You are here

Publications

  1. Paolo Avesani; Angelo Susi; Sara Ferrari,
    Case-Based Ranking for Decision Support Systems,
    Very often a planning problem can be formulated as a ranking problem: i.e. to find an order relation over a set of alternatives. The ranking of a finite set of alternatives can be designed as a preference elicitation problem. While the case-based preference elicitation approach is more effective with respect to the first principle methods, still the scaling problem remains an open issue because the elicitation effort has a quadratic relation with the number of alternative cases. In this paper we propose a solution based on the machine learning techniques. We illustrate how a boosting algorithm can effectively estimate pairwise preferences and reduce the effort of the elicitation process. Experimental results, both on artificial data and a real world problem in the domain of civil defence, showed that a good trade-off can be achieved between the accuracy of the estimated preferences, and the elicitation effort of the end user,
    2003
  2. Paolo Avesani; P. Ferrari; Angelo Susi,
    Case-Based Environmental Risk Assessment,
    The traditional approaches to environmental risk analysis based on first principle methods don't cover assessment systemic vulnerability, a meaningful component of the natural hazard. We argue that complex task can be accomplished with a case-based approach through a process of pairwise preference elicitation. Up to now the main restriction of the case-based risk assessment was related to the scalability issue and the cognitive overload of the experts. In this paper we propose a methodology based on a mixed initiative strategy that combines the user preference elicitation and the machine rank approximation. The work includes both the case-based model and the related computational tools. We illustrate how a boosting algorithm can effectively estimate pairwise preferences and reduce the effort of the elicitation pro?cess. Experimental results, both on artificial data and on a real world in the problem domain of civil defence, showed that a good trade-off can achieved between accuracy of the estimated preferences, and elicitation effort of user,
    2003
  3. Giordano Adami; Paolo Avesani; Diego Sona,
    Self Organization of Documents in a Given Taxonomy,
    Hierarchical supervised classifiers are highly demanding in terms of labelled examples, because the number of categories are proportional to the size of a given taxonomy. In this case the bootstrapping process plays a key role because a small amount of labelled examples could prevent a successful exploitation of the learning techniques. This paper proposes a method to make a first hypothesis of categorization for a set of unlabelled documents with respect to a given empty hierarchy of concepts. The goal is to support a semi-automated management of the bootstrapping. The proposed solution is based on a revised model of self organizing maps in such a way that the unsupervised learning is biased by a taxonomy given as input to the model,
    2003
  4. Sriharsha Veeramachaneni; Paolo Avesani,
    Active Sampling for Feature Selection,
    In many knowledge discovery applications the data mining step is followed by further data acquisition. New data may consist of new instances and/or new features for the old instances. When new features are to be added an acquisition policy can help decide what features have to be acquired based on their predictive capability and the cost of acquisition. This can be posed as a feature selection problem where the feature values are not known in advance. We propose a technique to actively sample the feature values with the ultimate goal of choosing between alternative candidate features with minimum sampling cost. Our algorithm is based on extracting candidate features in a "region" of the instance space where the feature value is likely to alter our knowledge the most. An experimental evaluation on a standard database shows that it is possible outperform a random subsampling policy in terms of the accuracy in feature evaluation,
    2003
  5. Alessio Micheli; Diego Sona; Alessandro Sperduti,
    Contextual Processing of Structured Data by Recursive Cascade Correlation,
    We propose a first approach to deal with contextual information in structured domains by Recursive Neural Networks. The proposed model, i.e. Contextual Recursive Cascade Correlation (CRCC), a generalization of the Recursive Cascade Correlation (RCC) model, is able to partially remove the causality assumption by exploiting contextual information stored in frozen units. We formally characterize the properties of CRCC showing that it is able to compute contextual transductions and also some causal supersource transductions that RCC cannot compute. Experimental results on controlled sequences and on a real-world task involving chemical structures confirm the computational limitations of RCC, while assessing the efficiency and efficacy of CRCC in dealing both with pure causal and contextual prediction tasks. Moreover, results obtained for the real-world task show the superiority of the proposed approach versus RCC when exploring a task for which it is not known whether the structural causality assumption holds,
