Method for automatically generating submission demand abstract based on strategy gradient algorithm

Abstract

The invention discloses a method for automatically generating a submission demand abstract based on a strategy gradient algorithm. The method comprises the steps of extracting submission information and source code annotation in a submission demand relationship and a text semantic tree structure corresponding to the submission information and the source code annotation; encoding into a hidden state through a bidirectional recurrent neural network encoder; mapping into a vector sequence with a fixed length; and carrying out soft control between a word selected from a vocabulary table and a wordcopied from a source sequence through the generation probability of the vocabulary by using a pointer generator to obtain final vocabulary distribution. A strategy gradient algorithm with a base lineis combined with N times of Monte Carlo search; the method comprises the following steps: calculating an average reward containing a sequence of an action through N times of Monte Carlo search, finding out an action sequence with the maximum average reward, taking the action as an action to be selected, performing the action according to the action to obtain a complete sequence, updating a strategy gradient by utilizing a difference value between the sequence searched by Monte Carlo search and a baseline sequence, and finally generating a submission demand abstract.

Description

technical field [0001] The invention relates to the technical field of information data processing, in particular to a method for automatically generating a summary of submission requirements based on a strategy gradient algorithm. Background technique [0002] A commit-based development model, popular on modern collaborative coding platforms, eases the developer's contribution to the project. In this model, the developer does not need to directly access the central repository to contribute to the project. He only needs to Clone the central repository by yourself, make changes independently in the central repository of personal clones, such as fixing bugs or implementing functions, and then submit the changes to the central repository by submitting requirements, and then the reviewers of the central repository review the submitted requirements Review, and if the review passes, the developer has made a contribution to the project. So developers can easily contribute to the m...

AI Technical Summary

Problems solved by technology

[0003] The current method of generating a summary of submission requirements has some shortcomings. Only text information is used as input, and the generated submission requirements description lacks logic and does not meet human evaluation standards; only using the policy gradient algorithm in ordinary reinforcement learning can only be considered to generate The reward of the whole sentence, without considering the reward of the words generated at the intermediate moment, which makes the generated sequence far from the target sequence and affects the accuracy of the generated sentence

Images