In the last two decades, user profiles have been used in several areas of information technology. We can mention, for example, its use in recommendation systems [BOHG13, ZYST19], in adaptable user interfaces [Bro16], in personalized systems [LMNS19], in cognitive or physical rehabilitation systems [DPQZ18]; etc.
In general, the systems store personal preference profiles as a set of items, which, in the vast majority, are represented by numeric attributes or linguistic labels (e.g. "Very low / Low / Medium / High / Very high", "single / married / separated / widowed”, etc.).
In the literature, the vast majority of research works and systems focus on the creation of profiles (using Data Mining techniques based on user navigation history) and their dynamics are made by means of systematic recreation of the profiles, without using the last used profile. These creations and dynamics are made ad-hoc for each particular implementation, without (as far as we know) a formal study on the creation of profiles and their dynamics. On the other hand, when using Data-Driven techniques, systems are unable to produce explanations neither on the creation of the profile nor on its dynamics.
In this project we propose to formalize the creation, representation and dynamics of profiles in a Knowledge-Driven perspective. In particular, we aim to:
(1) create a formal profile representation structure, based on the definition of a formal language that allows to:
(a) clearly represent a user profile and its attributes.
(b) describe properties and operations between the different items in a profile;
(2) Define three types of change operations for profiles:
(a) updating: it is the most common case, it happens as a consequence of the evolution of the individual or of his interaction with the system;
(b) identification: given a change in a certain profile or set of profiles, this operation identifies the cause or causes that motivated that change;
(c) targeted training: Given a profile and a goal (a particular new profile or a condition to be satisfied by a profile), this operation determines what are the processes (e.g. training) that must be carried out in order to achieve the goal.
The proposed operators will be based on the theory of belief revision (for an overview of the model and applications see [FH18]). Belief revision systems are logical frameworks for modelling the way how agents modify their beliefs when they receive new information (sometimes inconsistent with previous beliefs). In order to integrate the new information, the agent may need to reject some previous information, but he should preserve as much as possible of the original information. The AGM model [AGM85] is the standard model for formalizing knowledge dynamics. The adjustment of the belief revision model will allow to create the dynamics of computational profiles in order for it to be possible to predict the evolution of a profile based on partial, incomplete or inconsistent information.
(3) Test the language and models proposed on two platforms that use profiles, both with numerical items. The first one is a cognitive training platform (BRaNT) for stroke patients, the second one is a home banking platform with an adaptive interface. For both platforms, it is envisaged to enrich the profile representation language, and to characterize the operations described in item (2). Finally, a comparison will be made between the original models and the proposed models.
(4) Extend the proposed models to non-prioritized profile change models, i.e., models where new information is not always accepted and multiple change models, i.e., models where the input is not simply a single piece of information, but a set of new simultaneous pieces of information.