Research On Key Technologies Of Semantic-based Information Retrieval And Association Recommendation

With the Internet technology continues to evolve,the speed of information quantity growth is quick. Users searching the information they need has becoming very difficult. Search engine is proposed to solve the trouble that users feel difficult to retrieve information. When users submit the search keywords to search engine, they can easily find the information they need in some way. However, the traditional search engine encountered problems. First of all, the traditional search engine based on simple keyword matching makes it hard to understand the users’ meaning accurately base on the user’s query. Secondly, search engines often returns a large number of results, and the user will usually only browse the first page or less than ten pages of results. Because of the low ranking of the result which the user needs, it can’t be seen by the user. Finally, although the recommended system has been used in a large number of business field, like online music and so on, it still have problems such as the data sparse problem which lead to the rate of recommendation accuracy drops etc. Based on the above issues, the semantic-based information retrieval and recommendation system is studied. And based on the existing achievements, I puts forward some new methods. The main work of this thesis includes:Based on the thought of software engineering, a method is proposed to build a new ontology: spiral ontology method. A risk assessment process is added to the method of constructing ontology, and I propose that the process of ontology construction has to be continuing to iterate to ensure the correctness of ontology. In the chapter the spiral ontology method and other common ontology construction methods are compared and an ontology will be constructed through the protégé software base on the spiral ontology method.I propose a new index structure suitable for Ontology and Lucene’s scoring method is improved. First of all, based on the analysis of Lucene’s code and the structure of existing index, I propose a new index structure which will be suitable for ontology. Secondly, based on the analysis of Lucene’s scoring algorithm, I propose a method which comprehensively considerate the keywords’ frequency of the query, users’ number of clicks on statistics, and the experience that can automatically mark the weight of keywords. After that I score the retrieval results and sort them. The experiments prove that new scoring algorithms and the index structure can improve the retrieval result.At last I propose a collaborative filtering recommendation algorithm base on Ontology. According to the analysis of the existing collaborative filtering algorithm, and focusing on the sparse user data matrix problem, I propose a new collaborative filtering algorithm based on existing theory. The experiment prove that using the collaborative filtering recommendation algorithm base on ontology can raise the accuracy of the recommended results.