| 英文摘要 | Many Web 2.0 networks surge up, leading to the tremendous online propagation of online data. On one hand, users suffer from information overload and it is non-trivial to locate the resources they are interested in. On the other hand, many users create and maintain multiple accounts on different Web 2.0 networks (such as Google+, Twitter, YouTube) where their daily behaviors are presented and shared. From the perspective of a single network, user data is usually very sparse. The notorious cold-start and sparsity issues have severely hindered accurate single network-based user modeling and practical personalized services. However, if we can integrate the user's overall behavior data across the networks he/she is engaged in, an intact user model which reflects his/her exact preference is very likely to be set up. Therefore, we are devoted to addressing the above issues by exploiting cross-network user modeling, which includes two key problems: 1. User accounts linkage across social media networks is implicit. Most of the user accounts in different networks are not the same, and there are no explicit correlation among them, i.e., it is not known which user account in one network and which user account in another network correspond to the same individual in the physical world. How to correlate the cross-network user accounts comes first for cross-network user modeling. 2. User data across different social media networks are heterogenous and redundant. The heterogeneity is beyond modalities, such as text, image, video, and audio. Actually, the heterogeneity is obvious even within the same modality, such as upload, favor, share, and comment. Besides, these information may be redundant and even contradictory. How to aggregate these information is the key problem in cross-network user modeling. To cope with the above two issues, we have conducted the following researches on cross-network user identification and cross-network user modeling: 1. We present a cross-network user identification method based on user behavior pattern matching. User behaviors distribute on different networks and there are often certain temporal patterns among them, which are decided by the network attributes. For example, user behavior in social textual stream network platform is faster than multimedia sharing network platform. Based on this, we match user temporal behaviors across networks and combine the username similarity between networks, which effectively improve the accuracy ... |
修改评论