information network is called the seed node, and then nodes with high similarity with the seed node are added through the iterative way. Accepting the idea of clustering algorithm, this method can not only find its community according to the customization node, but also find the outlier nodes of the community. Experiments on the public test set and data set of Sina micro-blog have demonstrated the effectiveness of the method.

Exploring Event Evolution Patterns at the Atomic Level

The event evolution mining for news corpus is beneficial for people who are less interested in the set of documents related by a topic rather than the underlying stories. Most of state-of-the-art approaches which derived from the TDT field considered events at the document level, which made different granularity for each event in evolution graph. In this paper, we consider events from a unified perspective by introducing the IE techniques into this task. We propose an unsupervised approach to explore event evolution patterns through extracting atomic event from documents, identifying their co-reference, and measuring their relationship automatically. And then we construct the event evolution graph. Meanwhile, we propose two policies to find a newsworthy subset from a mass of atomic events in a news corpus to simplify the event evolution graph. Finally, we show experimentally that our method which works on a Chinese news corpus can construct the atomic event evolution graph, whose vertexes are standing for important events of the topic, and edges are reasonable relationship between their adjacent events.

Event Evolution Analysis in Microblogging Based on a View of Public Opinion Field

Event evolution analysis, which focus on discovering underlying relationships among events by using methods of data mining on text corpus, is a meaningful and challenge problem. In recent years, more and more people began to express their opinion on public events though microblogging services. It makes that the microblogging corpus contains not only the facts related to the events, but also the public concerns. Therefore, we believe that the event evolution analysis in microblogging should take different approaches and perspectives with the state-of-the-arts. In this paper, we employ the concept of public opinion field, on which event information and public opinion in text corpus are distinguished. Based on this view, we focus on how does the public opinion affect the evolution of events, propose a method to measure the influence, and represent it on the event evolution graph. The preliminary experiments on our dataset about H7N9 bird flu collected from Weibo shows that our method can get consistent results with our intuitive feel, that illustrates the effectiveness of the method.

Predicting the social influence of upcoming contents in large social networks

Online social networks, such as twitter and facebook, are continuously generating the new contents and relationships. To fully understand the spread of topics, there are some essential but remaining open questions. Why do some seemingly ordinary topics actually received widespread attention? Is it due to the attractiveness of the content itself, or social network structure plays a larger role in the dissemination of information? Can we predict the trend of information dissemination? Analyzing and predicting the influence and spread of up-coming contents is an interesting and useful research direction, and has brilliant perspective on web advertising and spam detection. For solving the problems, in this paper, a novel time series model has been proposed. In this model, the existing user-generated contents are summarized with a set of valued sequences. An early predictor is adopted for analyzing the topical/structural properties of series, and the influence of newly coming contents are estimated with the predictor. The empirical study conducted on large real data sets indicates that our model is interesting and meaningful, and our methods are effective and efficient in practice.

Using Social Intelligence for New Event Detection in Microblog Stream

New event detection from microblog has a very practical significance for people who would like be aware of events in the first place. Although it is a traditional task of TDT, most of the state-of-the-art approaches are not designed for microblog, so they could not take full advantages of the social media, such as users and their relationships. In this paper, we propose an on-line method to detect new events from microblog stream. We establish the Individual Language/Memory model (ILM Model) to formalize the social intelligence of users, using their knowledge help us find out new events. The experiment on our corpus collected from Chinese microblog, shows the effectiveness of our model.

Local community detection using seeds expansion

The hidden knowledge in the information network has attracted a large number of researchers from different subjects such as sociology, physics and computer science. Community discovery has great significance for the analysis of information network structure, the understanding of its function, the discovery of its hidden patterns, and the predication of its behavior. In the practical life, people tend to analyze the information network with a heuristic method, that is, analyze the partial structure which meets the specific needs abstracted from the huge amounts of relational data. For this case, a method of community discovery based on seeds expansion is put forward in this paper. The node that should be paid special attention to in the information network is called the seed node, and then nodes with high similarity with the seed node are added through the iterative way. Accepting the idea of clustering algorithm, this method can not only find its community according to the customization node, but also find the outlier nodes of the community. Experiments on the public test set and data set of Sina micro-blog have demonstrated the effectiveness of the method.

Infer the probability of read in microblogs

In microblogs contexts like Twitter, a large number of users follow others. In case the author is not protecting his tweets, they appear in the so-called public timeline and his followers will receive all the messages from him. However, if followers of the author do not browse the personal page of the author, or they do not browse the timeline of themselves, they will not read messages of the author. So, followers of the author could not read all messages of the author. In this paper, we will infer the probability of read in microblogs according to the daily time-series model of posting and the similarity of personal interest. Experiments were conducted on a real dataset from Twitter containing about 0.26 million users and 2.7 million tweets. Experimental results indicate that out method is effective to infer the probability of read in microblogs.

Chinese New Words Detection Using Mutual Information

New words detection is one of the most important problems in Chinese information processing. Especial in the application of new event detection, new words show the current trend of hot event and public opinion. With the fast development of Internet, the existing work based on lexicon will not be capable for the effectiveness and efficiency. In this paper, we proposed a novel method to detect new words in domain-specific fields based on Mutual Information. Firstly, the framework of detecting new word is introduced based on the mathematical feature of Mutual Information. Then, we propose a new method for measuring the distance of Mutual Information by word instead of character. Comprehensive experimental studies on People’s Daily corpus show that our approach well matches the practice.