Foundations of Intelligent Systems

Histograms are introduced as interesting patterns for data mining. An application of the procedure CF-Miner mining for various types of histograms is described. Possibilities of using domain knowledge in a process of mining interesting histograms are outlined.

The phenotype is the result of a genotype expression in a given environment. Genetic and eventually protein mutations and/or environmental changes may affect the biological homeostasis leading to a pathological status of a normal phenotype. Studying the alterations of the phenotypes on a temporal basis becomes thus relevant and even determinant whether considering the biological re-assortment between the involved organisms and the cyclic nature of the pandemic outbreaks. In this paper, we present a computational solution that analyzes phenotype data in order to capture statistically evident changes emerged over time and track their repeatability. The proposed method adopts a model of analysis based on time-windows and relies on two kinds of patterns, emerging patterns and variability patterns. The first one models the changes in the phenotype detected between time-windows, while the second one models the changes in the phenotype replicated over time-windows. The application to Influenza A virus H1N1 subtype proves the usefulness of our in silico approach.

We continue our research on dynamically adjusted approximate reducts (DAAR). We modify DAAR computation algorithm to take into account dependencies between attribute values in data. We discuss a motivation for this improvement and analyze its performance impact. We also revisit a filtering technique utilizing approximate reducts to create a ranking of attributes according to their relevance. As an illustration we study a data set from AAIA’14 Data Mining Competition.

This paper is about modeling and querying evidential databases. This kind of databases copes with imperfect data which are modeled via the evidence theory. Existing works on such data deal only with the compact form of the database. In this article, we propose a new formalism for modeling and querying evidential databases based on the possible worlds form. This work is a first step toward the definition of a strong representation system.

Aim of the paper is to illustrate a methodology to implement an ubiquitous city platform called Wi-City provided with centralized and mobile Decision Support Systems (DSSs) that take advantage from all the data of city interest including location, social data and data sensed by monitoring devices. The paper proposes to extend the existing Wi-City DSSs based on location intelligence with an advanced version based on multi criteria customer satisfaction expressed by the users grouped by age where the weights of the criteria are provided by the users instead of expert decision makers, and the rating of the aspects involved in the criteria depends on the evaluation expressed by all the service customers. Including such advanced DSSs in Wi-City makes the platform ready to provide information to users of intelligent cities where recommendations should depend on location and collective intelligence.

The issue of devising efficient and effective solutions for supporting the analysis of process logs has recently received great attention from the research community, as effectively accomplishing any business process management task requires understanding the behavior of the processes. In this paper, we propose a new framework supporting the analysis of process logs, exhibiting two main features: a flexible data model (enabling an exhaustive representation of the facets of the business processes that are typically of interest for the analysis) and a graphical query language, providing a user-friendly tool for easily expressing both selection and aggregate queries over the business processes and the activities they are composed of. The framework can be easily and efficiently implemented by leveraging either “traditional” relational DBMSs or “innovative” NoSQL DBMSs, such as Neo4J.

This paper addresses the problem of the incremental construction of an indexing structure, namely a proximity graph, for large image collections. To this purpose, a local update strategy is examined. Considering an existing graph G and a new node q, how only a relevant sub-graph of G can be updated following the insertion of q? For a given proximity graph, we study the most recent algorithm of the literature and highlight its limitations. Then, a method that leverages an edge-based neighbourhood local update strategy to yield an approximate graph is proposed. Using real-world and synthetic data, the proposed algorithm is tested to assess the accuracy of the approximate graphs. The scalability is verified with large image collections, up to one million images.

Recommender systems aim at providing suggestions of interest for end-users. Two main types of approach underlie existing recommender systems: content-based methods and collaborative filtering. In this paper, encouraged by good results obtained in classification by analogical proportion-based techniques, we investigate the possibility of using analogy as the main underlying principle for implementing a prediction algorithm of the collaborative filtering type. The quality of a recommender system can be estimated along diverse dimensions. The accuracy to predict user’s rating for unseen items is clearly an important matter. Still other dimensions like coverage and surprise are also of great interest. In this paper, we describe our implementation and we compare the proposed approach with well-known recommender systems.

