A Full Information Enhanced Question Answering System Based on Hierarchical Heterogeneous Crowd Intelligence Knowledge Graph

With the development of deep learning technology, generative question answering models based on neural networks have gradually become a mainstream research direction in academia and industry. The current question answering models fail to make full use of the multi-level knowledge embedded in the learned corpus, and the interpretability and robustness of the models in the face of attack samples have certain shortcomings. From the perspective of information theory, this paper constructs the semantic, pragmatic and syntactic knowledge contained in the large amount of crowd intelligence corpora obtained from the Internet platform into a hierarchical and heterogeneous natural language knowledge graph. The graph-based full information enhanced question answering model (GFIQA) is proposed, and the hierarchical heterogeneous knowledge graph is incorporated in the model. Through the crowd intelligence knowledge interpretation module, knowledge-enhanced generation module and single-layer anisotropic decoder, the relevant knowledge in the crowd intelligence natural language knowledge graph is appropriately selected based on the attention mechanism, and the ability of question understanding and answer generation is improved. The experimental results show that the GFIQA model has a large improvement in PPL, BLEU, and ENC (PPL: −11.76, BLEU: +0.126, ENC: + 0.232) compared with the baseline model, and can generate fluent and smooth answers with reasonable grammatical modifications and rich semantics.