Information Retrieval in large digital document repositories is at the same time a hard and crucial task. While the primary type of information available in documents is usually text, images play a very important role because they pictorially describe concepts that are dealt with in the document. Unfortunately, the semantic gap separating such a visual content from the underlying meaning is very wide. Additionally image processing techniques are usually very demanding in computational resources. Hence, only recently the area of Content-Based Image Retrieval has gained more attention. In this paper we describe a new technique to identify known objects in a picture based on a comparison of the shapes to known models. The comparison works by progressive approximations to save computational resources, and relies on novel algorithmic and representational solutions to improve preliminary shape extraction.

The current abundance of electronic documents requires automatic techniques that support the users in understanding their content and extracting useful information. To this aim, improving the retrieval performance must necessarily go beyond simple lexical interpretation of the user queries, and pass through an understanding of their semantic content and aims. It goes without saying that any digital library would take enormous advantage from the availability of effective Information Retrieval techniques to provide to their users. This paper proposes an approach to Information Retrieval based on a correspondence of the domain of discourse between the query and the documents in the repository. Such an association is based on standard general-purpose linguistic resources (WordNet and WordNet Domains) and on a novel similarity assessment technique. Although the work is at a preliminary stage, interesting initial results suggest to go on extending and improving the approach.

When using Horn Clause Logic as a representation formalism, the use of uninterpreted predicates cannot fully account for the complexity of some domains. In particular, in Machine Learning frameworks based on Horn Clause Logic, purely syntactic generalization cannot be applied to these kinds of predicates, requiring specific problems to be addressed and tailored strategies and techniques to be introduced. Among others, outstanding examples are those of numeric, taxonomic or sequential information. This paper deals with the case of (multidimensional) sequential information.Coverage and generalization techniques are devised and presented, and their integration in an incremental ILP system is used to run experiments showing its performance.

Document layout analysis is crucial in the automatic document processing workflow, because its outcome affects all subsequent processing steps. A first problem concerns the possibility of dealing not only with documents having easy layout, but with so-called Non-Manhattan layout documents as well. Another problem is that most available techniques can be applied to scanned document, due to the emphasis in previous decades being put on legacy documents digitization. Conversely, nowadays most documents come directly in digital format, and thus new techniques must be developed. A famous approach proposed in the literature for layout analysis was the RLSA, suitable to scanned black&white images and based the application of Run Length Smoothing and the AND logical operator. A recent variant thereof is based on the application of the OR operator, for which reason has been called RLSO. It exploits a bottom-up approach that proved able to handle even non-Manhattan layouts, on both scanned and natively digital documents. Like RLSA, it is based on the definition of thresholds for the smoothing operator, but the different approach requires different criteria than those that work in RLSA to define proper values. Since this is a hard and unnatural task for an (even expert) user, this paper proposes a technique to automatically define such thresholds for each single document, based on the distribution of spacing therein. Application on selected samples of documents, that aimed at covering a significant landscape of real cases, revealed that the approach is satisfactory for documents characterized by the use of a uniform text font size. It can provide a useful basis also for handling more complex cases.

This paper illustrates our work concerning the development of a layered architecture for deciding the situation-aware behavior of a Smart Home Environment (SHE). In the proposed approach, the surface level is directly embedded in the environment, while deeper levels represent the control software and perform progressively abstract and conceptual activities whose results can be fed back to the outside world (environment, user, supervisor). In particular, the reasoning layer is in charge of interpreting and transforming the data, collected through sensors of the smart environment, into high level knowledge about the situation. On the other hand, the learning layer, based on Inductive Logic Programming, suitably exploits the interaction of the user with the system to refine the user model and improve its future behavior. Finally, we provide the description of a typical scenario in which the proposed architecture might operate, along with a practical example of how the system might work.

Standardized processes are important for correctly carrying out activities in an organization. Often the procedures they describe are already in operation, and the need is to understand and formalize them in a model that can support their analysis, replication and enforcement. Manually building these models is complex, costly and error-prone. Hence, the interest in automatically learning them from examples of actual procedures. Desirable options are incrementality in learning and adapting the models, and the ability to express triggers and conditions on the tasks that make up the workflow. This paper proposes a framework based on First-Order Logic that solves many shortcomings of previous approaches to this problem in the literature, allowing to deal with complex domains in a powerful and flexible way. Indeed, First-Order Logic provides a single, comprehensive and expressive representation and manipulation environment for supporting all of the above requirements. A purposely devised experimental evaluation confirms the effectiveness and efficiency of the proposed solution.

Adapting the behavior of a smart environment means to tailor its functioning to both context situation and users’ needs and preferences. In this paper we propose an agent-based approach for controlling the behavior of a Smart Environment that, based on the recognized situation and user goal, selects a suitable workflow for combining services of the environment. We use the metaphor of a butler agent that employs user and context modeling to support proactive adaptation of the interaction with the environment. The interaction is adapted to every specific situation the user is in thanks to a class of agents called Interactor Agents.

This paper proposes an agent-based approach for proactively adapting the behavior of a Smart Environment that, based on the recognized situation and user goal, selects a suitable workflow for combining services of the environment. To this aim we have developed a multiagent infrastructure composed by different classes of agents specialized in reasoning and learning about the user and the context at different abstraction levels.

Digitization often introduces distortions in the form of odd perspective and curved text lines, especially toward the spine region of bound documents, that tamper both the creation of an acceptable digital reproduction of the document and the successful extraction of its textual content using OCR techniques. While already known for traditional acquisition means such as flatbed or planetary scanners, the problem gets even worse with the use of cameras, whose current widespread availability may open new opportunities for librarians and archivists. Dewarping is in charge of handling this kind of problems. This paper proposes a novel model-based dewarping method, aimed at solving some of the shortcomings of existing approaches through the use of a mixture of image processing and numerical analysis tools. The method is based on the construction of a curvilinear grid on each page of the document by means of piecewise polynomial fit and appropriate equipartition of a selection of its curved text lines. We show the method on sample documents and evaluate its impact on successful rate of OCR on a dataset.

