Can Vector Space Bases Model Context?

Current Information Retrieval models do not directly incorporate context, which is instead managed by means of techniques juxtaposed to indexing or retrieval. In this paper, direct modeling of context is addressed by proposing the full recover of the notion of vector space base, whose role has been understated in the formulation and investigation of the Vector Space Model. A base of a vector space models document or query content descriptors. Moreover the semantics of content descriptors depends on context. Therefore a base of a vector space is the construct to model context. Also, change of context can be modeled by linear transformations from one base to another. Matrix algebra can thus by employed to process context, its evolution and application. In particular, Relevance Feedback is shown to be an example of context change.