Descripció del projecte
Interactive Machine Learning in online professional networks
One of the main technological challenges in our society is how to make sense of the unprecedented volumes of data that are continuously generated in order to guide and support decision making. An important limitation of the current approaches to deal with is the lack of a learning process that can (1) obtain and evaluate feedback of the system’s performance from the environment and (2) adapt accordingly to improve the future performance. This is the basis Reinforcement learning (RL), the area of machine learning concerned with learning to control a system (or agent) that interacts with an environment so as to maximize a performance measure that expresses a long-term objective. What distinguishes reinforcement learning from other machine learning paradigms is that only partial feedback is given to the learner about the learner’s predictions. Further, the predictions may have long term effects through influencing the future state of the controlled system. Thus, time plays a special role.
The core objective of Fluttr is to change the current recruitment market. While the job market is rapidly changing mainly due to technology advancements, the recruitment market has not followed suit and the CV (created in the XV century) still represents the first fundamental milestone in the hiring process. Fluttr generates and uses large amounts of complex data from different sources that is used to build matches to create professional opportunities. Examples of matches are between members of a same community or different ones, between members and available jobs, provide pro-active recommendations for local events, training courses and products (for example books) that might be of interest to each member.
The present project proposal addresses the challenges faced by the recommendation engine of Fluttr. The first objective is to collect and study behavioral patterns and user profiles to extract knowledge and create a first predictive model, which is going to form the basis of the learning systems developed in the following objectives:
Integrating recommended matches (actions) in the predictive model and analyzing how these action spaces with different structures (e.g., actions organised in a graph) and different levels of interaction between actions (e.g., similarity between actions) may influence the learning performance.
Developing efficient algorithms that work in non-stationary environments when not only the performance of each action evolves, but also the structure of the action space can change over time.
The candidate is expected to understand contemporary topics and issues in Data Science. For example, multivariate statistics, machine learning, data mining and algorithmic complexity.
