We developed a context-aware mobile crowdsourcing
framework for recommending personalized packages of location-dependent tasks to workers (task-
doers). Our work is motivated by two major challenges facing existing commercial mobile
crowdsourcing platforms. First, tasks can take a long time to complete as they are dependent on
workers’ mobility. Second, location-dependent tasks are expensive, mainly because they may require
workers to travel. We recommend tasks to workers based on their current and predicted future
contexts, as well as their personal preferences toward different types of tasks. The key ideas are (i)
finding the right person at the right time for a particular task and (ii) bundling tasks to reduce the cost
of travel per task.

We formulate the task recommendation problem for two scenarios: a fixed-price
scenario where the objective is to maximize the expected number of completed tasks, and a dynamic-
pricing scenario where the objective is to maximize the systems utility, which is a function of both task
completion rate and priority. A logistic-regression based approach is used to model workers behavior
based on their statistical information. We show that the objective function in each scenario is
monotonic and submodular. Thus, we are able to exploit computationally feasible algorithms with tight
optimality bounds to solve the optimization problems. Our approach and solutions have been supported
by a successful real-world case-study evaluation,
"Paper-1".