Learning Mappings between Models of Data
msra
摘要
Information Integration systems today deal with data originating at multiple autonomous and het- erogeneous data sources. This heterogeneity means that applications simultaneously manipulate data available in multiple models or schemas. Such interactions are enabled by the presence of semantic map- pings between the dieren t models. Mappings make it possible to reason about data and answer queries across dieren t data models. However, determining these semantic mappings is a dicult task because of a variety of reasons - models use dieren t vocabularies, dieren t design methodologies and might have dieren t design goals. Further the large sizes of models, the need to incorporate multiple and varied evidences and the ubiquitous nature of applications that require mappings, necessitate the building of tools for automatic or semi-automatic (involving some user interaction) generation of mappings. In this proposal I look at issues in learning mappings between models - specically , building tools that learn mappings by incorporating multiple types of evidences. I discuss the various types of mappings that are needed for applications and the requirements they place on the representation of the mappings that we are trying to learn. I present a proposal for research investigating particular aspects of mappings.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络