Chinese Question Answering System Based On Distributed Representation

Question answering system is a kind of special type and form of information retrieval. Given a collection of documents, a question answering system is committed to find the answers to questions raised in natural language. Q&A is a multi-disciplinary subject, it includes information technology, artificial intelligence, Natural Language Processing, knowledge and data management and cognitive science. From a technical point of view, the question answering system uses natural or statistical language processing, information retrieval technology, knowledge representation and reasoning technology as potential building blocks. It includes text categorization, information extraction and summary techniques.Generally speaking, there are three parts of the question answering system: question classification, information extraction, answer extraction, these components play an indispensable role in the question answering system. Question classification plays the role of the primary role in question answering system, according to the problem of the type of entity classification problem. Information retrieval technology through their intelligent question answering system extract the answers that can be applied to obtain a success of the recognition. In the end, the answer extraction module presents topics, which are usually required to sort and validate candidate answers.The content of this paper is as follows:(1) This article studys the distributed representation of semantic units of different sizes, from words, phrases, sentences to paragraphs and chapters. We introduce the concept of distributed semantic representation, compare different types of distributed representation methods, and compare the advantages and disadvantages of different methods. The existing semantic distributed representation learning mainly focuses on the representation of words and phrases, and it is still in the initial stage of the sentence and discourse.(2) This article studys the question answering system based on knowledge base.The semantic knowledge base is an important part of the question answering system,At present, most of the semantic knowledge acquisition comes from structured data, and more concentrated in English, the acquisition of Chinese semantic knowledge still needs further exploration.Moreover, the knowledge base of each language is independent of each other, and how to make the fusion of multi language knowledge base needs further research. At the same time, we also construct the knowledge base of Chinese knowledge about geography, which is needed by the system and composed of a large number of three tuples.(3) This article introduces the semantic distributed representation to the question answering system based on knowledge base,By setting up the neural network structure, the question sentence and the candidate answer are transformed into the vector of the distributed semantic representation,Training through the gradient descent algorithm makes the problem and the correct answer in the semantic space similarity. Further work is devoted to the use of more complex structures and learning ability of the neural network to improve the performance of the system.The experiments show that using the distributed semantic expression binding to the question answering system based on knowledge,with questions and candidate answers by neural network mapping to the same vector space to calculate semantic similarity, achieved good results.