A method of intelligent question answering in virtual learning environment based on collaborative attention

Abstract

The invention relates to an intelligent question answering method for a virtual learning environment based on a stacked Bi-LSTM network and collaborative attention, and belongs to the technical fields of natural language processing and virtual reality. The method includes: S1: building a knowledge base and data preprocessing; S2: feature extraction and vector representation: configuring the co-attention mechanism and the attention mechanism in the model of building a stacked Bi-LSTM network; S3: considering the question-and-answer pair space vector Position and direction, harmonic cosine similarity and Euclidean distance to calculate the matching degree between question and answer pair vectors; S4: Build a virtual learning environment based on the Unity3D platform and introduce a network model to realize intelligent question and answer in a virtual classroom in a specific domain. The invention combines a deep network model with a multi-type attention mechanism to realize a deeper interactive understanding and representation of questions and answers, and meanwhile, the introduction of a specific domain virtual classroom broadens the application range of intelligent questions and answers.

Description

technical field [0001] The invention belongs to the technical field of natural language processing and virtual reality, and relates to an intelligent question answering method for a virtual learning environment based on a stacked Bi-LSTM network and collaborative attention. Background technique [0002] In recent years, deep learning has played a pivotal role in natural language processing (NLP). All tasks of NLP, such as information retrieval, intelligent question answering, machine translation, dialogue systems and voice manipulation, etc., can be attributed to the understanding and understanding of natural language. application. Compared with traditional methods, deep learning can automatically learn features based on raw data, automatically extract the relationship between words from a large number of samples, and it can combine the structural information in phrase matching and the hierarchical characteristics of text matching, through learning The process extracts effe...

AI Technical Summary

Problems solved by technology

However, these models currently have corresponding shortcomings, which are mainly reflected in the following aspects: First, the existing models based on deep neural networks lack the consideration of the interaction and influence between questions and answers, especially the effect of answer sentences on The impact of feature extraction and generation of question sentences; the second is that most models focus on feature extraction and representation, ignoring the final question and answer pair vector matching calculation is also a key step to improve the accuracy of the model; the third is the existing Most of the network models in the world use open-domain question-answering data sets for training and learning, lack of construction and use of data sets for specific fields, and have little practical application value

Images