The Naïve Bayes Model for Unsupervised Word Sense Disambiguation

Aspects Concerning Feature Selection

verfasst von: Florentina T. Hristea

Verlag: Springer Berlin Heidelberg

Buchreihe : SpringerBriefs in Statistics

Enthalten in: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Einloggen, um Zugang zu erhalten

Über dieses Buch

This book presents recent advances (from 2008 to 2012) concerning use of the Naïve Bayes model in unsupervised word sense disambiguation (WSD).

While WSD, in general, has a number of important applications in various fields of artificial intelligence (information retrieval, text processing, machine translation, message understanding, man-machine communication etc.), unsupervised WSD is considered important because it is language-independent and does not require previously annotated corpora. The Naïve Bayes model has been widely used in supervised WSD, but its use in unsupervised WSD has led to more modest disambiguation results and has been less frequent. It seems that the potential of this statistical model with respect to unsupervised WSD continues to remain insufficiently explored.

The present book contends that the Naïve Bayes model needs to be fed knowledge in order to perform well as a clustering technique for unsupervised WSD and examines three entirely different sources of such knowledge for feature selection: WordNet, dependency relations and web N-grams. WSD with an underlying Naïve Bayes model is ultimately positioned on the border between unsupervised and knowledge-based techniques. The benefits of feeding knowledge (of various natures) to a knowledge-lean algorithm for unsupervised WSD that uses the Naïve Bayes model as clustering technique are clearly highlighted. The discussion shows that the Naïve Bayes model still holds promise for the open problem of unsupervised WSD.

Inhaltsverzeichnis

Frontmatter

Chapter 1. Preliminaries

Abstract

This chapter describes the problem we are investigating and trying to solve in all other chapters. It introduces word sense disambiguation (WSD) and Naïve Bayes-based WSD, as well as local type features for unsupervised WSD with an underlying Naïve Bayes model.

Florentina T. Hristea

Chapter 2. The Naïve Bayes Model in the Context of Word Sense Disambiguation

Abstract

This chapter discusses the Naïve Bayes model strictly in the context of word sense disambiguation. The theoretical model is presented and its implementation is discussed. Special attention is paid to parameter estimation and to feature selection, the two main issues of the model’s implementation. The EM algorithm is recommended as suitable for parameter estimation in the case of unsupervised WSD. Feature selection will be surveyed in the following chapters.

Florentina T. Hristea

Chapter 3. Semantic WordNet-Based Feature Selection

Abstract

The feature selection method we are presenting in this chapter makes use of the semantic network WordNet as knowledge source for feature selection. The method makes ample use of the WordNet semantic relations which are typical of each part of speech, thus placing the disambiguation process at the border between unsupervised and knowledge-based techniques. Test results corresponding to the main parts of speech (nouns, adjectives, verbs) will be compared to previously existing disambiguation results, obtained when performing a completely different type of feature selection. Our main conclusion will be that the Naïve Bayes model reacts well in the presence of semantic knowledge provided by WN-based feature selection when acting as clustering technique for unsupervised WSD.

Florentina T. Hristea

Chapter 4. Syntactic Dependency-Based Feature Selection

Abstract

The feature selection method we are presenting in this chapter makes use of syntactic knowledge provided by dependency relations. Dependency-based feature selection for the Naïve Bayes model is examined and exemplified in the case of adjectives. Performing this type of knowledge-based feature selection places the disambiguation process at the border between unsupervised and knowledge-based techniques. The discussed type of feature selection and corresponding disambiguation method will once again prove that a basic, simple knowledge-lean disambiguation algorithm, hereby represented by the Naïve Bayes model, can perform quite well when provided knowledge in an appropriate way. Our main conclusion will be that the Naïve Bayes model reacts well in the presence of syntactic knowledge of this type and that dependency-based feature selection for the Naïve Bayes model is a reliable alternative to the WordNet-based semantic one.

Florentina T. Hristea

Chapter 5. N-Gram Features for Unsupervised WSD with an Underlying Naïve Bayes Model

Abstract

The feature selection method we are presenting in this chapter relies on web scale N-gram counts. It uses counts collected from the web in order to rank candidates. Features are thus created from unlabeled data, a strategy which is part of a growing trend in natural language processing. Disambiguation results obtained by web N-gram feature selection will be compared to those of previous approaches that equally rely on an underlying Naïve Bayes model but on completely different feature sets. Test results corresponding to the main parts of speech (nouns, adjectives, verbs) will show that web N-gram feature selection for the Naïve Bayes model is a reliable alternative to other existing approaches, provided that a “quality list” of features, adapted to the part of speech, is used.

Florentina T. Hristea

Backmatter

Titel: The Naïve Bayes Model for Unsupervised Word Sense Disambiguation
verfasst von: Florentina T. Hristea
Verlag: Springer Berlin Heidelberg
Electronic ISBN: 978-3-642-33693-5
Print ISBN: 978-3-642-33692-8
DOI: https://doi.org/10.1007/978-3-642-33693-5