A Question Analysis Method On Military Question Answering

In military domain,a weapon knowledge-based question answering systems can provide information support for commanders in decision making.Question answering systems accept questions from commanders and return answers in natural language form by querying and inferencing the knowledge base,which is of great significance for assisting military decision making.Question analysis is an important part in weapon knowledge-based question answering.This thesis study question analysis on several common weapon related questions towards military decision making assistance.The major work in this thesis includes:(1)studying question representation model of common weapon related questions in military decision making assistance according to the features of common weapon related questions.Question representation model formally represents semantics of weapon questions,and is the target output and foundation of question analysis;(2)Proposing question analysis method composed of seven steps including question classification,entity and literal linking,middle question semantic graph generation,relation disambiguation,question semantic graph ranking and intent filling,based on question representation model.Studying core techniques in military knowledge-based question analysis method,including entity disambiguation method based on classification,and method on constructing relation pattern library to disambiguate relations.And conducting experiments to certify the effectiveness of the question analysis method,and the core techniques in this method.(3)Designing and implementing a question analysis subsystem on the foundation of the question analysis method.The subsystem provides question analysis interface for a question answering system.The major contributions in this thesis includes the following:(1)Proposing weapon question representation model which is composed of question intention and question semantic graph.(2)Proposing a question semantic graph based question analysis method with entity disambiguation method and relation disambiguation method as core techniques.