Software reviewer mixed recommendation method based on deep learning and multi-Agent optimization

Abstract

The invention discloses a software reviewer mixed recommendation method based on deep learning and multi-Agent optimization. The software reviewer mixed recommendation method comprises the following steps: (1) mainly responsible for resource scheduling and allocation in a system; (2) performing low-order feature processing; (3) performing reviewer recall; and (4) sorting and recommending by reviewers. According to the method, an implicit factor matrix of an LFM learning project PR and an implicit factor matrix of a reviewer are adopted, the inner product of the two implicit factor matrixes is calculated to fill the vacancy value of a behavior characteristic matrix, the sparseness of the behavior characteristic matrix is reduced, and then a collaborative filtering recommendation algorithm based on the project PR is used for recommendation. In the sorting part, a deep neural network (DNN) is used for learning universal feature vectors of reviewers, and the DNN can learn high-order feature combinations, so that recommendation is more accurate. Collaborative learning among all parts is realized by designing a multi-Agent system, so that the recommendation efficiency in a mass data environment is improved.

Description

technical field [0001] The invention relates to the technical field of group intelligence software development, in particular to a mixed recommendation method for software reviewers based on deep learning and multi-Agent optimization. Background technique [0002] Traditional recommendation technologies mainly include collaborative filtering recommendation technology and content-based recommendation technology. These two technologies have been better applied in some scenarios. Collaborative filtering technology mainly recommends similar objects for the recommended subject based on the behavior similarity of similar objects. The core idea is similar to "groups of like gather people into groups". Collaborative filtering recommendation can be divided into user-based collaborative filtering recommendation and item-based collaborative filtering recommendation according to the calculation of object similarity. The user-based collaborative filtering recommendation calculates the s...

AI Technical Summary

Problems solved by technology

These recommendation methods do not consider the problems caused by massive data and high label feature dimensions, nor do they build deep models to mine the high-level feature information of developers and PRs. Therefore, there is still a lot of room for optimization in these recommendation methods

Images