Context-Centric Pricing
Turki Alelyani, Ke Yang Mao, Ye Yang · 2017
In software crowdsourcing, task price is one of the most important incentive to attract broad worker participation and contribution. Underestimating or overestimating a task's price may lead to task starvation or resource inefficiency. Nevertheless, few studies have addressed pricing support in software crowdsourcing. In this study, we propose Context-Centric Pricing approach to support software crowdsourcing pricing based on limited information available at early planning phase, i.e. textual task requirements. In the proposed approach, the global models include a list of 6 pricing factors and employ different natural language processing techniques for prediction; in addition, local models can be derived w.r.t. more relevant context, i.e. a set of similar tasks identified based on Topic modeling and we evaluate 7 predictive models. The proposed approach is evaluated on 450 software tasks extracted from TopCoder, the largest software crowdsourcing platform. The results show that: 1) the proposed models can be used at early crowdsourcing planning phase, when information on traditional metrics are not available; 2) the best model achieves 65% in accuracy measure; 3) the local model exhibits 27% increases superior to the global model. The proposed work can stimulate future research into crowdsourcing pricing estimation and inform ideas for crowdsourcing decision-makers.