Um sistema de recomendação para usuários das plataformas de crowdsourcing

Abstract

The success in software development on crowdsourcing platforms depends on many developers who are involved in registering and submitting tasks every time. One of the main challenges faced by users on this type of platform is the difficulty in choose tasks, due to the large number available and finding tasks according to their profile. This work presents a recommendation system with the objective of recommending tasks in real-time based on the user's last activities in these platforms that use TopCoder as a study base. For that, a task similarity analysis model (SIM-Crowd), a user history evaluation model (UHR-Crowd), and an algorithm responsible for grouping the models and generating the analyzes (RA-Crowd) were developed. To evaluate the recommendation system, experiments were carried out to measure the accuracy and coverage and the metrics propose by the models to improve the initial configuration of the system. The objectives were achieved because it is possible to generate recommendations with a good precision and coverage rate using the proposed models as a means. Although the experiments are carried out with historical data extracted from the platform, it is estimated that by detailing the process, the possibility of generating recommendations in real time is demonstrated.