Employee skills prediction using collaborative-filtering techniques

Abstract

During the last decades, the size of the IT companies grows constantly to reach several hundreds or thousands of employees. With a such large number of developers rises the issue of skills management. The different company managers should have a tool to analyze how many developers master a specific computer language or framework. With this tool, the managers would have the guarantee that their company has the resources to lead a specific project until the end. To achieve this, many IT companies have designed a system where each employee records his IT skills. Such system is difficult to manage for both the system administrator and the employees due to the large number of possible skills in computer sciences. This often leads to a database containing incomplete employees profiles and duplicated skills. In this master thesis, we will analyze how we could solve the incomplete profiles issue by implementing a recommender system. This engine will make periodically skills recommendations to the employees using collaborative filtering techniques. With this system, we replace the manual completion of the profiles by an automatic process where the employees just have to validate or not the skills proposed by the recommendation engine. This will improve the profiles level completion by proposing skills an employee could have missed. Another purpose of the system is to improve the consistency of the data by removing the manual encoding of the skills by the employees. In the first part of the thesis, we review the different recommender system techniques which are feasible for such application. Afterwards, we present the three collaborative filtering algorithms which have been implemented. In the second part of the document, we review how we designed the testing of our algorithms and what are the results obtained with these tests. We also present the results obtained when deploying the best algorithm in a real situation. In the last part of the thesis, we present how was designed the web application which uses our best solution to provide the skills predictions to the employees through a mailing system and a web interface.