Crucial Dates Detection In Continuous Social Media Event

With the rapid development of Internet,the popularity of smart devices has caused the explosive growth of social media user.When a sub-event happens,the majority of network users will report,forward and track the sub-event in the first time.The large amount of data contains a wealth of information that is difficult to find.The automatic timeline summary automatic detects key sub-events from social media corpus and displays the overall process of the event in a timeline.The key date selection is a very important sub-task in the social media automatic timeline summarization.The purpose is to detect the key date of the important sub-event and improve the accuracy of the detected date in the event.The existing work on traditional news gains candidate dates through date expressions,and then selects crucial dates according to the frequency of date mentions or the association among dates.While the posts in social media usually have fewer date expressions,traditional methods for crucial date detection on news corpus are not suitable for social media situation.On social media corpus,we observe that(1)there is an obvious burst of posts about the sub-event when an sub-event happens;(2)Different sub-events are subject to different durations of attention,and sub-events usually interact with each other.Therefore,we propose a Graphical Models with Content Relevance for Crucial Date Detection in Social Media Event(CRDD)to rank the dates and select the key dates.The model treats the publishing dates set of tweet streams as candidate dates set and uses the bag of word model to represent the texts collected by each date.Then we calculate the relationships among different dates based on the cosine similarity of the keywords embedded in different dates.A graph is constructed by combining the characteristics of burst and content correlation among sub-events.Based on the graph,we propose a random walk method is used to rank dates and select dates.The experiments on Twitter datasets about Arab Spring show that the proposed model is effective.