Research On Temporal Recommendation Algorithms For Crowdsourced Testing Task Allocation

Crowdsourced Testing is a kind of production organization that assigns the subtasks of software testing to non-specific groups through the Internet.The purpose of applying the recommendation algorithm in the task allocation process of the crowdsourced testing platform is to promote the fast and high-quality completion of the crowdsourced testing task.The automatic recommendation algorithm on the crowdsourced testing platform often uses the user behavior records to discover their ability,preference and other characteristics,and then recommends these appropriate testing tasks.This is conducive to solve the problem that testers face too many tasks to be tested,that is,information overload.It is also conducive for the testing task getting the attention of professional testers.The temporal characteristics of user behavior in crowdsourced testing scenarios are concerned by recommendation algorithms,but the temporal characteristics of tasks are not paid enough attention.And there are different types of crowdsourced testing tasks,such as competitive tasks and collaborative tasks.Different task types will make user behavior records show different characteristics.In addition,the interval of user behavior in previous models is relatively simple.In this paper,we focus on how to capture the evolution of user interest better by using time sequence information,enhance the influence of testing task in the recommendation system,and the challenge of multi domain interest brought by the diversity of testing tasks to the recommendation algorithm.The main work of this paper is as follows:(1)We experimentally verify the effectiveness and necessity of temporal correlation and temporal modeling for crowdsourced testing task recommendation algorithm.Considering the unique crowdsourced testing application scenarios,temporal correlation and special temporal modeling are focused in this paper.Comparative experiments show that the recommendation algorithm focusing on temporal correlation and temporal modeling can achieve better results on crowdsourced testing data set;(2)In order to solve the problem of paying attention to the temporal correlation of users while ignoring the temporal correlation of tasks in the crowdsourced testing recommendation algorithm,a crowdsourced testing recommendation algorithm based on temporal correlation is proposed based on temporal modeling and neural collaborative filtering.In this method,the consumption network is used to calculate the temporal correlation of testers and tasks respectively,and the two are combined to construct double temporal correlation.Then,the feature matrix of the time slice to be predicted is constructed.Finally,the nonlinear interaction between testers and tasks is modeled by neural collaborative filtering.A variety of comparative experiments show that the method is effective and achieves better results on the crowdsourced testing data set;(3)In order to solve the problem of user’s multi-domain preference caused by the diversity of testing tasks and the lack of explicit temporal modeling in crowdsourced testing recommendation algorithm,a LSTM crowdsourced testing temporal recommendation algorithm based on attention mechanism is proposed.In this method,attention mechanism is used to focus on the distributed multi-domain differential preferences in the historical data of testers,and some explicit time interval modeling methods are proposed,and then enough temporal information is input into LSTM network to capture the dynamic preferences of testers.A variety of comparative experiments show the effectiveness of this method,and also verify the effectiveness of explicit temporal modeling methods.