Information Retrieval in digital libraries is at the same time a hard task and a crucial issue. While the primary type of information available in digital documents is usually text, images play a very important role because they pictorially describe concepts that are dealt with in the document. Unfortunately, the semantic gap separating such a visual content from the underlying meaning is very wide, and additionally image processing techniques are usually very demanding in computational resources. Hence, only recently the area of Content-Based Image Retrieval has gained more attention. In this paper we describe a new technique to identify known objects in a picture. It is based on shape contours, and works by progressive approximations to save computational resources and to improve preliminary shape extraction. Small (controlled) and more extensive experiments are illustrated, yielding interesting results.

In the last decades speaking and writing habits have changed. Many works faced the author identification task by exploiting frequencybased approaches, numeric techniques or writing style analysis. Following the last approach we propose a technique for author identification based on First-Order Logic. Specifically, we translate the complex data represented by natural language text to complex (relational) patterns that represent the writing style of an author. Then, we model an author as the result of clustering the relational descriptions associated to the sentences. The underlying idea is that such a model can express the typical way in which an author composes the sentences in his writings. So, if we can map such writing habits from the unknown-author model to the known-author model, we can conclude that the author is the same. Preliminary results are promising and the approach seems viable in real contexts since it does not need a training phase and performs well also with short texts.

Layout analysis is a fundamental step in automatic document processing, because its outcome affects all subsequent processing steps. Many different techniques have been proposed to perform this task. In this work, we propose a general bottom-up strategy to tackle the layout analysis of (possibly) non-Manhattan documents, and two specializations of it to handle both bitmap and PS/PDF sources. A famous approach proposed in the literature for layout analysis was the RLSA. Here we consider a variant of RLSA, called RLSO (short for “Run-Lengh Smoothing with OR”), that exploits the OR logical operator instead of the AND and is particularly indicated for the identification of frames in non-Manhattan layouts. Like RLSA, RLSO is based on thresholds, but based on different criteria than those that work in RLSA. Since setting such thresholds is a hard and unnatural task for (even expert) users, and no single threshold can fit all documents, we developed a technique to automatically define such thresholds for each specific document, based on the distribution of spacing therein. Application on selected sample documents, that cover a significant landscape of real cases, revealed that the approach is satisfactory for documents characterized by the use of a uniform text font size.

Assessing whether two documents were written by the same author is a crucial task, especially in the Internet age, with possible applications to philology and forensics. The problem has been tackled in the literature by exploiting frequency-based approaches, numeric techniques or writing style analysis. Focusing on this last perspective, this paper proposes a novel technique that takes into account the structure of sentences, assuming that it is strictly related to the author's writing style. Specifically, a (collection of) text(s) in natural language written by a given author is translated into a set of First-Order Logic descriptions, and a model of the author's writing habits is obtained as the result of clustering these descriptions. Then, if an overlapping exists between the models of a known author and of an unknown one, the conclusion can be drawn that they are the same person. Among the advantages of this approach, it does not need a training phase, and performs well also on short texts and/or small collections.

Automatic processing of text documents requires techniques that can go beyond the lexical level, and are able to handle the semantics underlying natural language sentences. A support for such techniques can be provided by taxonomies that connect terms to the underlying concepts, and concepts to each other according to different kinds of relationships. An outstanding example of such a kind of resources is WordNet. On the other hand, whenever automatic inferences are to be made on a given domain, a generalization technique, and corresponding operational procedures, are needed. This paper proposes a generalization technique for taxonomic information and applies it to WordNet, providing examples that prove its behavior to be sensible and effective.

Detecting the reading order among the layout components of a document's page is fundamental to ensure effectiveness or even applicability of subsequent content extraction steps. While in single-column documents the reading flow can be straightforwardly determined, in more complex documents the task may become very hard. This paper proposes an automatic strategy for identifying the correct reading order of a document page's components based on abstract argumentation. The technique is unsupervised, and works on any kind of document based only on general assumptions about how humans behave when reading documents. Experimental results show that it is effective in more complex cases, and requires less background knowledge, than previous solutions that have been proposed in the literature.

Process Management techniques are useful in domains where the availability of a (formal) process model can be leveraged to monitor, supervise, and control a production process. While their classical application is in the business and industrial fields, other domains may profitably exploit Process Management techniques. Some of these domains (e.g., people’s behavior, General Game Playing) are much more flexible and variable than classical ones, and, thus, raise the problem of predicting which activities will be carried out next, a problem that is not so compelling in classical fields. When the process model is learned automatically from examples of process executions, which is the task of Process Mining, the prediction performance may also provide indirect indications on the correctness and reliability of the learned model. This paper proposes and compares two strategies for activity prediction using the WoMan framework for workflow management. The former proved to be able to handle complex processes, the latter is based on the classic and consolidated Naïve Bayes approach. An experimental validation allows us to draw considerations on the pros and cons of each, used both in isolation and in combination.

In this paper we propose an agent-based approach for controlling the behavior of a Smart Home Environment that, based on the recognized situation and user goal, selects a suitable workflow for combining services of the environment. To this aim we have developed a butler agent that employs user and context modeling for supporting proactive adaptation of the interaction with the house. The user can interact with the proposed services by accepting, declining or changing them. Such a feedback is exploited by the learning component of the butler to refine the user model and improve its future behavior accordingly. In order to provide a description of how the system might work, a practical example is shown.

Many document collections are private and accessible only by selected people. Especially in business realities, such collections need to be managed, and the use of an external taxonomic or ontological resource would be very useful. Unfortunately, very often domain-specific resources are not available, and the development of techniques that do not rely on external resources becomes essential. Automated learning of conceptual graphs from restricted collections needs to be robust with respect to missing or partial knowledge, that does not allow to extract a full conceptual graph and only provides sparse fragments thereof. This work proposes a way to deal with these problems applying relational clustering and generalization methods. While clustering collects similar concepts, generalization provides additional nodes that can bridge separate pieces of the graph while expressing it at a higher level of abstraction. In this process, considering relational information allows a broader perspective in the similarity assessment for clustering, and ensures more flexible and understandable descriptions of the generalized concepts. The final conceptual graph can be used for better analyzing and understanding the collection, and for performing some kind of reasoning on it.

