A Multi-task Multi-kernel Transfer Learning Method for Customer Response Modeling in Social Media

Customer response modeling is essential for a firm to allocate the marketing resources to active customers who have potential values. With the development of social media, customer response modeling in social media plays important roles in the firms’ marketing decisions. For customer response modeling in social media, the inputs involve multiple types of data and the purposes are to identify respondents to multiple items. In this study, a multi-task multi-kernel transfer learning (MT-MKTL) method is proposed to integrate shared, task-specific and transferred features in a framework for customer response modeling in social media. A two-phase algorithm is applied to solving the MT-MKTL problem. A computational experiment is conducted on microblog data. The experimental results show that the MT-MKTL method exhibits good performance.