The paper presents one of the main modules of HAMIS recommender system built for 34 business companies (clients) involved in heavy equipment repair in the US and Canada. This module is responsible for meta-actions discovery from a large collection of comments, written as text, collected from customers about their satisfaction with services provided by each client. Meta-actions, when executed, trigger action rules discovered from customers data which are in a table format. We specifically focus on the process of mining meta-actions, which consists of four representative and characteristic steps involving sentiment analysis and text summarization. Arranging these four steps in proposed order distinguishes our work from others and better serves our purpose. Compared to procedures presented in other works, each step in our procedure is adapted accordingly with respect to our own observations and knowledge of the domain. Results obtained from the experiments prove the high effectiveness of the proposed approach for mining meta-actions.

This article considers the bipolar approach to define database queries expressing users’ preferences (flexible queries). An algebraic framework for the definition of flexible queries of relational databases using fuzzy bipolar conditions of type and-if-possible and or-else has been considered. This paper defines some qualitative calibrations of such queries to specify a minimal quality of answers and to reduce their number. Different operators (extended $$\alpha $$-cuts) are defined and studied in this article. They can apply on the set of answers to express a qualitative calibrations of bipolar fuzzy queries. Some properties of these extended $$\alpha $$-cuts are pointed out and some of their applications for query evaluation are shown.

Sampling is an important technique for parameter estimation and hypothesis testing widely used in statistical analysis, machine learning and knowledge discovery. Sampling is particularly useful in data mining when the training data set is huge. In this paper, we present a new sampling-based method for learning by Boosting. We show how to utilize the adaptive sampling method in [2] for estimating classifier accuracy in building an efficient ensemble learning method by Boosting. We provide a preliminary theoretical analysis of the proposed sampling-based boosting method. Empirical studies with 4 datasets from UC Irvine ML database show that our method typically uses much smaller sample size (and is thus much more efficient) while maintaining competitive prediction accuracy compared with Watanabe’s sampling-based Boosting learner Madaboost.

Predicate Invention aims at discovering new emerging concepts in a logic theory. Since there is usually a combinatorial explosion of candidate concepts to be invented, only those that are really relevant should be selected, which cannot be done manually due to the huge number of candidates. While purely logical automatic approaches may be too rigid, statistical solutions provide more flexibility in assigning a degree of relevance to the various candidates in order to select the best ones. This paper proposes a new Statistical Relational Learning approach to Predicate Invention. It was implemented and tested on a traditional problem, yielding interesting results.

In the Probabilistic Graphical Model (PGM) community there is an interest around tractable models, i.e., those that can guarantee exact inference even at the price of expressiveness. Structure learning algorithms are interesting tools to automatically infer both these architectures and their parameters from data. Even if the resulting models are efficient at inference time, learning them can be very slow in practice. Here we focus on Cutset Networks (CNets), a recently introduced tractable PGM representing weighted probabilistic model trees with tree-structured models as leaves. CNets have been shown to be easy to learn, and yet fairly accurate. We propose a learning algorithm that aims to improve their average test log-likelihood while preserving efficiency during learning by adopting a random forest approach. We combine more CNets, learned in a generative Bayesian framework, into a generative mixture model. A thorough empirical comparison on real word datasets, against the original learning algorithms extended to our ensembling approach, proves the validity of our approach.

The fusion of imperfect data within the framework of belief functions has been studied by many researchers over the past few years. Up to now, there are some proposed combination rules dealing with dependent information sources. Moreover, the choice of one rule among several alternatives is crucial but the criteria to be based on are still non clear. Thus, in this paper, we evaluate and compare some dependent combination rules for selecting the most efficient one under the framework of classifier fusion.

