There has been a lot of research in the area of constraint programming [12,11,10], but not much attention has been paid to the interplay between information gathering, constraint propagation, and user interaction, which is the focus of Heracles. Bressan and Goh [1] have applied constraint reasoning technology to information integration. However, the specific problem they are addressing is quite different from ours. They use constraint logic programming to find relevant sources and construct an evaluation plan for a user query. In our system the relevant sources have already been identified. We focus on user interactivity, flexible integration of information gathering and other computational constraints in an uniform framework, and on information propagation in service of the user tasks. Petrie et al. [9] developed an architecture for constraint-based agents in engineering design applications. We share the idea of combining multiple reasoning systems. However, their focus is on distributed constraint management issues, such as adding and removing constraints, and in recording the rationale for the design choices and inconsistencies. Real-time information gathering is not considered in their system.
The growth and changes to the constraint network that occur in Heracles as a result of the hierarchical expansion of templates can be seen as a form of dynamic constraint satisfaction [6]. In dynamic constraint satisfaction the variables and constraints present in the network are allowed to change with time. Heracles imposes a structure to these changes as they correspond to meaningful units in the application: the templates.
Lamma et al. [4] propose a framework for interactive constraint satisfaction problems (ICSP) in which the acquisition of values for each variable is interleaved with the enforcement of constraints. The interactive behavior of our constraint reasoner also can be seen as a form of ICSP. However, our approach includes a notion of hierarchical decomposition and task orientation. Their application domain is on visual object recognition, while our focus is on information integration