The Research On Technology Of The Natural Language Transportable Interface Of The Database

With the rapid development and broad popularization of Database Application and Information Retrieval (IR) technology, more and more non-specialists need access data from all kinds of databases, and they want a kind of interface which could be used easily and conveniently. The Natural Language Interface of the Database (NLIDB) emerges as the times require. NLIDB is not only the hot in the artificial intelligence (AI) domain, but also an important branch of Question Answering (QA). The NLIDB means that it allows users to access the information in the Database by the way of some natural languages (such as Chinese and so on). It relates to many knowledge domains such as Database, natural language processing, AI, human-computer Interface and so on. Although NLIDB has made a great progress in the latest 30 years, the problems such as transportable application have still existed. So we need the more work of the NLIDB. Aimming at the point above, we go on researching on Natural Language Transportable Interface of Database (NLTIDB), which is transportable knowing from NLIDB.Named Entity Recognition (NER) is the key to QA. The Named Entity of airplanes will be needed when querying the NLTIDB.To recognize these entites, we use an improved method, which helps to enhance the convenience of querying the NLTIDB, to account the Mutual Information (MI).We construct both knowledge base and pattern base of a special domain. To make NLTIDB transportable, we design two bases: we build both the General Knowledge Base (GKB) and the Domain Knowledge Base (DKB) on the design of knowledge; on the design of the pattern base, we design a extending pattern structure which contains many kinds of information tagged, and have the pattern of the question, the pattern of Information Extraction, and some information of the query of the Database etc. saved in pattern base. The constructions both knowledge base and pattern base may decrease the degree of the domain's information implanted in program code sharply, and make advantage of the query and analysis in the NLTIDB system, and then help to fulfill the work of NLTIDB together.Commonly, the methods of the Information Retrieval contain Keyword Matching and Similarity Computing. We propose to use the both method of Hierarchal Maximum Entropy (HME) and Support Vector Machine (SVM) to realize Information Retrieval. Which are the important methods of the Machine Learning (ML). In this paper, we regard the Pattern Base as the resources of the Information Retrieval, and attempt to find out features of information through Lexical Analysis, Chunk Analysis, Syntactic Analysis and Semantic Analysis which will be trained, and match the query sentence with the pattern by the method of machine learning. The experiment shows this feasibility of this method we propose to.