Predictive Team Formation Analysis via Feature Representation Learning on Social Networks
| Autor: | Cheng-Te Li, Kun-Ta Chuang, Lo Pang Yun Ting |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Social network
Recall business.industry Computer science Flexibility (personality) 02 engineering and technology Machine learning computer.software_genre 020204 information systems 0202 electrical engineering electronic engineering information engineering Feature (machine learning) Graph (abstract data type) 020201 artificial intelligence & image processing Artificial intelligence business Feature learning computer |
| Zdroj: | Advances in Knowledge Discovery and Data Mining ISBN: 9783319930398 PAKDD (3) |
| DOI: | 10.1007/978-3-319-93040-4_62 |
| Popis: | Team formation is to find a group of experts covering required skills and well collaborating together. Existing studies suffer from two defects: cannot afford flexible designation of team members and do not consider whether the formed team is truly adopted in practice. In this paper, we propose the Predictive Team Formation (PTF) problem. PTF provides the flexibility of designated members and delivers the prediction-based formulation to compose the team. We propose two methods by learning the feature representations of experts based on node2vec [4]. One is Biased-n2v that models the topic bias of each expert in the social network. The other is Guided-n2v that refines the transition probabilities between skills and experts to guide the random walk in a heterogeneous graph of expert-expert, expert-skill, and skill-skill. Experiments conducted on DBLP and IMDb datasets exhibit that our methods can significantly outperform the state-of-the-art optimization-based approaches in terms of prediction recall. We also reveal that the designated members with tight social connections can lead to better performance. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|