Research On Algorithm Of Web Event Oriented Heterogeneous Media

With the rise of smartphone and the mobile Internet, people are easier to catch the events which are occurring and uploading on the Internet. We are in the world which is so-called “Sensing World”. While the information of Internet can reflect the events of real lives and using Internet to detect social events is really faster and lower cost, it gets more and more people to concern how to use Internet data to detect social events in real world.But Internet data has these characteristics: large size, high real-time, complex heterogeneous data type, the lack of high degree of information, seriously of information fragmentation. These characteristics have made the textural-based event detection become less effective. How to reduce the size of data, integration the different kinds of heterogeneous metadata, solve the problem of the lack of high degree of information and the seriously of information fragmentation which are commonly existence become more and more urgently. We design and realize multimedia event detection algorithms by these works.The first work focus on the large size of data, complex heterogeneous data type and the lack of high degree of information which exist in Internet data, this research puts forward a new approach based on time-slice and multi metadata fusion for multimedia social event detection. Firstly, we build a user-time model by time-slicing with user information and time information to reduce the scale of data. Secondly, the multi metadata fusion method and density-based clustering(DBSCAN) algorithm are applied to detect social events. The comparison experiment indicates that the new approach is faster and more accurate to detect the network social event compared with the existing methods.The second work focuses on the high real-time and unbalanced of Internet data, then combines with the incremental clustering algorithms which is called single pass to complete the social event detection. This paper designed a new approach based on Improved Single Pass Algorithm for online network event detection. Firstly, we using sliding time window to improve the algorithm runtime, as well as to reduce the shortcomings of single pass that items are more likely to clustering to big clusters. Secondly, we designed a new similarity method to solve the problem of unbalanced. At last, we find the best threshold by a series experiments. The result indicates that the new approach is more accurate to detect the network social event compared with the existing methods.