Research On Relevance Feedback In Vector Graphics Retrieval

The objective of relevance feedback is to learn from the actual interaction between the user and retrieval system, discover and capture the user's actual retrieval intention, and thus to amend the retrieval strategy of the system to get the search results that will as much as possible fit the user's actual needs. Introducing relevance feedback technology into vector graphics retrieval and by means of non-memory feedback and memory feedback can effectively make use of the user's feedback information to capture the user's retrieval intention, so as to improve the performance of the system.Firstly, the research situations of vector graphics retrieval and relevance feedback technology are analyzed. According to the retrieval requirements of vector graphics, a system framework of the vector graphics retrieval based on relevance feedback is proposed and the involved key technologies are elaborated. After that, on the basis of extensive analysis and research on existing learning algorithms and relevance feedback algorithms, a relevance feedback algorithm based on combining classifiers is proposed, which combines the expected results from the independent nearest neighbor classifiers with only one training sample formed by each positive or negative feedback sample, computes the relevance score of every vector graphics and optimizes the relevance score by introducing the technique called"Bayesian Query Shifting". The algorithm can take full advantage of the information proposed by each positive or negative feedback sample, besides it can further improve the precision of the vector graphics retrieval system. Next, in view of the problem that the existing retrieval system is lack of long-term learning about the user's intensions, a personalized retrieval algorithm is proposed. The algorithm adopts a long-term learning strategy, defines the semantic correlation between two vector graphics according to feedback log and establishes personalized and common semantic correlation matrix. On this basis, establishing the user model to realize the personalized retrieval which can make the system gradually adapt to the cognitive habits of different users and effectively improve the system retrieval performance. Finally, the implementation of vector graphics retrieval system based on relevance feedback is presented. Based on the system, the experiment platform is built to verify the effectiveness of the proposed algorithms in this paper.