Winner prediction on crowdsourcing contest
Project Leader: Tingru Cui
Primary Contact: Tingru Cui (firstname.lastname@example.org)
Keywords: data mining; information systems; machine learning
Disciplines: Computing and Information Systems
In crowdsourcing, a firm uses an open call to outsource tasks to a ‘crowd’ or an independent network of people. Examples are Designhill (logo and graphic design), Innocentive (scientific problems), Zooppa (advertisement campaigns), as well as Dell’s IdeaStorm (R&D ideas and problems). With the proliferation of crowdsourcing contests, seekers engage solvers to develop ideas and solutions to solve innovation-related problems.
While the crowd is able to generate large amounts of ideas in such contests, the challenge shifts to subsequent idea selection. The resource demands to identify valuable ideas is high and remains challenging for firms that apply crowdsourcing initiatives. For example, an innovation contest at IBM required fifty senior executives and professionals to review, cluster, and identify valuable ideas from a pool of over 46,000 ideas down to a converged set of 31 ‘big ideas’ in a week-long process.
This project aims to automate the idea evaluation process and predict the winner of crowdsourcing contest with multimodal deep learning. It will design and evaluate advanced neural network schemes that combine information from different modalities to study the influence of sophisticated interactions among textual, visual, and metadata on winner prediction.
In this project, the students can practice data mining and machine learning methods on a dataset from a crowdsourcing site and develop innovative solution. The students can also practice problem solving, critical thinking and communication skills.
Expected skills: programming in Python; strong knowledge of statistics, machine learning and data mining; some knowledge of big data platform AWS/Azure.