Research On The Applications Of Agent Coalition And Manifold Learning In Chinese Question Answering System

One of the important features of information society is information retrieval. Various searching engines are useful to people. But how to make searching engines understand the needs of users'search to get more exact result is the goal of question answering system (QA). Question answering system is a branch of information retrieval, it belongs to accurate retrieval. This thesis describes content of question answering system, relevant technologies and status of Chinese question answering system (Chinese QA) systematically, introduces multi-agent technology and manifold learning into research and implementation of Chinese question answering system. The main works in this thesis are as follows:1) Manifold learning is applied to improve precision of question classification.Question classification plays a very important role in question answering system, its result is a very good guide to answer extract, and affects the precision of answering question. This thesis introduces manifold learning into question classification. Combining with multi-class characteristic of Chinese questions, we designed an algorithm for Chinese question classification based on Local Linear Embedding (LLE), and the experience showed that the precision of question classification was significantly improved.2) Meta search technology is applied to improve precision of question answering.Aiming at the shortages of lower searching efficiency and lower ability in information resource covering of single general search engine, meta search technology is applied to find the most appropriate search method for each question. Especially utilization of knowledge search engine and call of special question answering system are introduced for the first time in question answering system. It can improve semantic search ability and searching efficiency of the system, so the precision of question answering can be improved.3) Multi-agent technology is applied to improve overall performance of the system.There are many methods at each step of the Chinese question answering system. Every method has its characteristic and suitability. Because of the diversity of questions of open field question answering system (whether question class or related field), no single method is universal. This thesis introduces multi-agent technology into Chinese question answering system, put forward Chinese question answering system model based on multi-agent, in which each step uses agent technology, and each step has many agents of the same kind but different abilities. We transformed solution of question answering to solution of multi-agent coalition, described the agent coalition question of Chinese question answering system model, then we used ant colony algorithm, improved ant colony algorithm, genetic algorithm, genetic algorithm and ant colony algorithm to optimize the solution, thus able to improve the overall performance of the system.