A shape filling type reading understanding analysis model and method based on reinforcement learning

Abstract

The invention discloses a complete form filling type reading understanding analysis model and method based on reinforcement learning. The model comprises a coding layer, which is used for vectorizingwords of an original text, coding the words, taking a hidden vector of the last word of each sentence, outputting the hidden vector as a sentence vector, coding the text into a sequence of sentence vectors, and transmitting the sequence to a sentence extraction layer; a sentence extraction layer which is used for selecting sentence vectors, taking obtained sentences as current given text segmentsand encoding the current given text segments; a classification layer which takes each vacancy to be filled as a problem, takes the obtained text segment codes and the word vectors of the four candidate words as input, and calculates an output probability through a multi-feature classification network; a prediction layer which is used for normalizing the probability value obtained by the upper layer and the probability value of the language model to obtain the probabilities of the four final options; And an output layer which is used for calculating the cross entropy of the probability and theactual probability obtained by the previous layer, optimizing the classification network and updating the parameters of the network by taking the loss value as a delay reward.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a cloze-type reading comprehension analysis model and method based on reinforcement learning. Background technique [0002] The cloze-type reading comprehension task has been one of the important index tasks used to measure machine reading ability in recent years. Different from common question-and-answer machine tasks, cloze-type reading comprehension tasks cannot directly extract answers from given texts, but can only choose answers from alternative options. The model cannot be directly applied to this problem and achieve better results. The most popular cloze dataset right now is the CLOTH dataset. [0003] The technologies commonly used to solve such problems are mainly language models and attention mechanisms. The language model is trained by using a deep network on a huge corpus to mine grammar and other information in the text. In the cloze task, the probabilit...

AI Technical Summary

Problems solved by technology

[0011] In order to overcome the deficiencies in the above-mentioned prior art, the purpose of the present invention is to provide a cloze-type reading comprehension analysis model and method based on reinforcement learning, to improve prediction accuracy and efficiency, and to overcome the high computational complexity of the prior art. , problems with low computational efficiency and low accuracy on derivation problems

Images