Research And Implementation Of A Distributed Intelligent Information Retrieval System

With rapid development of scientific technology and the varied forthcoming information, there emerge explosively various sci-fi documents, news materials and internet information, which requires an efficient search engine to support people to select the information they need. So far, the existing search engine systems are inclined to be in parallel and distributed architecture in order to improve users' response rate and to enlarge systems'retrieving scope. A brand new method of realizing distributed intelligence search engine is put forward in this thesis, which is different from traditional solution of distributed search engine. It applies Machine Learning technologies into Search Broker section that enables search engine with sufficient learning capacity and foundation for effective information filter. Meanwhile, the system provides Personalized Services according to users'research domain and interests. The thesis first introduces three classical models of information retrieval system and explains relative concepts and realization principles of large-scale distributed information system; then it gives out the quantum formula of evaluating the effectiveness of information system; finally it concerns about the application status of Machine Learning in information retrieval field and brings some ways of realizing intelligent information system. On above points, the thesis conducts the model of distributed intelligence info-retrieval system, which is based on the traditional information system architecture. The system uses vector space model as architecture in every distributed search joint and sort according to joint feedback. There is a search broker in every joint that function to distribute users'search request, centralize all joint feedbacks, and then return to users after enough processing. It describes that a CC4 Neural Network is added to search broker. Through the joint feedback as the sample, it extracts quantities of positive or negative examples for training and study so that the neural network can judge relevance among remained samples, filter out useless information and finally satisfy users'information request. Many experiment results indicate that the distributed intelligent info-retrieval system in this thesis is of great practical value.