SAPI: SWARM-OPTIMIZATION-BASED APPROACH FOR PERSONALIZATION OF WEB-BASED USER INTERFACES

The growing amount of information overload on Internet makes it very difficult to find and select pieces of information needed by users. Personalized user interfaces improve the ability of web-based systems to interact with a user by constructing a user model based on partial information and experience with that user. They can present solution of information overload problem. Machine learning supports the acquisition of user models from memorization of user's interaction, which is a difficult problem. Usually, the information about the user is limited and it is not easy to infer assumptions about the user. In this paper, a leaning approach for user interface personalization (SAPI) is proposed. It is based on a swarm intelligence model – the particle swarm optimization (PSO) algorithm. In human-computer interaction, user interface events and frequencies can be recorded and organized into episodes. By computing episode frequencies and implication relations automatically can be derived application-specific episode associations and therefore enabled adaptation of interaction structure to a specific user. When SAPI applied to sequences of user actions would allow a user interface to adapt over time to an individual’s pattern of use. During the first step of SAPI, the initial interaction structure of the web-based system to be adapted is determined as a tree hierarchy. Then, the transition matrix for relevant training patterns is established. For this purpose are used dialogue protocols of particular work sessions with the web-based system from the beginning to completing the selected tasks. To each training pattern corresponds a transition matrix describing the transitions between the interaction points. On the basis of the transition matrices, a neural network is determined. The network weights are trained by the PSO algorithm, and the weights are used for construction of the optimal interaction structure with minimum weighted path length. SAPI is illustrated by an example with the Electronic Research Funding Information server ELFI (http://www.elfi.info/elfi). The initial and adapted interaction trees for German national research funders are shown. SAPI can automatically assist users in their tasks by detecting and analyzing users’ behavior patterns. It can, to a certain extent, increase user productivity. The advantages of the approach proposed consist in: 1) training of web-interaction structure weights based on a PSO model; 2) creation of optimal interaction structure adapted to user allowing significant quicker access to interaction points; 3) capability to build implicit user model, etc. Possible applications of SAPI in adaptive web-based systems, especially in personalized eCommerce services and benefits are discussed