In this HDR dissertation, I focus on my research works addressing some of the challenges related to service-oriented dataintegration. Specifically, I focus on the aspects of data privacy, data uncertainty and ranking in data service composition. Data service compositions (a.k.a., Data mashups) are situational applications that combine data elements from different data sources to provide value-added in-formation to users and solve immediate business data needs. Typically, the access to data sources is carried out through data services. Data service composition involves important challenges that need to be resolved. First, data services could be employed to access confidential and privacy-sensitive data, such as the medical record in the healthcare domain. In our works, we proposed new models and techniques to preserve the confidentiality and the privacy of data when services are executed in isolation, but also when they are composed with each other to resolve complex queries. With our models, service providers can enforce their privacy and security policies without being constrained to change the implementation of their services, i.e., data services are regarded as black boxes. Our solutions prevent data leakage among services when they are composed to answer complex queries. Second, several data web services with similar semantics (or also similar func-tionalities) may exist, and the same data service may return several data items that match, to a certain extent, the user query, we proposed a host of service ranking algorithms to select and rank data services, as well as their outputs (when they are executed), based on how well they match user’s query and preferences. Our algorithms can be used when data services are executed in isolation but also composed with each other to answer complex queries. Finally, we addressed the data uncertainty aspects that may be associated with data services. Uncertainty in data services can happen at two levels: semantics and data. The former is when the semantics of the service is uncertain (i.e. the service can have multiple possible interpretations and thus semantics). The latter is when the data returned by the service is uncertain. We proposed models, techniques and algorithms to take into account those uncertainty aspects when uncertain data services are selected, composed and executed.