Abstract argumentation allows to determine in an easy, formal way which claims survive in a conflicting dispute. It works by considering claims as abstract entities, and expressing attack relationships among them. However, this level of expressiveness prevents abstract argumentation systems from being directly applied to reasoning processes where the context is relevant. An outstanding example is when a claim is supported by appealing to authority, so that the audience assigns reliability to the claim's justication based on the authority's renowned experience in the domain. To handle this, we propose to enrich the classical representation used in abstract argumentation by associating arguments with weights that express their degree of authority. The weights' values define their strength in the given domain, which in turn should affect the evaluation of their degree of justification. This paper defines a strategy to combine these weights in order to determine which arguments withstand in a dispute concerning a given domain. Such a strategy was implemented in the ARCA system, that allows to comfortably set up argumentation problems and solve them using both traditional extension-based semantics and the proposed evaluation approach. ARCA is used to illustrate the proposed strategy by means of sample use cases.

Contemporary libraries have changed quickly their social role and function due to the proliferation and diversification of multimedia digital documents, becoming complex networks able to support communication and collaboration among the various distributed users communities. Technologies have not grown in step with the needs generated by this new approach, except in specific areas and implications. Hence the need to design an integrated digital library architecture that covers by advanced techniques the whole spectrum of functionality, without which the same social and cultural function of a modern digital library is at risk. This paper briefly describes an architecture that aims to bridge this gap, bringing together the experience, expertise and software systems developed by university and companies researchers. A prototype of the system is under development.

The recent explosion in Internet usage and the growing amount of digital images caused by the more and more ubiquitous presence of digital cameras has created a demand for effective and flexible techniques for automatic image retrieval. As the volume of the data increases, memory and processing requirements need to correspondingly increase at the same rapid pace, and this is often prohibitively expensive. Image collections on this scale make performing even the most common and simple image processing and machine learning tasks non trivial. In this paper we present a method to reduce the computational complexity of a widely known method for image indexing and retrieval based on a second order statistical measure. The aim of the paper is twofold: Q1) is it possible to efficiently extract an approximate distribution of the image features with a resulting low error? Q2) how the resulting approximate distribution affects the similarity-based accuracy? In particular, we propose a sampling method to approximate the distribution of correlograms, adopting a Monte Carlo approach to compute the distribution on a subset of pixels uniformly sampled from the original image. A further variant is to sample the neighborhood of each pixel too. Validation on the Caltech 101 dataset proved that the proposed approximate distribution, obtained with a considerable decrease of the computational time, has an error very low when compared to the exact distribution. Result obtained in the second experiment on a similarity-based ranking task are encouraging.

For many real-world applications it is important to choose the right representation language. While the setting of First Order Logic (FOL) is the most suitable one to model the multi-relational data of real and complex domains, on the other hand it puts the question of the computational complexity of the knowledge induction process. A way of tackling the complexity of such real domains, in which a lot of relationships are required to model the objects involved, is to use a method that reformulates a multi-relational learning task into an attribute-value one. In this chapter we present an approximate reasoning method able to keep low the complexity of a relational problem by using a stochastic inference procedure. The complexity of the relational language is decreased by means of a propositionalization technique, while the NP-completeness of the deduction is tackled using an approximate query evaluation. The proposed approximate reasoning technique has been used to solve the problem of relational rule induction as well as the task of relational clustering. An anytime algorithm has been used for the induction, implemented by a population based method, able to efficiently extract knowledge from relational data, while the clustering task, both unsupervised and supervised, has been solved using a Partition Around Medoid (PAM) clustering algorithm. The validity of the proposed techniques has been proved making an empirical evaluation on real-world datasets.

While multimedia digital documents are progressively spreading, most of the content of Digital Libraries is still in the form of text, and this predominance will probably never be questioned. Except pure display of these documents, all other tasks are based on some kind of Natural Language Processing, that must be supported by suitable linguistic resources. Since these resources are clearly language-specific, they might be unavailable for several languages, and manually building them is costly, time-consuming and error-prone. This paper proposes a methodology to automatically learn linguistic resources for a natural language starting from texts written in that language. The learned resources may enable further high-level processing of documents in that language, and/or be taken as a basis for further manual refinements. Experimental results show that its application may effectively provide useful linguistic resources in a fully automatic manner.

The current spread of digital documents raised the need of effective content-based retrieval techniques. Since manual indexing is infeasible and subjective, automatic techniques are the obvious solution. In particular, the ability of properly identifying and understanding a document’s structure is crucial, in order to focus on the most significant components only. At a geometrical level, this task is known as Layout Analysis, and thoroughly studied in the literature. On suitable descriptions of the document layout, Machine Learning techniques can be applied to automatically infer models of classes of documents and of their components. Indeed, organizing the documents on the grounds of the knowledge they contain is fundamental for being able to correctly access them according to the user’s needs. Thus, the quality of the layout analysis outcome biases the next understanding steps. Unfortunately, due to the variety of document styles and formats, the automatically found structure often needs to be manually adjusted. We propose the application of supervised Machine Learning techniques to infer correction rules to be applied to forthcoming documents. A first-order logic representation is suggested, because corrections often depend on the relationships of the wrong components with the surrounding ones. Moreover, as a consequence of the continuous flow of documents, the learned models often need to be updated and refined, which calls for incremental abilities. The proposed technique, embedded in a prototypical version of the document processing system DOMINUS, using the incremental first-order logic learner INTHELEX, revealed good performance in real-world experiments.

Markov Logic (ML) combines Markov networks (MNs) and first-order logic by attaching weights to first-order formulas and using these as templates for features of MNs. State-of-the-art structure learning algorithms in ML maximize the likelihood of a database by performing a greedy search in the space of structures. This can lead to suboptimal results because of the incapability of these approaches to escape local optima. Moreover, due to the combinatorially explosive space of potential candidates these methods are computationally prohibitive. We propose a novel algorithm for structure learning in ML, based on the Iterated Local Search (ILS) metaheuristic that explores the space of structures through a biased sampling of the set of local optima. We show through real-world experiments that the algorithm improves accuracy and learning time over the state-of-the-art algorithms. On the other side MAP and conditional inference for ML are hard computational tasks. This paper presents two algorithms for these tasks based on the Iterated Robust Tabu Search (IRoTS) metaheuristic. The first algorithm performs MAP inference and we show through extensive experiments that it improves over the state-of-the-art algorithm in terms of solution quality and inference time. The second algorithm combines IRoTS steps with simulated annealing steps for conditional inference and we show through experiments that it is faster than the current state-of-the-art algorithm maintaining the same inference quality.

