Study Of Knowledge Graph Question Answering Systems Based On Generative Adversarial Learning

With the wide adoption of the Model Internet and the coming era of Artificial Intelligence,the demand of rapid capture of precise answers is becoming increasingly higher,making Question Answering system a new generation of approach for people to acquire information.At the same time,the tremendous developments of Knowledge Graphs have a massive boost to the research as well as the implementation of Question Answering systems.As a consequence,research on knowledge-based Question Answering systems has received unprecedented attention.In this paper,we focused on knowledge-based Question Answering systems and conducted our research,including:(1)We designed a question generation model for knowledge graphs based on the famous encoder-decoder architecture,given a triple fact in knowledge graph,the generation model can automatically generate corresponding questions in natural languages.On the testing set,our model achieves 40.2 and 37.38 in BLEU and METEOR respectively.(2)We proposed a neural-network-based model for knowledge graph question answering matching task,combining character-level and word-level semantic information as well as the structure information of the whole knowledge graph to find a best-matching fact for the given question in knowledge graphs and achieves 72.4%accuracy.(3)We designed a semi-supervised architecture for knowledge graph question answering systems based on Generative Adversarial Networks,which utilizes the internal relationships between question generation and question answering matching tasks and conducts joint adversarial learning.The performance of generative model and matching model both get improved.