National Technical University of Athens (NTUA)
Σημασιολογική απάντηση ερωτημάτων στον παγκόσμιο ιστό
Abstract
dc:descriptionWith the rapid growth of the World Wide Web (WWW) the management of information becomes more difficult. Day by day, millions of new pages are added to the already existing ones, which in turn, are constantly updated, increasing the volume of available information. The volume has a negative impact on the exploitation of available information as the users find it difficult. Moreover, search becomes more difficult because of the fact that the information is heterogeneous and written in human language. The existence of metadata, which will illustrate the data and make the content of the information more clear, is a condition that would be particularly helpful. Therefore, there is an enormous need to develop techniques and built systems that will be able to search, classify and categorize the available information and, automatically generate metadata whenever it is possible. The second step deals with the number of those metadata so as to be possible for the computers to process. As a solution to this problem there have been proposed question answering systems, which take advantage of the existing metadata of documents in order to improve performance. The contribution of this thesis is to study and develop question answering systems. Moreover, specific problems are studied, that would help comprehension of semantic documents by search engines, so as to achieve better information extraction.Question answering systems are essentially systems which take as input queries and a set of documents, in which they search for answers. The question is usually given in natural language and searches for answers are related to a particular knowledge domain. When a human is reading from the web a text written in Natural Language, he/she is able to clearly distinguish the boundaries of words, their meaning and can easily understand which is the object or abstract concept to which they refer. However, this procedure is not very obvious for a computer system. So, text processing and representation techniques will be needed so as to bridge the gap between the language understood by the world of machines and the human world language. For this reason, various text processing techniques have been developed over the past years. These techniques are moving in both directions. The first direction regards the recognition of the limits of a word or phrase in the text. This phrase contains a conceptual content and represents a physical object, or an abstract concept, in Natural Language. Of course, the computer can identify that a particular phrase represents an object by using these techniques, but it is still not able to conceptually distinguish the subject by itself. Therefore, we need to add some kind of external knowledge, which is similar to the experience that a human gains. In order to be stored, this knowledge needs to be modeled in a way that can be easy understood and readily processed by a computer.In the context of this thesis, algorithms and methods to recognize the semantics of the phrases and texts are developed. The meaning of semantics is that phrases and texts are mapped to clearly defined objects in various knowledge databases and ontologies. This semantic annotation enables machines to understand the semantics of the text and to improve the searching results.The first part of this thesis is dedicated to the presentation of a question answering system. This system is to some extent based on the identification of the semantics of the documents, from which the information is to be exported. The system is based on the combination of classical information retrieval methods with external semantic information stored in knowledge bases. First of all, a search engine is used to gather the web pages and, then, the sentences that are candidates to be answers to the question are extracted. The phrases are ranked by using a linear combination of various retrieval metrics in order to save the most important. Then, the web page in Wikipedia for each of these phrases is retrieved. This is done to enable exploitation of structured information, which exists in the structure of the Wikipedia page. Then, a novel method for the exploitation of this information is proposed. Using this method the best fitting phrases on the semantics of the query are selected as an answer.Beyond the recognition of the semantics of the text words, the semantic enrichment of whole documents is equally useful in question answering systems. The prior knowledge of the semantic information which is contained in each document helps for the better management of the documents. In this context, a series of algorithms that enrich the texts with metadata is presented, which contain semantic information, by using supervised learning techniques. Moreover, in this part of the thesis, we study the problem of exploiting document collections with unbalanced data regarding an ontology.In conclusion, the contribution of this thesis lies in studying and trying to bridge the gap between natural language and the Semantic Web. Based on this, a study is carried out about Question Answering by combining natural language processing techniques and semantic technologies. The questions, which are usually given by people in Natural Language, may have more than one answer. In addition, we present a new way of document representation, which helps classification and the Question Answering. Then, a study about the identification of the meaning of the words that exist in the texts in natural language and are usually part of the answer is presented. Finally, another study on the semantic enrichment of documents, which are based on ontologies, is presented, in order to exploit the advantages of the semantic web in terms of speed and organization of information.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Stratogiannis, Georgios
- Στρατογιάννης, Γεώργιος
Subjects
dc:subject × 16- Απάντηση ερωτημάτων
- Σύνδεση οντοτήτων
- Παγκόσμιος ιστός
- Σημασιολογικός ιστός
- Διανυσματική αναπαράσταση
- Σημασιολογικός εμπλουτισμός εγγράφων
- Question answering
- Entity linking
- World wide web
- Semantic web
- Vector representation
- Document semantic enrichment
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
dc:rights- Statement dc:rights
-
- BY
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/39761