Statistical Relational Learning (SRL) is a growing field in Machine Learning that aims at the integration of logic-based learning approaches with probabilistic graphical models. Markov Logic Networks (MLNs) are one of the state-of-the-art SRL models that combine first-order logic and Markov networks (MNs) by attaching weights to first-order formulas and viewing these as templates for features of MNs. Learning models in SRL consists in learning the structure (logical clauses in MLNs) and the parameters (weights for each clause in MLNs). Structure learning of MLNs is performed by maximizing a likelihood function (or a function thereof) over relational databases and MLNs have been successfully applied to problems in relational and uncertain domains. However, most complex domains are characterized by incomplete data. Until now SRL models have mostly used Expectation-Maximization (EM) for learning statistical parameters under missing values. Multistrategic learning in the relational setting has been a successful approach to dealing with complex problems where multiple inference mechanisms can help solve different subproblems. Abduction is an inference strategy that has been proven useful for completing missing values in observations. In this paper we propose two frameworks for integrating abduction in SRL models. The first tightly integrates logical abduction with structure and parameter learning of MLNs in a single step. During structure search guided by conditional likelihood, clause evaluation is performed by first trying to logically abduce missing values in the data and then by learning optimal pseudo-likelihood parameters using the completed data. The second approach integrates abduction with Structural EM of [17] by performing logical abductive inference in the E-step and then by trying to maximize parameters in the M-step.

The coalition structure generation problem represents an active research area in multi-agent systems. A coalition structure is defined as a partition of the agents involved in a system into disjoint coalitions. The problem of finding the optimal coalition structure is NP-complete. In order to find the optimal solution in a combinatorial optimization problem it is theoretically possible to enumerate the solutions and evaluate each. But this approach is infeasible since the number of solutions often grows exponentially with the size of the problem. In this paper we present a greedy adaptive search procedure (GRASP) to efficiently search the space of coalition structures in order to find an optimal one.

Studying, understanding and exploiting the content of a digital library, and extracting useful information thereof, require automatic techniques that can effectively support the users. To this aim, a relevant role can be played by concept taxonomies. Unfortunately, the availability of such a kind of resources is limited, and their manual building and maintenance are costly and error-prone. This work presents ConNeKTion, a tool for conceptual graph learning and exploitation. It allows to learn conceptual graphs from plain text and to enrich them by finding concept generalizations. The resulting graph can be used for several purposes: finding relationships between concepts (if any), filtering the concepts from a particular perspective, keyword extraction and information retrieval. A suitable control panel is provided for the user to comfortably carry out these activities.

Studying, understanding and exploiting the content of a digital library, and extracting useful information thereof, require automatic techniques that can effectively support the users. To this aim, a relevant role can be played by concept taxonomies. Unfortunately, the availability of such a kind of resources is limited, and their manual building and maintenance are costly and error-prone. This work presents ConNeKTion, a tool for conceptual graph learning and exploitation. It allows to learn conceptual graphs from plain text and to enrich them by finding concept generalizations. The resulting graph can be used for several purposes: finding relationships between concepts (if any), filtering the concepts from a particular perspective, extracting keyword, retrieving information and identifying the author. ConNeKTion provides also a suitable control panel, to comfortably carry out these activities.

Information Retrieval in digital libraries is at the same time a hard task and a crucial issue. While the primary type of information available in digital documents is usually text, images play a very important role because they pictorially describe concepts that are dealt with in the document. Unfortunately, the semantic gap separating such a visual content from the underlying meaning is very wide, and additionally image processing techniques are usually very demanding in computational resources. Hence, only recently the area of Content-Based Image Retrieval has gained more attention. In this paper we describe a new technique to identify known objects in a picture. It is based on shape contours, and works by progressive approximations to save computational resources and to improve preliminary shape extraction. Small (controlled) and more extensive experiments are illustrated, yielding interesting results.

The current abundance of electronic documents requires au- tomatic techniques that support the users in understanding their con- tent and extracting useful information. To this aim, it is important to have conceptual taxonomies that express common sense and implicit relationships among concepts. This work proposes a mix of several tech- niques that are brought to cooperation for learning them automatically. Although the work is at a preliminary stage, interesting initial results suggest to go on extending and improving the approach.

The main goal of the project was the development of a District Service Center for the SMEs of the Textile and Clothing sector. In particular, it investigates the introduction of innovative technologies to improve the process/product innovation of the sector. In this direction, the research unit proposal consisted in introducing document processing and indexing techniques on a variety (both for structure and content) of document formats whit the aim of improving the exchange of data among companies and the semantic content-based retrieval for the real companies’ needs.

Store, process and manage collections of different types of documents is one of the needs of most organizations, especially universities. The management of libraries, scientific conferences, research projects are just some of the practical cases that require advanced solutions. DOMINUSplus is an open project born with the aim of harmonizing the Artificial Intelligence approaches developed at the LACAM laboratory with the research on Digital Libraries in a general software backbone for document processing and management, extensible with ad-hoc solutions for specific problems and context (such as universities).

Activities of most organizations, and of universities in particular, involve the need to store, process and manage collections of different kinds of documents. Examples that require advanced solutions to such issues include the management of libraries, scientific conferences, research projects. DOMINUSplus is an open project born with the aim of harmonizing the Artificial Intelligence approaches developed at the LACAM laboratory with the research on Digital Libraries in a general software backbone for document processing and management, extensible with ad-hoc solutions for specific problems and context (such as universities).

We present high performing SLS algorithms for learning and inference in Markov Logic Networks (MLNs). MLNs are a state-of-the-art representation formalism that integrates first-order logic and probability. Learning MLNs structure is hard due to the combinatorial space of candidates caused by the expressive power of first-order logic. We present current work on the development of algorithms for learning MLNs, based on the Iterated Local Search (ILS) metaheuristic. Experiments in real-world domains show that the proposed approach improves accuracy and learning time over the existing state-of-the-art algorithms. Moreover, MAP and conditional inference in MLNs are hard computational tasks too. This paper presents two algorithms for these tasks based on the Iterated Robust Tabu Search (IRoTS) schema. The first algorithm performs MAP inference by performing a RoTS search within a ILS iteration. Extensive experiments show that it improves over the state-of the-art algorithm in terms of solution quality and inference times. The second algorithm combines IRoTS with simulated annealing for conditional inference and we show through experiments that it is faster than the current state-of-the-art algorithm maintaining the same inference quality.

