Wojciech Jaworski

Rule Induction: Combining Rough Set and Statistical Approaches

In this paper we propose the hybridisation of the rough set concepts and statistical learning theory. We introduce new estimators for rule accuracy and coverage, which base on the assumptions of the statistical learning theory. Then we construct classifier which uses these estimators for rule induction. These estimators allow us to select rules describing statistically significant dependencies in data. We test our classifier on benchmark datasets and show its applications for KDD.

Syntactic Approximation of Semantic Roles

The aim of this paper is to propose a method of simulating - in a syntactico-semantic parser - the behaviour of semantic roles in case of a language that has no resources such as VerbNet of FrameNet, but has relatively rich morphosyntax (here: Polish). We argue that using an approximation of semantic roles derived from syntactic (grammatical functions) and morphosyntactic (grammatical cases) features of arguments may be beneficial for applications such as text entailment.

Generalized indiscernibility relations: applications for missing values and analysis of structural objects

In this paper, we discuss an approach to structural objects based on a generalisation of indiscernibility relation used in rough set theory. The existing results in rough set theory are based on the assumption that objects are perceived by attribute value vectors. We propose the new point of view on rough set theory. We replace information systems with the knowledge representation models that incorporate information relative to the structure of objects.We redefine the indiscernibility relation as a relation on objects characterised by some axioms. Such a definition can be naturally applied for information systems with missing values and multivalued attributes. We extend the approach on structural objects. We introduce the meaning representation language for expressing properties of structural objects and we show how to select relevant formulae from this language for sequential data.

Learning compound decision functions for sequential data in dialog with experts

In this paper, we investigate the problem of learning the decision functions for sequential data describing complex objects that are composed of subobjects. The decision function maps sequence of attribute values into a relational structure, representing properties of the object described by the sequence. This relational structure is constructed in a way that allows us to answer questions from a given language. The decision function is constructed by means of rule system. The rules are learned incrementally in a dialog with an expert. We also present an algorithm that implements the rule system and we apply it to real life problems.

Model selection and assessment for classification using validation

We address the problem of determination of the size of the test set which can can guarantee statistically significant results in classifier error estimation and in selection of the best classifier from a given set. We focus on the case of the 0-1 valued loss function and we provide one and two sides optimal bounds for Validation (known also as Hold-Out Estimate and Train-and-Test Method). We also calculate the smallest sample size, necessary for obtaining the bound for given estimation accuracy and reliability of estimation, and we present the results in tables. Finally, we propose strategies for classifier design using the bounds derived.

Semantic Roles in Grammar Engineering

The aim of this paper is to discuss difficulties involved in adopting an existing system of semantic roles in a grammar engineering task. Two typical repertoires of semantic roles are considered, namely, VerbNet and Sowa’s system. We report on experiments showing the low inter-annotator agreement when using such systems and suggest that, at least in case of languages with rich morphosyntax, an approximation of semantic roles derived from syntactic (grammatical functions) and morphosyntactic (grammatical cases) features of arguments may actually be beneficial for applications such as textual entailment.

A note on the Regularization Algorithm

Regularization Algorithm (also called Regularization Network) is a technique for solving problems of learning from examples – in particular, the problem of approximating a multivariate function from sparse data. We analyze behavior of Regularization Algorithm for regularizator parameter equal to zero. We propose an approximative version of algorithm in order to overcome the computational cost for large data sets. We give proof of convergence and estimation for error of approximation.

APPLICATION OF LINGUISTIC CUES IN THE ANALYSIS OF LANGUAGE OF HATE GROUPS

Hate speech and fringe ideologies are social phenomena that thrive on-line. Members of the political and religious fringe are able to propagate their ideas via the Internet with less eort than in traditional media. In this article, we attempt to use linguistic cues such as the occurrence of certain parts of speech in order to distinguish the language of fringe groups from strictly informative sources. The aim of this research is to provide a preliminary model for iden- tifying deceptive materials online. Examples of these would include aggressive marketing and hate speech. For the sake of this paper, we aim to focus on the political aspect. Our research has shown that information about sentence length and the occurrence of adjectives and adverbs can provide information for the identification of dierences between the language of fringe political groups and mainstream media.

Hybridization of rough sets and statistical learning theory

In this paper we propose the hybridization of the rough set concepts and statistical learning theory. We introduce new estimators for rule accuracy and coverage, which base on the assumptions of the statistical learning theory. These estimators allow us to select rules describing statistically significant dependencies in data. Then we construct classifier which uses these estimators for rule induction. In order to make our solution applicable for information systems with missing values and multiple valued attributes, we propose axiomatic representation of information systems and we redefine the indiscernibility relation as a relation on objects characterized by axioms. Finally, we test our classifier on benchmark datasets.

Identifying Web Users on the Base of their Browsing Patterns

Our aim is to develop methodology for recognition of Internet portal users on the base of their browsing patterns. This is a classification task in which we have thousands values of decision attribute and objects are described by means of sequences of symbols. We develop feature selectors which make our task tractable. Since the behaviour usually does not distinguish users, we introduce user profiles which are clusters of indiscernible users. Then, we construct classifiers which assign descriptions of user behaviour to user profiles. We also derive quality measures specific for our task.