Polls, as the most common method of eliciting crowd insights are getting more and more popular. From predicting election results, to customer satisfaction, to scientific surveys, polls play a crucial rule in revealing the true sentiment of a community. However, when participation in a poll is subject to having some specific expertise and skills, finding sufficient number of participants for a given poll is a serious challenge. In this paper, we propose a new method of polling crowd contextual insight which is based on decomposing a poll into some sub-polls and recruiting participants to answer the given questions. We also take into account the probability of engagement of a participant in a poll to make sure that we recruit sufficient number of suitable workers. The proposed method is implemented and tested using the simulated data, build based on a public data dump from Stack overflow. The evaluation results show the superiority of our proposed method over the other related work.