In Artificial Intelligence with Coalition Structure Generation (CSG) one refers to those cooperative complex problems that require to find an optimal partition (maximizing a social welfare) of a set of entities involved in a system. The solution of the CSG problem finds applications in many fields such as Machine Learning (set covering machines, clustering), Data Mining (decision tree, discretization), Graph Theory, Natural Language Processing (aggregation), Semantic Web (service composition), and Bioinformatics. The problem of finding the optimal coalition structure is NP-complete. In this paper we present a greedy adaptive search procedure (GRASP) with path-relinking to efficiently search the space of coalition structures. Experiments and comparisons to other algorithms prove the validity of the proposed method in solving this hard combinatorial problem.

The current tools to create OWL-S annotations have been designed starting from the knowledge engineer’s point of view. Unfortunately, the formalisms underlying Semantic Web languages are often incomprehensible to the developers of Web services. To bridge this gap, it is desirable that developers are provided with suitable tools that do not necessarily require knowledge of these languages in order to create annotations on Web services. With reference to some characteristics of the involved technologies, this work addresses these issues, proposing guidelines that can improve the annotation activity of Web service developers. Following these guidelines, we also designed a tool that allows a Web service developer to annotate Web services without requiring him to have a deep knowledge of Semantic Web languages. A prototype of such a tool is presented and discussed in this paper.

The spread and abundance of electronic documents requires automatic techniques for extracting useful information from the text they contain. The availability of conceptual taxonomies can be of great help, but manually building them is a complex and costly task. Building on previous work, we propose a technique to automatically extract conceptual graphs from text and reason with them. Since automated learning of taxonomies needs to be robust with respect to missing or partial knowledge and flexible with respect to noise, this work proposes a way to deal with these problems. The case of poor data/sparse concepts is tackled by finding generalizations among disjoint pieces of knowledge. Noise is handled by introducing soft relationships among concepts rather than hard ones, and applying a probabilistic inferential setting. In particular, we propose to reason on the extracted graph using different kinds of relationships among concepts, where each arc/relationship is associated to a weight that represents its likelihood among all possible worlds, and to face the problem of sparse knowledge by using generalizations among distant concepts as bridges between disjoint portions of knowledge.

Users of Digital libraries require more intelligent interaction functionality to satisfy their needs. In this perspective, the most important features are flexibility and capability of adapting these functionalities to specific users. However, the main problem of current systems is their inability to support different needs of individual users due both to their inability to identify those needs, and, more importantly, to insufficient mapping of those needs to the available resources/services. The approaches considered in this paper to tackle such problems concern the use of Machine Learning techniques to adapt the set of user stereotypes with the aim of modelling user interests and behaviour in order to provide the most suitable service. A purposely designed simulation scenario was exploited to show the applicability of the proposal.

We present a suite of Machine Learning and knowledge-based components for textual-profile based gene prioritization. Most genetic diseases are characterized by many potential candidate genes that can cause the disease. Gene expression analysis typically produces a large number of co-expressed genes that could be potentially responsible for a given disease. Extracting prior knowledge from text-based genomic information sources is essential in order to reduce the list of potential candidate genes to be then further analyzed in laboratory. In this paper we present a suite of Machine Learning algorithms and knowledge-based components for improving the computational gene prioritization process. The suite includes basic Natural Language Processing capabilities, advanced text classification and clustering algorithms, robust information extraction components based on qualitative and quantitative keyword extraction methods and exploitation of lexical knowledge bases for semantic text processing.

With the increasing amount of information in electronic form the fields of Machine Learning and Data Mining continue to grow by providing new advances in theory, applications and systems. The aim of this paper is to consider some recent theoretical aspects and approaches to ML and DM with an emphasis on the Italian research.

Studying, understanding and exploiting the content of a document collection require automatic techniques that can effectively support the users in extracting useful information from it and reason with this information. Concept networks (e.g., taxonomies) may play a relevant role in this perspective, but are seldom available, and cannot be manually built and maintained cheaply and reliably. On the other hand, automated learning of these resources from text needs to be robust with respect to missing or partial knowledge, because often only sparse fragments of the target network can be extracted. This work presents ConNeKTion, a tool that is able to learn concept networks from plain text and to structure and enrich them by finding concept generalizations. The proposed methodologies are general and applicable to any language. It also provides functionalities for the exploitation of the learned knowledge, and a control panel that allows the user to comfortably carry out these activities. Several experiments and applications are reported, showing the usefulness and flexibility of ConNeKTion.

The availability of automatic support may sometimes determine the successful accomplishment of a process. Such a support can be provided if a model of the intended process is available. Many realworld process models are very complex. Additionally, their components might be associated to conditions that determine whether they are to be carried out or not. These conditions may be in turn very complex, involving sequential relationships that take into account the past history of the current process execution. In this landscape, writing and setting up manually the process models and conditions might be infeasible, and even standard Machine Learning approaches may be unable to infer them. This paper presents a First-Order Logic-based approach to learn complex process models extended with conditions. It combines two powerful Inductive Logic Programming systems. The overall system was exploited to learn the daily routines of the user of a smart environment, for predicting his needs and comparing the actual situation with the expected one. In addition to proving the efficiency and effectiveness of the system, the outcomes show that complex, human-readable and interesting preconditions can be learned for the tasks involved in the process.

Tables are among the most informative components of documents, because they are exploited to compactly and intuitively represent data, typically for understandability purposes. The needs are to identify and extract tables from documents, and, on the other hand, to be able to extract the data they contain. The latter task involves the understanding of a table structure. Due to the variability in style, size, and aims of tables, algorithmic approaches to this task can be insufficient, and the exploitation of machine learning systems may represent an effective solution. This paper proposes the exploitation of a first-order logic representation, that is able to capture the complex spatial relationships involved in a table structure, and of a learning system that can mix the power of this representation with the flexibility of statistical approaches. The obtained encouraging results suggest further investigation and refinement of the proposal.