Concept learning methods for Web ontologies inspired by Inductive Logic Programming and the derived inductive models for class-membership prediction have been shown to offer viable solutions to concept approximation, query answering and ontology completion problems. They generally produce human-comprehensible logic-based models (e.g. terminological decision trees) that can be checked by domain experts. However, one difficulty with these models is their inability to provide a way to measure the degree of uncertainty of the predictions. A framework for inducing terminological decision trees extended with evidential reasoning has been proposed to cope with these problems, but it was observed that the prediction procedure for these models tends to favor cautious predictions. To overcome this limitation, we further improved the algorithms for inducing/predicting with such models. The empirical evaluation shows promising results also in comparison with major related methods.

In this paper we present a new method to measure the degree of dissimilarity of a pair of linear classifiers. This method is based on the cosine similarity between the normal vectors of the hyperplanes of the linear classifiers. A significant advantage of this method is that it has a good interpretation and requires very little information to exchange among datasets. Evaluations on a synthetic dataset, a dataset from the UCI Machine Learning Repository, and facial expression datasets show that our method outperforms previous methods in terms of the normalized mutual information.

Many nearest neighbor based classification approaches require that atypical and mislabeled examples be removed from the training dataset in order to achieve high accuracy. Current editing approaches often remove excessive amount of examples from the training set which does not always lead to optimum accuracy rate. We introduce a new editing method, called HC-edit,—for the k-nearest neighbor classifier —that recognizes areas in the dataset of high purities and removes minority class examples from those regions. The proposed method takes hierarchical clusters with purity information as its input. To edit the data, clusters with purities above a user-defined purity threshold are selected and minority class examples are removed from the selected clusters. Experiments carried out on real datasets using trees generated by traditional agglomerative hierarchical approaches, and trees generated by a supervised taxonomy method which incorporates class label information in the clustering process show that the new approach leads to improved accuracy and does well in comparison to other editing methods.

Probabilistic topic models based on Latent Dirichlet Allocation (LDA) are increasingly used to discover hidden structure behind big text corpora. Although topic models are extremely useful tools for exploring and summarizing large text collections, most of time the inferred topics are not easy to understand and interpret by human. In addition, some inferred topics may be described by words that are not much relevant to each other and are thus considered low quality topics. In this paper, we propose a novel method that not only assigns a label to each topic but also identifies low quality topics by providing a reliability score for the label of each topic. Our rationale is that a topic labeling method cannot provide a good label for a low quality topic, and thus predicting label reliability is as important as topic labeling itself. We propose a novel measure (Ontology-Based Coherence) that can assess coherence of topics with respect to an ontology structure effectively. Empirical results on a real dataset and our user study show that the proposed predictive model using the defined measures can predict the label reliability better than two alternative methods.

Twitter feeds provide data scientists with a large repository for entity based sentiment analysis. Specifically, the tweets of individual users may be used in order to track the ebb and flow of their sentiments and opinions. However, this domain poses a challenge for traditional classifiers, since the vast majority of tweets are unlabeled. Further, tweets arrive at high speeds and in very large volumes. They are also suspect to change over time (so-called concept drift). In this paper, we present the PyStream algorithm that addresses these issues. Our method starts with a small annotated training set and bootstraps the learning process. We employ online analytic processing (OLAP) to aggregate the opinions of the individuals we track, expressed in terms of the votes they would cast in a national election. Our results indicate that we are able to capture the sentiments of individuals as they evolve over time.

Previous works on Polish sentiment dictionaries revealed the superiority of machine learning on vectors created from word contexts (concordances or word co-occurrence distributions), especially compared to the SO-PMI method (semantic orientation of pointwise mutual information). This paper demonstrates that this state-of-the-art method could be improved upon when extending the vectors by word embeddings, obtained from skip-gram language models. Specifically, it proposes a new method of computing word sentiment polarity using feature sets composed of vectors created from word embeddings and word co-occurrence distributions. The new technique is evaluated in a number of experimental settings.

