Reinforcement Learning for Expert Finding from Web Search Results

Finding experts is a crucial problem for developing countries, since these highly qualified expatriates might be able to contribute to the local development. Reaching members of the highly qualified diaspora is thus a major challenge for policy makers. This paper presents a Deep Reinforcement Learning method for tracing expert’s mobility trajectories from web search engine results. Our method queries search engines to identify and extract relevant information, such as the affiliation institution and the affiliation years of each expert. The goal of this work is to implement an intelligent expert finding agent capable of assisting this search by generating and observing as few automatic queries as possible. We are using as an agent a Deep-Q Network with two architectures based on neural networks to approximate the value of the Q-value function. The source code is available here: https://github.com/rcln/unoporunoDQN .