Understanding what the user is doing in a Smart Environment is important not only for adapting the environment behavior, e.g. by providing the most appropriate combination of services for the recognized situation, but also for identifying situations that could be problematic for the user. Manually building models of the user processes is a complex, costly and error-prone engineering task. Hence, the interest in automatically learning them from examples of actual procedures. Incremental adaptation of the models, and the ability to express/learn complex conditions on the involved tasks, are also desirable. First-order logic provides a single comprehensive and powerful framework for supporting all of the above. This paper presents a First-Order Logic incremental method for inferring process models, and show its application to the user's daily routines, for predicting his needs and comparing the actual situation with the expected one. Promising results have been obtained with both controlled experiments that proved its efficiency and eectiveness, and with a domain-specic dataset.

Information retrieval effectiveness has become a crucial issue with the enormous growth of available digital documents and the spread of Digital Libraries. Search and retrieval are mostly carried out on the textual content of documents, and traditionally only at the lexical level. However, pure term-based queries are very limited because most of the information in natural language is carried by the syntactic and logic structure of sentences. To take into account such a structure, powerful relational languages, such as first-order logic, must be exploited. However, logic formulæ constituents are typically uninterpreted (they are considered as purely syntactic entities), whereas words in natural language express underlying concepts that involve several implicit relationships, as those expressed in a taxonomy. This problem can be tackled by providing the logic interpreter with suitable taxonomic knowledge. This work proposes the exploitation of a similarity framework that includes both structural and taxonomic features to assess the similarity between First-Order Logic (Horn clause) descriptions of texts in natural language, in order to support more sophisticated information retrieval approaches than simple term-based queries. Evaluation on a sample case shows the viability of the solution, although further work is still needed to study the framework more deeply and to further refine it.

Markov Logic Networks (MLNs) combine Markov networks (MNs) and firstorder logic by attaching weights to first-order formulas and using these as templates for features of MNs. Learning the structure of MLNs is performed by state-of-the-art methods by maximizing the likelihood of a relational database. This leads to suboptimal results for prediction tasks due to the mismatch between the objective function (likelihood) and the task of classification (maximizing conditional likelihood (CL)). In this paper we propose two algorithms for learning the structure of MLNs. The first maximizes the CL of query predicates instead of the joint likelihood of all predicates while the other maximizes the area under the Precision-Recall curve (AUC). Both algorithms set the parameters by maximum likelihood and choose structures by maximizing CL or AUC. For each of these algorithms we develop two different searching strategies. The first is based on Iterated Local Search and the second on Greedy Randomized Adaptive Search Procedure.We compare the performances of these randomized search approaches on realworld datasets and show that on larger datasets, the ILS-based approaches perform better, both in terms of CLL and AUC, while on small datasets, ILS and RBS approaches are competitive and RBS can also lead to better results for AUC.

Terrestrial Laser Scanner surveys performed in coastal area have generated 3D cloud points used to obtain digital elevation model and standard deviation of the micro-topography of coastal surfaces. Starting from data collected, roughness. coefficients have been estimated for each surface typology characterizing the coastal area (sand calcarenite, vegetation, etc). Applying Machine Learning techniques on digital images, the extension and the surface typology of these areas have been obtained. All data collected have been elaborated by means of software implemented stalling from known hydrodynamic formula to evaluate the inland penetration of a hypothesized tsunami.

The discovery of Web services is often in uenced by the rigid structure of registries containing their XML description. In recent years some methods that replace the traditional UDDI registry with Peer To Peer networks for the creation of catalogs of Web services have been proposed in order to make this structure exible and usable. This paper proposes a dierent view by placing the semantic description of services as content of P2P networks and showing that all the needed information for an ecient Web service discovery is already contained in its OWL-S description.

Horn clause Logic is a powerful representation language exploited in Logic Programming as a computer programming framework and in Inductive Logic Programming as a formalism for expressing examples and learned theories in domains where relations among objects must be expressed to fully capture the relevant information. While the predicates that make up the description language are defined by the knowledge engineer and handled only syntactically by the interpreters, they sometimes express information that can be properly exploited only with reference to a suitable background knowledge in order to capture unexpressed and underlying relationships among the concepts described. This is typical when the representation includes numerical information, such as single values or intervals, for which simple syntactic matching is not sufficient. This work proposes an extension of an existing framework for similarity assessment between First-Order Logic Horn clauses, that is able to handle numeric information in the descriptions. The viability of the solution is demonstrated on sample problems.

A paper document processing system is an information sys- tem component which transforms information on printed or handwritten documents into a computer-revisable form. In intelligent systems for pa- per document processing this information capture process is based on knowledge of the specific layout and logical structures of the documents. In this project we design a framework which combines technologies for the acquisition and storage of printed documents with knowledge-based techniques to represent and understand the information they contain. The innovative aspects of this work strengthen its applicability to tools that have been developed for building digital libraries.

Protein fold recognition is the problem of determining whether a given protein sequence folds into a previously observed structure. An uncertainty complication is that it is not always true that the structure has been previously observed. Markov logic networks (MLNs) are a powerful representation that combines first-order logic and probability by attaching weights to first-order formulas and using these as templates for features of Markov networks. In this chapter, we describe a simple temporal extension of MLNs that is able to deal with sequences of logical atoms. We also propose iterated robust tabu search (IRoTS) for maximum a posteriori (MAP) inference and Markov Chain-IRoTS (MC-IRoTS) for conditional inference in the new framework. We show how MC-IRoTS can also be used for discriminative weight learning. We describe how sequences of protein secondary structure can be modeled through the proposed language and show through some preliminary experiments the promise of our approach for the problem of protein fold recognition from these sequences.

Pedagogical Conversational Agents (PCAs) have the advantage of offering to students not only task-oriented support but also the possibility to interact with the computer media at a social level. This form of intelligence is particularly important when the character is employed in an educational setting. This paper reports our initial results on the recognition of users' social response to a pedagogical agent from the linguistic, acoustic and gestural analysis of the student communicative act.

