The Study On Agricultural Information Mining On The Internet Based On Intelligent Agents

Traditional search engine can't meet people's information needs. Facing the ocean of information on Internet, people always get too much information to select and to digest, so it's hard for them to find the necessary information and knowledge. Web mining is a new technology and it can discover resources and knowledge quickly and efficiently, therefore, it can compensate the shortcomings of search engine.The aim of the thesis is to help people who search and mine domain information on Internet. Based on the study of Web mining and intelligent agent technologies, a domain oriented information mining model is presented. Around the building and software development of the model, the main work and innovation of this thesis are listed as the following:Firstly, the presented model has a multi-agent architecture and orients the practical domain. In the model system, domain, user, and document representations are depicted. Documents are the objects of information process and the foundation of document model is the base of information process such as feature extraction, document filtration and so on. The representation of domain model by using the topic concepts and the key words emphasizes the domain oriented feature of information service, so it can realize the personalization, intelligence of the information service system. User model embodies user interests and intention. It is used to express and mining user interests.Secondly, in the information retrieval agents, a key problem is how to extract text features and how to deal with the high number of the feartures. The system firstly learns the domain training samples by using thesaurus to process word-separation and word-frequency statistics. According to word-frequency distribution, it chooses the feature collection and their weights to formulate feature vector and generate domain model and user model. Then, retrieving-monitoring agent uses the feature vector to search information and document on Internet. Analysis-filtrating agent extracts the document feature to formulate structured representation of the document.Thirdly, selecting-recommending agent combines the content-based and social-based document recommendation methods. Content-based method focuses on the relation establishment between documents and users by text contents, i.e. the key words. However, social-based method focuses on the relation establishment between documents and users by the evaluation of users on domain. In content-based method, this kind of agent searches the user with the most similarities and then recommends the chosen documents to the user by comparing the structured document with user model. In social-based method users can provide URLs and other documents to the system, and the agent can recommend URLs and other objects according to compare the similarities among users.Fourthly, leaning agent adjusts the domain model and user model adaptively by reinforcement learning and genetic algorithm. The learning process of the agent is the process of user feedback, and also the mining process of user interests. The system adopts both the explicit and implicit user feedbackmethods. Recording user browsing behaviors, obtaining user evaluation on documents and the documents retrieved form Internet, the three factors are the original reinforcement signal. The signal is spread to domain model and user model by their interaction and then adaptively adjusts the system. The system also employs genetic algorithm to adjust user model actively.Finally, the thesis realized an agricultural information mining system named WMS.