Text Retrieval System Research Based On Frame And Description Logic

At present, most information retrieval methods are based on keywords matching.Its core is the mechanical matching of key characters, ignoring the relation of thedocuments on semantic level. If users can not use exactly keywords to describe thewillingness, the precision of the retrieval system will be low. Knowledgerepresentation method based on the framework can establish a similar context toretrieve documents, and according to users’ understanding, retrieve relevantinformation in the framework. This method can effectively compensate for the defectsof keyword searching.At present, most information retrieval methods are based on keywords matching.Its core is the mechanical matching of key characters, ignoring the relation betweenthe documents on semantic level. In order to solve the problem of low accuracy in theexisting text retrieval system, we proposed the method combined with semanticframework and description logic in this dissertation. In knowledge representationlevel， we used knowledge representation method based on semantic frame torepresent information structure and content of a document through a multi-levelsemantic slot; In the query reasoning, we used description logic reasoning method tofill semantic slot information in the semantic framework completely. The researchwork in this dissertation mainly includes the following aspects:1. The text retrieval system both combined with framework and description logicis proposed. In this method, the semantic framework is used in knowledgerepresentation level in order to convert unstructured text into structured semanticframework.2. In the processing of mining semantic slot information in the default semantic framework, this paper proposed the reasoning method based on description logic. Theproposed method adopts the Racer inference engine based on Tableaux algorithm. TheRule-based reasoning model is proposed according to the reasoning. The method canreason the default and hidden information in the framework.3. In the processing of the similarity calculation, the weighted semanticsimilarity calculation method is proposed. In this calculation formula, Word Netdictionary is used to find all the synset IDs of the two words firstly, then the number ofsynset can be used to adjust the weight w.4. Finally, a binding framework and description logic text retrieval system wererealized, and the test data set provided by the Text Retrieval Conference TREC (Textretrieval Conference) were used to design and implement the experimental system, andmean average accuracy (MAP) was used to evaluate the experimental results.At last, by comparing binding framework and description logic text retrievalwith the traditional text retrieval, the precision and recall rate effects have beenimproved greatly, and the final quality for text retrieval have made great contributionsfor improving the quality of text retrieval.