The cube of opposition is a logical structure that underlies many information representation settings. When applied to multiple criteria decision, it displays various possible aggregation attitudes. Situations are usually assessed by combinations of properties they satisfy, but also by combinations of properties they do not satisfy. The cube of opposition applies to qualitative evaluation when criteria are weighted as well as in the general case where any subset of criteria may be weighted for expressing synergies between them, as for Sugeno integrals. Sugeno integrals are well-known as a powerful qualitative aggregation tool which takes into account positive synergies between properties. When there are negative synergies between properties we can use the so-called desintegral associated to the Sugeno integral. The paper investigates the use of the cube of opposition and of the if-then rules extracted from these integrals and desintegrals in order to better describe acceptable situations.

OpenStreetMap (OSM) is a crowd source geographical database that gives users a wide range of tools for searching and locating points of interest and to support the user in navigation on a map. This paper proposes to define a query language for OSM specifying user requests about the route to select between a source and a destination. To this purpose we use Uppaal (http://www.uppaal.org/) model checker: the user poses her query specifying desired points of interest via temporal logic. the method model checks the negation of the desired property, whose counterexample will retrieve the desired path.

Efficient management of big Rule-Based Systems constitutes an important challenge for Knowledge Engineering. This paper presents an approach based on Granular Sets and Granular Relations. Granules of data replace numerous low-level items and allow for concise definition of constraints over a single attribute. Granular Relations are used for specification of preconditions of rules. A single Granular Rule can replace numerous rules with atomic preconditions. By analogy to Relational Databases, a complete Granular Rule Frame consists of Rule Scheme and Rule Specification. Such approach allows for efficient and concise specification of powerful rules at the conceptual level and makes analysis of rule set easier. The detailed specifications of Granular Rules are much more concise than in the case of atomic attribute values, but still allow for incorporating all necessary details.

Ontology alignment process is seen as a key mechanism for reducing heterogeneity and linking the diverse data and ontologies arising in the Semantic Web. In such large infrastructure, it is inconceivable to assume that all ontologies dealing with a particular knowledge domain are aligned in pairs. Furthermore, the high performance of the alignment techniques is closely related to two major factors, i.e., time consumption and resource machine limitations. Indeed, good quality alignments are valuable and it would be appropriate to harness. This paper introduces a new indirect ontology alignment method. The proposed method implements a strategy of indirect ontology alignment based on a smart direct alignments composition and reuse. Results obtained after extensive carried experiments are very encouraging and highlight many useful insights about the new proposed method.

This paper proposes an interactive approach that guides users in the step-by-step composition of services by providing safe suggestions based on type convertibility. Users specify the points of the workflow (called the focus) they want to complete, and our approach suggests services and connections whose data types are compatible with the focus. We prove the safeness (every step produces a well-formed workflow) and the completeness (every well-formed workflow can be built) of our approach.

We propose MUSETS (multi-session total shortening) – a novel formulation of the query suggestion task, specified as an optimization problem. Given an ambiguous user query, the goal is to propose the user a set of query suggestions that optimizes a diversity-aware objective function. The function models the expected number of query reformulations that a user would save until reaching a satisfactory query formulation. The function is diversity-aware, as it naturally enforces high coverage of different alternative continuations of the user session. For modeling the topics covered by the queries, we also use an extended query representation based on entities extracted from Wikipedia. We apply a machine learning approach to learn the model on a set of user sessions to be subsequently used for queries that are under-represented in historical query logs and present an evaluation of the approach.

In this paper we define an extension of the description logic $$\mathcal {SROIQ}$$ based on a preferential semantics to introduce a notion of typicality in the language which allows defeasible inclusions to be represented in a knowledge base. We define a polynomial encoding of the resulting language into $$\mathcal {SROIQ}$$, thus showing that reasoning in the preferential extension of $$\mathcal {SROIQ}$$ has the same complexity as reasoning in $$\mathcal {SROIQ}$$.

Similar entity search by example is an important task in the area of retrieving information from semantic knowledge bases. In this paper we define a new interactive variant of this problem that is called iQbees for “Interactive Query-by-Example Entity Search” and is an extension of a previous QBEES approach. We also present a working on-line prototype demo which implements the proposed approach.

