Paper Type

full

Description

With the widespread adoption of information systems, a wide variety of digital data traces of users are exposed. However, as users use multiple information systems for different tasks that are not directly linked with each other, the data will be fragmented or disconnected. Interconnecting these data traces of a user from various information systems is an important research problem known as “User Entity Resolution” (UER). Much of the current UER methods depend on the similarity of user profile attributes and network data. With the decrease in the reliability and availability of user profile attributes and network data, it is important to develop methods that incorporate user activity data to improve UER. In this study, we propose a novel approach to incorporate user activity data in UER. A deep neural network model is proposed as a cross domain transfer model. Given the user activity data from a primary domain, the trained transfer model can be used to generate an estimate for a secondary domain of the user. The estimate could be used to search for the matching user in the secondary domain. Our evaluation results indicate that the proposed model has a far superior ability to rank users.

Share

COinS
 

Application of Deep User Activity Transfer Models for Cross Domain User Matching

With the widespread adoption of information systems, a wide variety of digital data traces of users are exposed. However, as users use multiple information systems for different tasks that are not directly linked with each other, the data will be fragmented or disconnected. Interconnecting these data traces of a user from various information systems is an important research problem known as “User Entity Resolution” (UER). Much of the current UER methods depend on the similarity of user profile attributes and network data. With the decrease in the reliability and availability of user profile attributes and network data, it is important to develop methods that incorporate user activity data to improve UER. In this study, we propose a novel approach to incorporate user activity data in UER. A deep neural network model is proposed as a cross domain transfer model. Given the user activity data from a primary domain, the trained transfer model can be used to generate an estimate for a secondary domain of the user. The estimate could be used to search for the matching user in the secondary domain. Our evaluation results indicate that the proposed model has a far superior ability to rank users.