Fabrizio Silvestri

Dynamic personalization of web sites without user intervention

A novel online recommender system builds profiling models and offers suggestions without the user taking the lead.

An Online Recommender System for Large Web Sites

In this paper we propose a WUM recommender system, called SUGGEST 3.0, that dynamically generates links to pages that have not yet been visited by a user and might be of his potential interest. Differently from the recommender systems proposed so far, SUGGEST 3.0 does not make use of any off-line component, and is able to manage Web sites made up of pages dynamically generated. To this purpose SUGGEST 3.0 incrementally builds and maintains historical information by means of an incremental graph partitioning algorithm, requiring no off-line component. The main innovation proposed here is a novel strategy that can be used to manage large Web sites. Experiments, conducted in order to evaluate SUGGEST 3.0 performance, demonstrated that our system is able to anticipate users requests that will be made farther in the future, introducing a limited overhead on the Web server activity.

On-line generation of suggestions for Web users

The knowledge extracted from the analysis of historical information of a Web server can be used to develop personalization or recommendation systems. Web usage mining (WUM) systems are specifically designed to carry out this task by analyzing the data representing usage data about a particular Web site. Typically these systems are composed by two parts. One, executed offline, that analyze the server access logs in order to find a suitable categorization, and another executed online which is aimed at classifying the active requests, according to the previous offline analysis. In this paper we propose a WUM recommendation system, implemented as a module of the Apache Web server that is able to dynamically generate suggestions to pages that have not yet been visited by a user and might be of his potential interest. Differently from previously proposed WUM systems, SUGGEST 2.0 incrementally builds and maintains the historical information, without the need for an offline component, by means of an incremental graph partitioning algorithm. In the last part, we also analyze the quality of the suggestions generated and the performance of the module implemented. To this purpose we introduce also a new quality metric, which try to estimate the effectiveness of a recommendation system as the capacity of anticipating users requests that will be made farther in the future.

K-Model: A New Computational Model for Stream Processors

We introduce K-model, a computational model to evaluate the algorithms designed for graphic processors, and other architectures adhering to the stream programming model. We address the lack of a formal complexity model that properly accounts for memory contention, address coalescing in memory accesses, or the serial control of instruction flows. We study the impact of K-model rules on algorithm design. We devise a coalesced and low contention data access technique for Batchers networks, and we evaluate the effectiveness of this technique within our K-model. To evaluate the benefits in using K-model in evaluating solutions for streaming architectures, we compare the complexity of a sorting network built using our technique, and quick sort. Although in theory quick sort is more efficient than bitonic sort, empirically, our bitonic sorting network has been shown to be faster than the state-of-the-art implementation of quick sort on graphics processing units (GPUs). We use our K-model to prove that this observation should generally hold. As a side result, our technique to perform a Batchers network on GPUs improves the performance of one the fastest comparison-based solution for integers sorting.

A Study on the Effect of Application and Resource Characteristics on the QoS in Service Provisioning Environments

This article deals with the problem of quality provisioning in business service-oriented environments, examining the resource selection process as an initial matching of the provided to the demanded QoS. It investigates how the application and resource characteristics affect the provided level of QoS, a relationship that intuitively exists but has not yet being mapped. To do so, it focuses on identifying the application and resource parameters that affect the customer-defined QoS parameters. The article realistically centres upon modeling a data mining application and simple PC nodes in order to study how they affect response times. It moves on, by proving the existence of these specific relations and maps them using simple artificial neural networks so as to be able to wrap them in a single mechanism for resource selection based on customer QoS requirements and real time provider QoS capabilities.

An Open Architecture for QoS Information in Business Grids

Grid Computing is now in the state of development that can offer dynamic management of various parameters that affect the applications’ properties such as performance and reliability capabilities. The importance of that achievement is great, given the trend of migrating traditional service markets to inter-enterprise infrastructures and the resulting demand in more or different guarantees on the level of the Quality of Service. In that frame, we present a design pattern for monitoring and evaluating SLA terms on service-oriented architectures. This mechanism takes into account the actual capabilities of the service provider infrastructure and maps them to customer-centric Quality of Service terms, thus ensuring that agreements will not be validated. In this way it enables the estimation of the actual capability of the service to provide Quality of Service at a certain degree.

Representing Document Lengths with Identifiers

The length of each indexed document is needed by most common text retrieval scoring functions to rank it with respect to the current query. For efficiency purposes information retrieval systems maintain this information in the main memory. This paper proposes a novel strategy to encode the length of each document directly in the document identifier, thus reducing main memory demand. The technique is based on a simple document identifier assignment method and a function allowing the approximate length of each indexed document to be computed analytically.