A model method based on paragraph internal reasoning and joint question answer matching

Abstract

The invention discloses a reading understanding model method based on paragraph internal reasoning and joint question answer matching, and the method comprises the following steps: S1, constructing avector for each candidate answer, the vector representing the interaction of a paragraph with a question and an answer, and then enabling the vectors of all candidate answers to be used for selectinganswers; S2, carrying out experiment. According to the model provided by the invention, paragraphs are firstly segmented into blocks under multiple granularities; an encoder is used for summing the intra-block word embedding vectors by utilizing neural word bag expression; then, a relationship between blocks with different granularities where each word is located through a two-layer forward neuralnetwork is modeled to construct a gating function, so that the model has greater context information and captures paragraph internal reasoning at the same time. Compared with a baseline neural network model such as a Stanford AR model and a GA Reader, the accuracy of the model is improved by 9-10%. Compared with a recent model SurfaceLR, the accurcay is at least improved by 3% and is about 1% higher than that of a single model of the TriAN, and in addition, the model effect can also be improved through pre-training on an RACE data set.

Description

technical field [0001] The invention belongs to the technical field of reading comprehension, and in particular relates to a model method based on paragraph internal reasoning and joint question answer matching. Background technique [0002] Machine Reading for Question Answering (MRQA) research has received widespread attention in recent years. How to enable machines to read and understand natural language is the main problem in the field of cognitive intelligence. Reading comprehension tasks need to be integrated and reasoned throughout the document Information about events, entities and their relationships, question answering is often used to assess reading comprehension. The main types of machine reading comprehension tasks are cloze, choice, and question-and-answer. Deep learning has been widely used in natural language processing in recent years, as well as in the field of machine reading comprehension. Compared with traditional feature-based methods, in neural In the...

AI Technical Summary

Problems solved by technology

[0003] Current reading comprehension methods are usually based on pairwise sequence matching, paragraph-to-sequence matching connecting questions and candidate answers, or paragraph-to-question matching, however, these methods may not be suitable for selective reading comprehension where questions and answers are often equally important, based solely on Questions matching paragraphs may not make sense and may lead to loss of information in paragraphs, such as "Which statement of the following is true?" Questions, on the other hand, concatenating question and answer into a single sequence for matching may lose both question and Interaction information between answers, for example, sometimes answers require questions for anaphora resolution, etc. In addition, most models usually employ recurrent neural networks as encoders, which sequentially parse text sequences word by word, although helpful in capturing linguistic Lexical and grammatical structure, but paragraphs tend to be long, which limits multi-sentence reasoning within paragraphs

[0004] For current machine reading comprehension methods, only matching the question to the paragraph will result in the loss of information in the paragraph or concatenating the question and answer into a single sequence. Matching the paragraph will lose the interaction between the question and the answer, and the traditional recurrent network sequentially parses the text Thus ignoring the problem of intra-paragraph reasoning, a model method based on intra-paragraph reasoning and joint question-answer matching is proposed

Images