Location-Dependent Task Allocation for Collaborative Mobile Users with Social Awareness

It has been found in many areas that crowd intelligence can be exploited to effectively handle complex tasks. For instance, sensing tasks can be allocated to a group of mobile users (known as workers) to complete them efficiently. A key to success is to match tasks with workers properly so that various constraints are satisfied while a mediator for the matching can also earn a profit as an incentive for their effort. This task allocation problem has been studied in the literature from different perspectives. One aspect that is less addressed is the collaboration efficiency when a group of workers need to work together to fulfill the requirements of a task. In this paper, we attempt to solve a collaborative task allocation problem, which takes into account social connections among workers and their impact on collaboration efficiency and achievable profits. As this problem is proved to be NP-hard, we formulate a temporal heterogeneous graph and develop a deep reinforcement learning method based on an expressive neural network model for the graph. By decomposing the heterogeneous graph into smaller and simpler subgraphs, we try to reduce the network dimensionality while extracting essential features. Our experiments also show that the proposed method offers competitive advantages over other heuristic and meta-heuristic algorithms.