Modern application software contains increasingly more customization features. Yet studies have shown that users often fail to adapt a system to their needs even if it makes their work more efficient. Two main barriers to customization are the difficulty in making adaptations and the lack of awareness that these adaptations exist. As a solution, systems that automatically recommend customizations when they appear to be useful have been proposed. In our work, we consider the problem of recommending software extensions – small components that provide additional functionality to an application. For applications like Emacs, Office, or Firefox, there exist thousands of such extensions on the web. However, classical recommender algorithms based on context or collaborative features often turn out to be insufficient, since the extensions are sometimes of highly varying quality. Existing systems that recommend software functions or customizations have also not succeeded for various other reasons. They are application-dependent, do not solve the privacy problem, and their mechanisms do not evolve in case the underlying application changes. We propose to improve recommendations by estimating the relevance and quality of an extension based on web link structure, extension authorship, and its dependencies with respect to other extensions. Our system is extensible and can be used with any application. We address the privacy issue by giving the user full control over the collected information. Finally, our feature computations evolve with the set of available extensions. ACM Classification H5.2 [Information interfaces and presentation]:

A scheme for domain and task representation in an interactive learning environment is described. The approach, based on recent enhancements to a classic AI planning formulation, enables suitable feedback to be incorporated economically into tutors for a range of procedural domains. The method is applied to the problem of simulating the operation of a video cassette recorder and representing tasks within this domain.