Research Of Constructing Multi-modal Semantic Knowledge Base

With the development of Internet, more and more data that are filled with all kinds offorms are generated. It becomes a very significant research topic that how to use thesedata well.Currently, there is no effective way to organize and manage these multi-modaldata. There is not dataset which is not good enough.Some contain a variety of modes, butthe organization of data in a single and semantic knowledge is not rich enough.Combining the advantages of the various data sets, proposed organizational approachbased on state-level semantic structure of the multi-modal data.Crawling multi-modal data is the foundation of the knowledge structure.In order toconstruct an effective multi-modal semantic knowledge base it reference method thatImagenet and other data sets use. At first, in order to guide crawling data and organize data,we construct a hierarchical semantic structure. For each node of the level of semanticstructure we use the method of theme crawling to get data sets corresponding to thissemantic node.Through analysing condtions of constructing this semantic knowledge base,the data in this paper are collected from the BBC News website.After series of data preprocessing,we need to re-organize the data. Since the data arebased on hierarchical semantic structure. Therefore, the data need to automaticallydiscover hidden semantics, so as to achieve the purpose of re-organization of the data. Inorder to better discover hidden semantics, semantic categories has two steps. Firstly, acombination of statistical mapping and Topic generation model is good for findingseveral large topic and achieving the purpose of subdivision of each topic,a combinationof hierarchical clustering and statistical mapping method performs well.Thus we get thegoalof re-organization of the data.Finally, the experiment proved the structure of multi-modal data sets in the semanticknowledge base classification is valid, as well as the multi-model semantic knowledgehierarchy is applied to the cross-media retrieval can improve the accuracy of retrieval.