This paper presents an analysis of the effect of features obtained from 3 different audio analysis tools on classifier accuracy during emotion detection. The research process included constructing training data, feature extraction, feature selection, and building classifiers. The obtained results indicated leaders among feature extraction tools used during classifier building for each emotion. An additional result of the conducted research was obtaining information on which features are useful in the detection of particular emotions.

Several studies report successful results on how social assistive robots can be employed as interface in the assisted living domain. In this domain, a natural way to interact with robots is to use a speech. However, humans often use particular intonation in the voice that can change the meaning of the sentence. For this reason, a social assistive robot should have the capability to recognize the intended meaning of the utterance by reasoning on the combination of linguistic and acoustic analysis of the spoken sentence to really understand the user’s feedback. We developed a probabilistic model that is able to infer the intended meaning of the spoken sentence from the analysis of its linguistic content and from the output of a classifier able to recognise the valence and arousal of the speech prosody starting from dataset. The results showed that reasoning on the combination of the linguistic content with acoustic features of the spoken sentence was better than using only the linguistic component.

Shifts in emotions towards given topics on social media are often related to momentous real world events, and for the researcher or journalist, such changes may be the first observable sign that something interesting is going on. Further research on why a topic t suddenly has become, say, more or less popular, may involve searching for topics $$t'$$ whose co-occurrence with t have increased significantly together with the change in emotion. We hypothesize that $$t'$$ and its increasing relationship to t may relate to a contributing cause why the attitude towards t is changing. A method and tool is presented that monitors a stream of messages, reporting topics with changing emotions and indicating explanations by means of related topics whose increasing occurrence are taken as possible clues of why the change did happen.

We present an extension to knowledge discovery in Music Information Retrieval (MIR) databases and the emotional indices associated with (i) various scalar theory, and (ii) correlative behavioral demographics. Certain societal demographics are set in their ways as to how they dress, behave in society, solve problems and deal with anger and other emotional states. It is also well documented that particular musical scales evoke particular states of emotion and personalities of their own. This paper extends the work that Knowledge Discovery in Databases (KDD) and Rough Set Theory has opened in terms of mathematically linking music scalar theory to emotions. We now, extend the paradigm by associating emotions, based from music, to societal demographics and how strong these relationships to music are as to affect, if at all, how one may dress, behave in society, solve problems and deal with anger and other emotional states.

The network sciences have provided significant strides for understanding complex systems. Those systems are represented by graphs. One of the most relevant features of graphs representing real systems is clustering, or community structure. The communities are clusters (groups) of nodes, with more edges connecting to nodes of the same cluster and comparatively fewer edges connecting to nodes of different clusters. It can be considered as independent compartments of a graph. There are two possible sources of information we can use for the community detection: the network structure, and the attributes and features of nodes. In this paper, we use the features of nodes to detect communities. There are nodes in network that are more able and susceptible to diffuse information and propagate influence. The main purpose of our approach is to find leader nodes of networks and to form community around those nodes. Unlike to most existing researches studies, the proposed algorithm doesn’t require a priori knowledge of k number of communities to be detected.

We address the problem of finding local patterns and related local knowledge, represented as implication rules, in an attributed graph. Our approach consists in extending frequent closed pattern mining to the case in which the set of objects is the set of vertices of a graph, typically representing a social network. We recall the definition of abstract closed patterns, obtained by restricting the support set of an attribute pattern to vertices satisfying some connectivity constraint, and propose a specificity measure of abstract closed patterns together with an informativity measure of the associated abstract implication rules. We define in the same way local closed patterns, i.e. maximal attribute patterns each associated to a connected component of the subgraph induced by the support set of some pattern, and also define specificity of local closed patterns together with informativity of associated local implication rules. We also show how, by considering a derived graph, we may apply the same ideas to the discovery of local patterns and local implication rules in non disjoint parts of a subgraph as k-cliques communities.

