Proceedings of the 1st International Workshop on Cross Domain Knowledge Discovery in Web and Social Network Mining

Bo Long,Yi Chang,Hang Li
2012-01-01
Abstract:In the field of Web and social network mining, more and more learning tasks can easily acquire multiple data sets from various domains. For example, a modern search engine system often conducts ranking learning tasks in various domains with different languages (e.g., English text search, Spanish text search, etc.), or different verticals/topics (e.g., news search, product search, etc.); and recently recommendation systems start to leverage multiple types of user data from different domains, such as user browsing history data, user shopping record data, and user social network data. At the same time, the need for knowledge transfer is increasingly evident as many new datasets, or parts of data, are only very sparsely annotated. Different from traditional single-domain learning problems based on the assumption that training and test data are drawn from identical distribution, cross domain learning problems are built on multiple domain data that may have different degrees of relatedness to target tasks, offering an opportunity to help one another. To better leverage multiple domain data, mining and transferring of shared knowledge across multiple domains is likely to become a crucial step in Web and social network mining in the future.
What problem does this paper attempt to address?