    2003
  6. Giordano Adami; Paolo Avesani; Diego Sona,
    Bootstrapping for Hierarchical Document Classification,
    Management of hierarchical organization of data (for example directories) is a process starting to play a key role in the knowledge management community, due to the great amount of human resources needed to create and maintain these organized repositories of information. This problem has been partially faced within the machine learning community by developing hierarchical supervised classifiers that help maintainers to categorize new resources within given hierarchies. Although such learning models seem to exploit the relational knowledge, they are highly demanding in terms of labelled examples, because the number of categories are related to the size of the corresponding hierarchy. Hence the creation of new directories or the modification of existing directories require strong investments. This paper proposes a semi-automatic process (interlived with human suggestions) which aim is to minimize (simplify) the work required to the administrators when creating, modifying, and maintaining directories. Within this process bootstrapping a taxonomy with examples represent a critical factor for the effective exploitation of any supervised learning model. For this reason we deepen the bootstrapping process proposing a method to make a first hypothesis of categorization for a set of unlabelled documents with respect to a given empty hierarchy of concepts. The proposed model, namely TaxSOM, which is based on a revisitation of self organizing maps, performs an unsupervised classification exploiting the a-priori knowledge encoded in a taxonomy structure both at the terminological and topological level. The ultimate goal of TaxSOM is to create the premise for a successful training of a supervised classifier,
    2003
  7. Giordano Adami; Paolo Avesani; Diego Sona,
    Bootstrapping of Supervised Hierarchical Classifiers,
    Hierarchical supervised classifiers are highly demanding in terms of labelled examples, because the number of categories are proportional to the size of a given taxonomy. In this case the bootstrapping process may represent a critical bottleneck for the deployment of learning techniques and the training of supervised classifiers. This work proposes a method to make a first hypothesis of categorization for a set of unlabelled documents with respect to a given empty hierarchy of concepts. The goal is to support a semi-automated management of the bootstrapping. The proposed solution is based on a revised model of self-organizing maps, namely TaxSOM, in such a way that the unsupervised learning is biased by a taxonomy given as input to the model,
    2003
  8. Giordano Adami; Paolo Avesani; Diego Sona,
    Unsupervised Categorization Exploiting a-priori Knowledge of a Taxonomy,
    Hiearchical categorization of documents is a task receiving growing interest both in the information retrieval and machine learning communities. Although hierarchical supervised classifiers seem to exploit the relational knowledge, at the same time they require an increasing amount of labelled examples. The bootstrap of hierarchical supervised classifiers is becoming a critical issue because the total amount of labelled examples is related to the size of concept hierarchy. This work proposes a solution to the bootstrap problem based on the self-organizing maps. This well known model is revised to enable the exploitation of the a-priori knowledge encoded in a taxonomy structure both at the terminological and topological level. An experimental evaluation has been performed on a collection of taxonomies extracted from the Google web directory,
    2003
  9. Paolo Avesani; Paolo Massa; Michele Nori; Angelo Susi,
    Collaborative Radio Community,
    Proceedings of the 2nd International Conference on Adaptive Hypermedia and Adaptive Web Based Systems,
    London,
    Springer Verlag,
    2002
    , (2nd International Conference on Adaptive Hypermedia and Adaptive Web Based Systems,
    Malaga, Spain,
    May 29-31, 2002)
  10. Angelo Susi; Anna Perini; Emanuele Olivetti,
    iEMSs 2002 Integrated Assessment and Decision Support,
    2002
    , pp. 426-
    431
    , (iEMSs 2002 Integrated Assessment and Decision Support,
    Lugano, Switzerland,
    24/06/2002 - 27/06/2002)

Pages