Silvia Suriani

Modeling Online User Behavior

A framework for online user behavior soft modeling is presented in this work. Behavior models of users in dynamic virtual environments has been described in the literature in terms of timed transition automata which can be compiled in a planning domain. The extended notion of soft timed transition automata is proposed in order to recognize a larger class of user histories. The notion of deviation from the user model allows to assess and evaluate in real time the dynamic behavior of users acting in virtual environments, such as e-learning and e-business platforms. The timed automata model allows to describe virtually infinite sequences of user actions subject to temporal constraints, while soft measures allows to assess recognition of behaviors by evaluating the amount of temporal deviation, additional or omitted actions contained in an observed behavior. The proposed model allow the partial recognition of user history also when the observed actions only partially meets the given behavior model constraints. This approach is more realistic for real time user support systems, with respect to standard boolean model recognition, when more than one user model is potentially available the amount of deviation from the models can be used as guide to generate the system support by anticipation, projection and other known techniques. Experiments based on logs from an e-learning platforms and plan compilation of the soft timed automaton shows the expressivity of the proposed model.

An evolutionary algorithm for adaptive online services in dynamic environment

An evolutionary adaptive algorithm for solving a class of online service provider problems in a dynamical web environment is introduced. In the online service provider scenario, a system continuously generates digital products and service instances by assembling components (e.g. headlines of online newspapers, search engine query results, advertising lists) to fulfill the requirements of a market of anonymous customers. The evaluation of a service instance can only be known by the feedback obtained after delivering it to the customer over the internet or through telephone networks. In dynamic domains available components and customer/agents preferences are changing over the time. The proposed algorithm employs typical genetic operators in order to optimize the service delivered and to adapt it to the environment feedback and evolution. Differently from classical genetic algorithms the goal of such systems is to maximize the average fitness instead of determining the single best optimal service/product. Experimental results for different classes of services, online newspapers and search engines, confirm the adaptive behavior of the proposed technique.

A multivalued planning model

In this work a model for planning with multivalued fluents and graded actions, based on the infinity-valued Lukasiewicz logic, is introduced. In multivalued planning, fluents and actions can assume truth values in [0, 1]. Multivalued fluents and graded actions allow to model many real situations where some features of the world cannot be modelled with boolean values and where actions can be executed with varying strength which produces graded effects as well. A correct/complete algorithm which solves bounded multivalued planning problems based on MIP compilation is also described and a prototype implementation is presented.

Soft User Behavior Modeling in Virtual Environments

A framework for online user behavior soft modeling is presented in this work. Behavior models of users in dynamic virtual environments has been described in the literature in terms of timed transition automata which can be compiled in a planning domain. The extended notion of soft timed transition automata is proposed in order to recognize a larger class of user histories. The notion of deviation from the user model allows to assess and evaluate in real time the dynamic behavior of users acting in virtual environments, such as e-learning and e-business platforms. The timed automata model allows to describe virtually infinite sequences of user actions subject to temporal constraints, while soft measures allows to assess recognition of behaviors by evaluating the amount of temporal deviation, additional or omitted actions contained in an observed behavior. The proposed model allow the partial recognition of user history also when the observed actions only partially meets the given behavior model constraints. This approach is more realistic for real time user support systems, with respect to standard boolean model recognition, when more than one user model is potentially available the amount of deviation from the models can be used as guide to generate the system support by anticipation, projection and other known techniques. Experiments based on logs from an e-learning platforms and plan compilation of the soft timed automaton show the expressivity of the proposed model.

Parallel Actions and Generalized Multivalued Constraints in Multivalued Planning

In this work an extension of the model for planning with multivalued fluents and graded actions introduced in [8] is proposed. This model is based on the infinity---valued Lukasiewicz logic, where the fluents can assume truth values in the interval [0,1] and actions can be executed at different application degrees also varying in [0,1]. Multivalued fluents and graded actions allow to model many real situations where some features of the world are fuzzy and where actions can be executed with varying strength. The main contributions of this paper are given by the introduction of the simultaneous executability of the graded actions and the extension of multivalued constraints to generalized multivalued constraints. An extension of the correct/complete algorithm which solves bounded multivalued planning problems is presented. It allows to solve problems with generalized constraints and simultaneous actions.