Research On Semantic Retrieval For Heterogeneous Engineering Documents

Knowledge-based economy promotes manufacturing informatization and intellectualization, and knowledge-intensive industry is further developed. Knowledge-intensive product design, analysis, and manufacturing use and generate a large amount of data, information, knowledge, for example, processes employ different computer systems for analysis and processing. As a result, the number of engineering documents is increasing rapidly. Besides, engineering environment is complicated, because manufacturing involves many parallel or serial processes, which need the support of multiple disciplines knowledge and collaboration between different engineers and teams. This is a challenge for knowledge management, sharing and reusing in knowledge-intensive industry. And the key is to understand the content of heterogeneous documents and retrieve them semantically in complicated engineering environment, which has great value for product innovation, product quality assurance, and enterprise informatization.This work is supported in part by a grant from the National High Technology Research and Development Program of China (No.2009AA04Z151) and the Zhejiang Natural Science Foundation of China (No.Z1090461), aiming at some problems in semantic retrieval for engineering documents, such as semantic representation and semantic inference. The main works are listed as follows:1) A heterogeneous engineering document semantic retrieval approach based on multi-dimensional association is proposed.For the heterogeneous engineering documents in product design, analysis, and manufacturing, current retrieval methods don’t consider the multiple resources and heterogeneous characteristics of engineering documents, and can’t analyze context semantically. The proposed framework for heterogeneous engineering document semantic retrieval based on multi-dimensional association includes multi-strategy content capture, semantic annotation, multi-dimensional association construction and representation, query analysis and structuring, context construction and inference, query refinement, semantic retrieval, result ranking, and feedback. The proposed approach can overcome document heterogeneity to understand document content semantically, associate documents at different facets and levels, infer query intents based on context semantically, improve query processing and retrieval precision, and realize intelligent retrieval across heterogeneous engineering documents.2) A multi-dimensional association representation method based on ontology and Topic Maps is proposed.Considering characteristics about engineering document association, we propose multi-dimensional association method for knowledge organization, which contains content-based internal association and processing and evolution-based external association. After analyzing the characteristics of ontology and Topic Maps, a multi-dimensional association representation method based on ontology and Topic Maps is proposed, which constructs semantic model based on ontologies, and expand association scope by Topic maps, which is used to associated special knowledge. The multi-dimensional association representation considers both general and special knowledge in documents, combines internal and external associations, analyzes and organizes heterogeneous engineering documents at semantic level, which supports knowledge management, retrieval, semantic inference and navigations.3) A local context model based on multi-dimensional association and some context operations are proposed.Context model is a knowledge support for semantic understanding and engineering knowledge applications. Considering the requirements of engineering context modeling, we propose context model based on multi-dimensional association, which is formally defined as a quad, and further introduce corresponding operations. The context model consists of core entity set, contextual association set, rule set and size space. According to the application scenarios, we propose two types of context operations, which are context enrichment and context shifting. Context enrichment contains context supplement, expansion, and contraction. Context shifting contains context lifting and lowering, core entity based change, and context merging. Compared with current researches, our method considers the dynamics of context, describes context knowledge at sematic level, meets the requirements for content diversity, and supports for semantic inference in various application scenarios.4) A context-based query processing and retrieval approach is proposed.Context can be used in query intents understanding and retrieval, but context models in current researches are static or ambiguous, which limit the flexibility and explicitness in semantic understanding and inference. To solve the problems above, we propose a context-based query processing and retrieval approach, including processes of query analysis, query structuring, context construction and inference, context-based query refinement and retrieval. The proposed approach utilizes semantic inference algorithm to construct context model for specific query task. The context provides semantic understanding of background knowledge for query and retrieval, which can guarantee the flexibility and explicitness of query processing. The proposed context-based query refinement and retrieval algorithm calculates confidences of query objects and conditions to identify the semantics in query and retrieval constrains and captures query intents, which can reduce complexity of retrieval and improve retrieval precision.