Automatic Summarization Of Multimedia Information And Related Technology Research,

The rapid Internet growth and the multimedia information explosion urge the research on effective techniques on information organization, generalization and analysis. The key of the research is the automatic summarization and inquiry of information. We need a system, which can answer the questions from text corpus and then synthesize and generalize the answers. At the same time the system should be capable of processing multimedia information. To build such a system, we must investigate automatic summarization and question answering system.Firstly we review the theory on automatic summarization, including the techniques of large scale text processing, text summarization and multimedia information retrieval. Then a cohesive framework of automatic summarization is proposed. It is expected to address the multiple media under the uniform framework.In this thesis we develop a multiple document summarization system which is the new tendency of text summarization. Our approach utilizes two level semantics relevancy among documents under same topic, that are the inter-documents relevancy and intra-document relevancy respectively. Our automatic multi-document summarization system combine a series of NLP techniques, such as text segmentation, text clustering and Vector Space Model. We also propose a new approach to compute similarity between two documents which utilize a relevancy lexicon. Domain-independent, un-supervisory and the easiness to implement feature in our summarization system.A useful complementarity exists between text summarization and question answering systems. QA system is one of the most important task of TREC. We develop our QA system on TREC-10. We make use of WordNet, a semantics lexicon, as a knowledge source to be integerated into our system. A Path-Finding algorithm is implemented on WordNet to find semantics relations between two words. Finally, in the Answer Processing module, we design and develop an algorithm which named "syntax constrained semantics verification". This approach of verification combines syntax and semantics features to verify the relevance between answer and question.Video summarization is another important task of multimedia information summarization. We propose a new approach of video summarization which combine NLP techniques and video analysis techniques. An automatic video summarization system which our multi-document system is integrated with is implemented. The approach is an attempt to broaden the research of automatic video summarization.