(SOURCE SELECTION [3]) Let S be a set of sources, F be an integration model, GF () be a gain function and CF () a cost function using model F, and c be a budget on cost. The Source Selection problem finds a subset SI S that maximizes GF (SI ) CF (SI ) under constraint CF (SI ) c. . Dong et al. [3] show how GF (SI ) can be defined in the context of data fusion as the accuracy of the integration result, and use an additive cost function over the costs of the selected sources.

DEFINITION 2. A data source S is dynamic when it is updating its content by capturing entity appearances, disappearances and value changes from a data domain with a frequency fS. We assume knowledge of the data evolution for a past time window T ending at time t0. We consider a fixed set of future time points, denoted by Tf , and we wish to maximize the profit of integration for and Tf . Let GF (SI , Tf ) be the overall gain of integrating SI using model F for Tf , and CF (SI , Tf ) be the corresponding integration cost. With GF (SI , t) denoting the gain of integrating SI for a single time point t Tf , and AtTf denoting an aggregate function (e.g., average or max) over the time points in Tf we define the overall gain as GF (SI , Tf ) = AtTf GF (SI , t). Similarly to Dong et al. [3] we assume an additive cost model with CF (SI , Tf ) = P SSI C(S, Tf ), where C(S, Tf ) denotes the cost of source S SI for Tf .

REWRITE ALL THE MATHEMATICAL EXPRESSION above USING EQUATION EDITOR in word file