Paths and Contextually Specific Recommendations

We describe the Recer system and its underlying model of user activity, paths. A path is a timestamped log of activity in one of more tools, such as web browsers and text editors, gained by client–side tracking. By default, recent user activity is taken as representing the current context, and as an implicit query to the database of paths. The result is a set of recommendations specific to that context, and also reflecting a group’s patterns of past activity with the logged tools. We give some examples of its use before going on to discuss how recommender systems’ ability to handle heterogeneous data types allows extension of the model of user context to include locations and artefacts in both the physical world and virtual worlds. Our aim here is to have a much more realistic and broad notion of context and activity implemented in our systems, moving beyond an implicit assumption that all that is significant about a person’s activity is on their desktop. We outline our current work extending Recer to track multiple devices and media so that recommendations refer to people, places and things—both physical and digital. Recer will become part of a larger system aiming to support awareness and collaboration amongst people exploring a collection of documents about Glasgow, the physical city and a VR model of the city centre.