Using user behaviour similarity for recommendation computation: the broadway approach

Recommendation systems have gained a lot of attention recently especially in the field of on-line information retrieval (IR) systems (e.g. searching documents on the web). Recommenders are traditionally partitioned into two main families: ones (Resnick et al 1997). Systems of the first type recommend items or actions to the user depending on an evaluation of the user own past actions, while those belonging to the second family recommend to a user items positively evaluated by other  " ‚ƒ q ‚„… I † users. In this work we describe a new recommendation approach, called the Broadway approach, where the system recommends to a user what have to that user. Other systems in IR field use user behavior similarity as a basis for recommendation computation (Yan et al 1996). However, the Broadway approach has the particularity to model user behaviors by observation variables rather than matching user actions to a pre-specified behavior model. Following the Broadway approach, the user interactions with the application are saved in a log-like file. This log file contains a set of ¢£ ‘ ¤ q ¥ u ¦ " ¥ R § " £¥¨¦ , each holds the evolution with time of a variable that is said to be relevant to describe the user behavior. Obviously the choice of these variables depends on the application field (cf. Section 2). Time series are grouped into that correspond to a user session with the system within a well specified period of time which has a well defined semantic in the application. Now CBR methodology is used (cf. Case-Based reasoning (Kolodner, 1993)): it is a problem solving methodology where, in order to find the solution to a current problem, one looks for a similar problem in an experience base, takes