| Today, the Internet has become the most efficient, convenient and widely spread media of news. Net news is characterized by having large amount of information, high instantaneity and rapid growth, but news that concerned by organizations and individuals have a characteristic of strong specialization and weak time varying. How to automatically recognize the information that users'long-term concern about from the dynamic and large amount of net information, and provide the users with information service actively has important social significance and practical value.The thesis is aiming at a special application for automatically building enterprise portal's news section, based on researches on related technologies about reorganization and retrieval of net news, using system architecture of meta search engine and combining technologies of Intelligent Information Processing such as distributed information retrieval and merge, text classification and reorganization based on context and domain ontology to design and implements a specialized topic news automatic retrieval system. The main work and contribution are listed as below:1. Researches are conducted on technologies of meta search engine realization, and studies and discussions are focused on result merging which is the key link of the research. Then we propose a method of using PSO(Particle swarm optimization) algorithm to optimize the merging results of multiple resource retrieval systems, and verify the validity of the algorithm by a group of experiments on the retrieval of electronic periodical literatures. Now we apply this system into the structure of special topic news automatic retrieval system.2. Studying and analyzing some key technology issues of selection and processing of text feature and kernel function in text classification and topic reorganization based on SVM. The experimental results show that when using IM algorithm and set the number of features as approximately 4000 and the kernel function as SIGMOID, the recognition accuracy of news text classification is above 97%. Now these works have been merged into the target system.3. A special topic news automatic retrieval system are designed and implemented. By a set of topic keywords and topic sample database provided by users, using query semantics in which topic keywords are extended by ontology, using meta search engine architecture to collect topic news, and push to users by mergence and filtering with topic reorganization. Now, the main architecture and prototype of the target system has been completed, some core modules have been able to excute. |
PDF Full Text Request