This paper proposes an architecture for agents that are in charge of handling a given environment in an Ambient Intelligence context, ensuring suitable contextualized and personalized support to the user’s actions, adaptivity to the user’s peculiarities and to changes over time, and automated management of the environment itself. Functionality involves multi-strategy reasoning and learning, workflow management and service composition. In Multi-Agent Systems, different types of agents may implement different parts of this architecture.

The decision-making process in stock investment requires not only the rational trading rules practices, but also faith that market information is reliable. A trust metric is an indication of the degree to which one social actor trusts another. In our Agent-Based Model we built a stock-trading model that issues a daily stock trading signal. This paper introduces an agent-based model for finding the optimal degree of trust in stock transactions. The system has been evaluated in the context of Bank of America stock in the period of 1987–2014. The model outperformed both S&P 500 and buy-and-hold Bank of America stock strategy by two to three times.

Nowadays almost everybody spends a lot of time commuting and traveling, so we are all very much interested in smooth use of various roads. Also governing bodies are concerned to assure efficient exploitation of the transportation system. The European Union announced a directive on Intelligent Transport Systems in 2010, to ensure that systems integrating information technology with transport engineering are deployed within the Union. In this paper we address automatic classification of vehicle type, based on audio signals only. Hierarchical classification of vehicles is applied, using decision trees, random forests, artificial neural networks, and support vector machines. A dedicated feature set is proposed, based on spectral ranges best separating the target classes. We show that longer analyzing frames yield better results, and a set of binary classifiers performs better than a single multi-class classifier.

In this paper we present a system for detecting shadows in dynamic indoor and outdoor environment. The algorithm we propose fuses together color and stereo disparity information. Some considerations on the nature of the shadow improves the algorithm’s ability to candidate the pixels as shadow or foreground. The candidate of both color and disparity information are then weighted by analyzing the effectiveness in the scene. The techniques employed allows separate computation and multithreading operations.

Interoperability of annotations in different domains is an essential demand to facilitate the interchange of data between semantic applications. Foundational ontologies, such as SKOS (Simple Knowledge Organization System), play an important role in creating an interoperable layer for annotation. We are proposing a multi-layer ontology schema, named SKOS-Wiki, which extends SKOS to create an annotation model and relies on the semantic structure of the Wikipedia. We also inherit the DBpedia definition of named entities. The main goal of our proposed extension is to fill the semantic gaps between these models to create a unified annotation schema.

This paper addresses issues associated with the real-time control of public transit operations to minimize passenger wait time: namely vehicle headway, maintenance of passenger comfort, and reducing the impact of control strategies. The randomness of passenger arrivals at bus stops and external factors (such as traffic congestion and bad weather) in high frequency transit operations often cause irregular headway that can result in decreased service reliability. The approach proposed in this paper, which has the capability of handling the uncertainty of transit operations based on Multi-objective evolutionary algorithm using a dynamic Bayesian network, applies preventive strategies to forestall bus unreliability and, where unreliability is evident, restore reliability using corrective strategies. “Holding”, “expressing”, “short-turning” and “deadheading” are the corrective strategies considered in this paper.

Facade-layout synthesis is a combinatorial problem that arises when insulating buildings with rectangular parameterizable panels. At the core of the problem lies the assignment of size to an unknown number of panels and their arrangement over a rectangular facade surface. The purpose of this communication is to give an overview of the facade-layout synthesis problem and its reasoning by constraint satisfaction problems. Then, we show the combinatorial characteristics of the problem, its modeling by means of constraint satisfaction and a decision support system that solves the problem using several constraint-based algorithms.

During deep brain stimulation (DBS) surgery for Parkinson disease, the target is the subthalamic nucleus (STN). STN is small, (9$$\,\times \,$$7 x 4 mm) and typically localized by a series of parallel microelectrodes. As those electrodes are in steps advanced towards and through the STN, they record the neurobiological activity of the surrounding tissues. By careful inspection and analysis of such recordings one can obtain a range of depth at which given electrodes passed through the STN. Both human made inspection and computer based analysis are performed during surgery in the environment of the operation theatre. There are several methods for the STN detection, one of them – developed by the authors – is described in [8]. While the detection of the STN interior can be obtained with good certainty its borders can be slightly fuzzy and sometimes it is difficult to classify whether given depth should be regarded as belonging to the STN proper or lying outside of it. Mapping of the borders is important as the tip of the final permanent stimulating electrode is often placed near the ventral (bottom) border of the STN [12]. In this paper we are showing that analysis focusing on narrow frequency bands can yield better discrimination of the STN borders and STN itself.

