MyPYTHIA: a recommendation portal for scientific software and services

this paper, we describe the MyPYTHIA Web recommendation portal that evolved from our earlier efforts in scientific recommendation systems [7,14]. Our design philosophy for scientific recommendation systems can be characterized as data-driven, i.e. a database of performance data on a benchmark set of problem instances is accumulated, and this database is mined to form the basis of a recommendation for future problem instances. A necessary assumption, thus, is that the performance database accumulated for mining is representative of the set of problem instances (and performance indicators) that will be encountered in the future. This assumption is valid in several problem domains where the Grid has enabled a recurring methodology of computation (e.g. solving linear algebra problems using component-based software from an OO library [15], solving partial differential equations using PSEs [16] and solving optimization problems using a service such as NEOS [17]). Hence, our focus is on application domains where there is substantial agreement over the methodology of computation and for which a large class of specialized algorithms and software have been developed. Later in the discussion, we address how MyPYTHIA's design and architecture can be extended to support other domains. Just as recommender systems in electronic commerce reduce information overload, recommender systems in the Grid setting provide targeted access to scientific softwar