Unstructured Information Search Based On Ontology Semantics And Object Feature

The rapid development of mobile Internet and the swift popularization of mobile intelligent products accelerate the booming of the Web 2.0 and social media,and then the vast amounts of unstructured information springs up on the Internet.How to search the needed information quickly and effectively from the massive unstructured information for users,which is an important challenging research subject in information retrieval.There are two main problems in existing information retrieval systems.On the one hand,the traditional keyword-based search engine utilizes the absolute matching technology of keywords for text indexing and searching,but the ambiguity of natural language leads to the low recall and precision for search,in addition,the ranking of the returned results is not reasonable.On the other hand,Content-Based Image Retrieval(CBIR)technology uses visual content to retrieve images,however,the semantic gap between low-level visual features and high-level semantics exists,which seriously interferes the performance of CBIR.In order to solve above problems,especially how to analyze effectively and search efficiently for the rapid growth of the mass data(such as text,image,etc.)on the Internet,guarantee the accuracy of the returned results and the rationality of the ranking of the returned results,and meet the query input with the diversity of sources for users,in this thesis,the related theoretical foundation of information retrieval based on ontology semantics is studied.Moreover,we propose an Unstructured Information Search Pattern based on Ontology Semantics and Object Feature(UISPOSOF).With the basis of the UISPOSOF,a prototype system,which is Unstructured Information Search System based on ontology semantics and object feature(UISSOSOF),is designed and implemented.Specifically,in this thesis,the main contributions of the work can be summarized as follows:(1)We propose an approach to constructing a domain ontology based on a cycle of Plan,Do,Check,Act and Evaluate(PDCAE)and construct ancient ceramics ontology.In view of the present recognized more mature domain ontology construction method(i.e."Seven Steps" is put forward by School of Medicine,Stanford University)without considering ontology evaluation stage,it will lead to the quality of domain ontologies varies tremendously.Therefore,we introduce PDCA cycle and add ontology evaluation stage,and then propose an approach to constructing a domain ontology based on a cycle of PDCAE to make the domain ontology have the characteristics of cycle and evaluation feedback,which is more benefical to construct domain ontology with high quality.Currently,the existing researches on domain ontology evaluation with comprehensive standard have not discussed the membership algorithm of index evaluation set in the ontology evaluation stage of this proposed approach.Hence,a membership algorithm of evaluation set,an evaluation index system and a computational model of comprehensive score for domain ontology are performed int this thesis,they provide a reference for quantitatively evaluating the quality of the domain ontology or selecting the domain ontology with high quality.There are few researches on ancient ceramics ontology,it limits the knowledge sharing and reuse of ancient ceramics,and the development of its related applications.Therefore,an ancient ceramics ontology is constructed.(2)We propose an approach to measuring semantic similarity and relatedness between two concepts in an ontology.In order to search the needed information based on semantics by using ontology concept,which can be used to express the user's query requirements,it is necessary to measure semantic similarity and relatedness between two concepts in an ontology,and then the semantic similarity and relatedness determine the degree of semantic matching between ontology concepts and user queries.Currently,the existing similarity and relatedness measures are not give a fully consideration for the factors that affect the semantic similarity or relatedness between two concepts,and they are not fully utilized the ontology semantic knowledge.In view of this,we propose a comprehensive metric of semantic similarity,a method of semantic relatedness measure and a comprehensive degree metric associated semantic similarity with relatedness.After that,we compare the proposed approach with other nine similarity/relatedness measures based on the benchmark data set and WordNet 3.0(which is a semantic knowledge base)in an experiment.The experimental results validate the effectiveness of our proposed approach.(3)We proprose an approach to extracting important local features of images based on the Singular Value Decomposition and Scale Invariant Feature Transform(SVD-SIFT).In view of the present researches of information retrieval based on ontology semantic are not considered the feature of unstructured information to sort the returned results,it may lead to that information search based on ontology semantics can not distinguish individuals of the user query input for describing the unstructured information with same semantics.Here,we combine semantic concepts and the feature of unstructured information(i.e.unstructured object feature)corresponding to individuals in an ontology for information retrieval,therefore,it can solve the problem that the returned results are more and the sorting of the returned results are unreasonable.How to extract the feature of unstructured objects is the key to solve this problem.In unstructured objects,for the image feature extraction,an approach to extracting important local features of images based on SVD-SIFT is proposed,and then we employ the global features(including invariant moment and colour histogram based on HSV space)and important local features(SVD-SIFT)of an image to describe the image.After that,we compare the proposed approach(which combines global feature(i.e.colour feature and shape feature)extraction and important local feature(i.e.SVD-SIFT)extraction)with four methods(i.e.colour histogram,invariant moment,SIFT,SVD-SIFT feature extraction)in an experiment of image search.The experimental results show that our proposed approach significantly reduces the complexity of computing the similarity for image visual features and balances the effectiveness and efficiency of content-based image retrieval.The proposed approach provides original foundations for unstructured information search based on ontology semantics and object feature.(4)The Unstructured Information Search Pattern based on Ontology Semantics and Object Feature(UISPOSOF)is proposed.Specifically,this thesis presents the computing strategy of a similarity measure,which associates ontology semantics with unstructured object feature(i.e.the feature of unstructured information).The strategy(we called it the semantics fusion grading similarity measure)is more beneficial for large-scale unstructured information retrieval.Therefore,we propose a search process for unstructured information based on ontology semantics and object feature,the search process involves some vital algorithms,which include semantic search algorithms based on concepts in an ontology or a query image and search algorithm based on semantics fusion grading interaction.On the basis of above research results,it lays a theoretical foundation for the implementation of a prototype system,which is an unstructured information search system based on ontology semantics and object feature(UISSOSOF).(5)Based on the above research works,a prototype system(i.e.UISSOSOF)is designed and implemented.The system supports ontology concept-based search,query image-based search,and their combinatorial search;especially,it has a certain ability of semantic reasoning.On the basis of UISSOSOF,a comprehensive analysis and evaluation are developed for the proposed search algorithms by using a large number of real individual information(mainly includes texts and images)in ancient ceramics,which include the analysis of search cases,and the comparative analysis and evaluation for different query patterns and algorithms.The experimental results validate the effectiveness and high-efficiency of the proposed search algorithms in this thesis.Moreover,by comparing the performance of the proposed search algorithms with the traditional keyword-based search and Content-Based Image Retrieval algorithms,we observed that the performance of the proposed algorithms have significantly improved.At the same time,the results also validate the feasibility and practicality of UISPOSOF.