In some real world applications like robotics, manufacturing, the same planning operator with single or multiple effects is instantiated to several objects. This is quite different from performing the same action (or plan) several times. In this paper we give an approach to construct an iterated form of these operators (actions). We call such actions iterated actions that are performed on sets of objects. In order to give a compact and formal specification of such actions, we define a new type of predicate called set predicate. We show that iterated actions on sets behave like classical planning actions. Thus any classical planner can be used to synthesize plans containing iterated actions. We formally prove the correctness of this approach. An implementable description of iterated actions is given in PDDL for an example domain. The implementations were carried out using the state-of-the-art BlackBox planner.

Pushing is often used by robots as a simple way to manipulate the environment and has in the past been well studied from kinematic and numerical perspective. The paper proposes a qualitative approach to pushing convex polygonal objects by a simple wheeled robot through a single point contact. We show that by using qualitative reasoning, pushing dynamics can be described in concise and intuitive manner, that is still sufficient to control the robot to successfully manipulate objects. Using the QUIN program on numerical data collected by our robot while experimentally pushing objects of various shapes, we induce a model of pushing. This model is then used by our planning algorithm to push objects of previously unused shapes to given goal configurations. The produced trajectories are compared to smooth geometric solutions. Results show the correctness of our qualitative model of pushing and efficiency of the planning algorithm.

This paper outlines first issues related to music genre classification and a short description of algorithms used for musical instrument separation. Also, the paper presents proposed optimization of the feature vectors used for music genre recognition. Then, the ability of decision algorithms to properly recognize music genres is discussed based on two databases. In addition, results are cited for another database with regard to the efficiency of the feature vector.

Parallel sentences are a relatively scarce but extremely useful resource for many applications including cross-lingual retrieval and statistical machine translation. This research explores our new methodologies for mining such data from previously obtained comparable corpora. The task is highly practical since non-parallel multilingual data exist in far greater quantities than parallel corpora, but parallel sentences are a much more useful resource. Here we propose a web crawling method for building subject-aligned comparable corpora from e.g. Wikipedia dumps and Euronews web page. The improvements in machine translation are shown on Polish-English language pair for various text domains. We also tested another method of building parallel corpora based on comparable corpora data. It lets automatically broad existing corpus of sentences from subject of corpora based on analogies between them.

Knowledge discovery from texts, particularly the identification of spatial information is a difficult task due to the complexity of texts written in natural language. Here we propose a method combining two statistical approaches (lexical and contextual analysis) and a text mining approach to automatically identify types of spatial relations. Experiments conducted on an English corpus are presented.

In this paper, we investigate the issue of modeling corporate entities’ online reputation. We introduce a bayesian latent probabilistic model approach for e-Reputation analysis based on Dimensions (Reputational Concepts) Categorization and Opinion Mining from textual content. Dimensions to analyze e-Reputation are set up by analyst as latent variables. Machine Learning (ML) Natural Language Processing (NLP) approaches are used to label large sets of text passages. For each Dimension, several estimations of the relationship with each text passage are computed as well as Opinion and Priority. The proposed automatic path modeling algorithm explains Opinion or Priority scores based on selected Dimensions. Model Robustness’ is evaluated over RepLab dataset.

This paper proposes a novel approach in incorporating several metadata such as citations, co-authorship, titles, and keywords to identify real authors in author disambiguation task. Classification schemes make use of these variables to identify authorship. The methodology performed in this paper is: (1) coarse grouping of article by the use of focus names, (2) Applying a model using paper metadata to identify same authorship, and (3) separate the true authors having the same focus name.