Ambient Intelligence aims at promoting an effective, natural and personalized interaction with the environment services. In order to provide the most appropriate answer to the user requests, an Ambient Intelligence system should model the user by considering not only the cognitive ingredients of his mental state, but also extra-rational factors such as affect, engagement, attitude, and so on. This paper describes a study aimed at building a multimodal framework for recognizing the social response of users during interaction with embodied agents in the context of ambient intelligence. In particular, we describe how we extended a model for recognizing the social attitude in text-based dialogs by adding two additional knowledge sources: speech and gestures. Results of the study show that these additional knowledge sources may help in improving the recognition of the users' attitude during interaction.

Pedagogical Conversational Agents (PCAs) have the advantage of offering to students not only task-oriented support but also the possibility to interact with the computer media at a social level. This form of intelligence is particularly important when the character is employed in an educational setting. This paper reports our initial results on the recognition of users' social response to a pedagogical agent from the linguistic, acoustic and gestural analysis of the student communicative act.

In multiagent adversarial environments, the adversary consists of a team of opponents that may interfere with the achievement of goals. In this domain agents must be able to quickly adapt to the environment and infer knowledge from other agents’ deportment to identify the future behaviors of opponents. We present a relational model to characterize adversary teams based on its behavior. A team’s deportment is represent by a set of relational sequences of basic actions extracted from their observed behaviors. Based on this, we present a similarity measure to classify the teams’ behavior. The sequence extraction and classification are implemented in the domain of simulated robotic soccer, and experimental results are presented.

Symbolic Machine Learning systems and applications, especially when applied to real-world domains, must face the problem of concepts that cannot be captured by a single denition, but require several alternate definitions, each of which covers part of the full concept extension. This problem is particularly relevant for incremental systems, where progressive covering approaches are not applicable, and the learn- ing and refinement of the various definitions is interleaved during the learning phase. In these systems, not only the learned model depends on the order in which the examples are provided, but it also depends on the choice of the specific definition to be refined. This paper proposes different strategies for determining the order in which the alternate definitions of a concept should be considered in a generalization step, and evaluates their performance on a real-world domain dataset

When used as an interface in the context of Ambient Assisted Living (AAL), a social robot should not just provide a task-oriented support. It should also try to establish a social empathic relation with the user. To this aim, it is crucial to endow the robot with the capability of recognizing the user’s affective state and reason on it for triggering the most appropriate communicative behavior. In this paper we describe how such an affective reasoning has been implemented in the NAO robot for simulating empathic behaviors in the context of AAL. In particular, the robot is able to recognize the emotion of the user by analyzing communicative signals extracted from speech and facial expressions. The recognized emotion allows triggering the robot’s affective state and, consequently, the most appropriate empathic behavior. The robot’s empathic behaviors have been evaluated both by experts in communication and through a user study aimed at assessing the perception and interpretation of empathy by elderly users. Results are quite satisfactory and encourage us to further extend the social and affective capabilities of the robot.

Conversational agents have been widely used in pedagogical contexts. They have the advantage of offering to users not only a task-oriented support, but also the possibility to relate with the system at social level. Therefore, besides endowing the conversational agent with knowledge necessary to fulfill pedagogical goals, it is important to provide the agent with social intelligence. To do so the agent should be able to recognize the social attitude of the user during the interaction in order to accommodate the conversational strategy. In this paper we illustrate how we defined and applied a model for recognizing the social attitude of the student in natural interaction with a Pedagogical Conversational Agent (PCA) starting from the linguistic, acoustic and gestural analysis of the communicative act.

One of the most appreciated functionality of computers nowadays is their being a means for communication and information sharing among people. With the spread of the Internet, several complex interactions have taken place among people, giving rise to huge Information Networks based on these interactions. Social Networks potentially represent an invaluable source of information that can be exploited for scientific and commercial purposes. On the other hand, due to their distinguishing peculiarities (huge size and inherent relational setting) with respect to all previous information extraction tasks faced in Computer Science, they require new techniques to gather this information. Social Network Mining (SNM) is the corresponding research area, aimed at extracting information about the network objects and behavior that cannot be obtained based on the explicit/implicit description of the objects alone, ignoring their explicit/implicit relationships. Statistical Relational Learning (SRL) is a very promising approach to SNM, since it combines expressive representation formalisms, able to model complex relational networks, with statistical methods able to handle uncertainty about objects and relations. This paper is a survey of some SRL formalisms and techniques adopted to solve some SNM tasks.

Metabolomics is increasingly becoming an important field. The fundamental task in this area is to measure and interpret complex time and condition dependent parameters such as the activity or flux of metabolites in cells, their concentration, tissues elements and other biosamples. The careful study of all these elements has led to important insights in the functioning of metabolism. Recently, however, there is a growing interest towards an integrated approach to studying biological systems. This is the main goal in Systems Biology where a combined investigation of several components of a biological system is thought to produce a thorough understanding of such systems. Biological circuits are complex to model and simulate and many efforts are being made to develop models that can handle their intrinsic complexity. A significant part of biological networks still remains unknown even though recent technological developments allow simultaneous acquisition of many metabolite measurements. Metabolic networks are not only structurally complex but behave also in a stochastic fashion. Therefore, it is necessary to express structure and handle uncertainty to construct complete dynamics of these networks. In this paper we describe how stochastic modeling and simulation can be performed in a symbolic-statistical machine learning (ML) framework. We show that symbolic ML deal with structural and relational complexity while statistical ML provides principled approaches to uncertainty modeling. Learning is used to analyze traces of biochemical reactions and model the dynamicity through parameter learning, while inference is used to produce stochastic simulation of the network.

The society model of the last years has given a key role to knowledge in terms of economic and social development. On this background, is allocated the success of the communities of practice, and more recently, of the Complex Learning Community model, whose strength is represented by the ability to provide students, educators, and professionals with a common creative space to develop not only knowledge and expertise, but also ideas, synergies, chances. In both cases, particular emphasis is given to the interactions between users as a "place" where knowledge emerges, is built and delivered. Our former paper was dedicated to the communities of practice, and in particular, to the use of an intelligent agent to control and lead the interaction among users; now our research concerns environments for collaborative learning, where the role of the animator is fundamental, but we take into account a blended animator, supported by specific tools to manage the community. Such tools are necessarily connected to the comprehension and elaboration of the natural language which highlights the appropriate knowledge elements to be used. In this paper we intend to analyse the available techniques in order to identify those that are more suitable for designing the supporting tools required by the blended animator. Particular attention will be given to the domain ontology.

