Research On Multi-dimensions Dynamic Relationship And Event Discovery In Dataspace

Discovering relationship between heterogeneous data including structured,semi-structured and unstructured data is an important research in dataspace area. There are two main approaches about relationship discovery: One is to use the Apriori algorithm to find the relationships confirming to the specific association rule, which is often used to discover the relationship between different items and attributes stored in structured data set such as database; The other one is to use semantic analysis, which discovering the underlying schema through the analysis of predefined relationship to find some new relationships. Furthermore some new relationships are discovered through these schema. However, such approaches are not applicable for discovering relationship that can not predefined and they process only one data structure. There is no uniform method of discovering the association of different data structures.Research on dynamic event discovery most focused on the discovery of topic and tracking technology. The methods of discovering topic use the cluster analysis to put the news which report the same topic together, every cluster represent a topic. Along with the development of social network some researchers proposed the method of discovering event based on weibo. This method use short text like weibo as data source, and then also use cluster analysis to generate events.First,this paper some of previous work on traditional methods of relationship and event discovery are conclusived. And then proposed a dynamic relationship discovery method in Dataspase. The data sources of this method are user query logs and news archive. First, we can discover the relation entities based on the time window and then build multi-dimensions inverted index for unstructured news archive. The final relationships are corroborated by computing relation strength between entities using index information. After discovering relationship entities, we use these entities to form an EDR graph. We propose two algorithms which are LTC and GTC for computing the event cluster, every cluster is an event. The method of discovering dynamic relationship and event in this paper can detect the associations that can not predefined and predicted, What's more, this method can update related entities along with the evolution of events and detect events based on the discovered relationship entities. Experiments show that this method can discover relationship entities and event more efficiently and has good performance.