A computer-implemented method is provided for identifying items in loosely-structured data. The method generally includes constructing a composite data definition for items which are to be identified and parsing input data using the data definition. The composite data definition is constructed by defining a layout for a composite data definition, where the layout indicates at least one of positional relationship of data items to each other and positional information for data items in the loosely-structured data; arranging data items in the layout, where each data item in the layout has a common meaning for applications that use the data item; creating an identification order list for the composite data definition, where the identification order list includes the data items in the layout and specifies an order in which the data items comprising the composite data definition are to be identified within the loosely-structured data.