This work extends a SWRL based OWL-S atomic services composition method in order to obtain and manage OWL-S composite services. After the identification of the OWL-S constructs in a SWRL plan, the steps for building the OWL-S control contructs tree, itself serializable with language syntax as well, is given. The obtained composed SWS can be considered as a Simple Process, encoded as a SWRL rule and fed to the SWRL composer for making up new compositions.

Since most content in Digital Libraries and Archives is text, there is an interest in the application of Natural Language Processing (NLP) to extract valuable information from it in order to support various kinds of user activities. Most NLP techniques exploit linguistic resources that are language-specific, costly and error-prone to produce manually, which motivates research for automatic ways to build them. This paper extends the BLA-BLA tool for learning linguistic resources, adding a Grammar Induction feature based on the advanced process mining and management system WoMan. Experimental results are encouraging, envisaging interesting applications to Digital Libraries and motivating further research aimed at extracting an explicit grammar from the learned models.

The problem of implementing socially intelligent agents has been widely investigated in the field of both Embodied Conversational Agents (ECAs) and Social Robots that have the advantage of offering to people the possibility to relate with computer media at a social level. We focus our study on the recognition of the social response of users to embodied agents in the context of ambient intelligence. In this paper we describe how we extended a model for recognizing the social attitude in natural conversation from text by adding two additional knowledge sources: speech and gestures.

The Semantic Web together with Web services technologies enable new scenarios in which the machines use the Web to provide intelligent services in an autonomus way. The orchestration of Semantic Web Services now can be defined from an abstract perspective where their formal semantics can be exploited by software agents to replace human input. This paper tackles the more difficult use case, automatic composition, providing a complete solution to create and manage service processes in a semantically interoperable environment.

The implementation of effective Semantic Web Services (SWS) platforms allowing the composition and, in general, the orchestration of services presents several problems. Some of them are intrinsic within the formalisms adopted to describe SWS, especially when trying to combine the dynamic aspect of SWS effects and the static nature of their ontological representation in Description Logic (DL). This paper proposes a mapping of OWL-S with a DL action formalism in order to evaluate executability and projection by means of the notion of Contexts.

Statistical Relational Learning (SRL) is a growing field in Machine Learning that aims at the integration of logic-based learning approaches with probabilistic graphical models. Markov Logic Networks (MLNs) are one of the state-of-the-art SRL models that combine first-order logic and Markov networks (MNs) by attaching weights to first-order formulas and viewing these as templates for features of MNs. Learning models in SRL consists in learning the structure (logical clauses in MLNs) and the parameters (weights for each clause in MLNs). Structure learning of MLNs is performed by maximizing a likelihood function (or a function thereof) over relational databases and MLNs have been successfully applied to problems in relational and uncertain domains. However, most complex domains are characterized by incomplete data. Until now SRL models have mostly used Expectation-Maximization (EM) for learning statistical parameters under missing values. Multistrategic learning in the relational setting has been a successful approach to dealing with complex problems where multiple inference mechanisms can help solve different subproblems. Abduction is an inference strategy that has been proven useful for completing missing values in observations. In this paper we propose two frameworks for integrating abduction in SRL models. The first tightly integrates logical abduction with structure and parameter learning of MLNs in a single step. During structure search guided by conditional likelihood, clause evaluation is performed by first trying to logically abduce missing values in the data and then by learning optimal pseudo-likelihood parameters using the completed data. The second approach integrates abduction with Structural EM of [17] by performing logical abductive inference in the E-step and then by trying to maximize parameters in the M-step.

The presence of mega-boulders scattered landward along gently sloping rocky coasts is attributed to the impact of tsunami or of exceptional storms. Considering the original position and the size of the largest boulder is possible to estimate the characteristics of the wave that moved them in the past, and then estimate the maximum inundation. The present roughness of the coastal area conditions the capacity of the tsunami inland penetration in case of a future event. The knowledge of the parameters of the possible tsunami together with the coastal topography and roughness make possible to estimate automatically scenarios of probable flooding

Il paper descrive il sistema di graphic matching ICRPad M-Evo, sviluppato con l’obiettivo di consentire agli studiosi di humanities di effettuare ricerche su grandi database di manoscritti storici applicando ai data humanities l’approccio metodologico definito dal “quarto paradigma” del data science (data intensive scientific discovery – Gordon Bell, 2012). Secondo tale approccio, gli algoritmi si sviluppano e applicano per trovare nuove ipotesi di lavoro tramite la scoperta di pattern estratti direttamente da database di grandi dimensioni.

Reaching high precision and recall rates in the results of term-based queries on text collections is becoming more and more crucial, as long as the amount of available documents increases and their quality tends to decrease. In particular, retrieval techniques based on the strict correspondence between terms in the query and terms in the documents miss important and relevant documents where it just happens that the terms selected by their authors are slightly different than those used by the final user that issues the query. Our proposal is to explicitly consider term co-occurrences when building the vector space. Indeed, the presence in a document of different but related terms to those in the query should strengthen the confidence that the document is relevant as well. Missing a query term in a document, but finding several terms strictly related to it, should equally support the hypothesis that the document is actually relevant. The computational perspective that embeds such a relatedness consists in matrix operations that capture direct or indirect term co-occurrence in the collection. We propose two different approaches to enforce such a perspective, and run preliminary experiments on a prototypical implementation, suggesting that this technique is potentially profitable.

Workflow management is fundamental to efficiently, effectively, and economically carry out complex working and domestic activities. Manual engineering of workflow models is a complex, costly, and error-prone task. The WoMan framework for workflow management is based on first-order logic. Its core is an automatic procedure that learns and refines workflow models from observed cases of process execution. Its innovative peculiarities include incrementality (allowing quick learning even in the presence of noise and changed behavior), strict adherence to the observed practices, ability to learn complex conditions for the workflow components, and improved expressive power compared to the state of the art. This paper presents the entire algorithmic apparatus of WoMan, including translation and learning from a standard log format for case representation, import/export of workflow models from/into standard formalisms (Petri nets), and exploitation of the learned models for process simulation and monitoring. Qualitative and quantitative experimental evaluation shows the power and efficiency of WoMan, both in controlled and in real-world domains.