Modelling and classification of large textual data

Bc. Vyacheslav Gatin

Master's thesis

Abstract:

The amount of data that the world creates has been increasing in geometrical progression (IBM, 2016), while textual data represents a big part of it. Manipulation with and classification of textual data are dealt with Natural Language Processing field of machine learning. This thesis explores the paradigm of machine learning, as being a key technique in modern data analysis and manipulation. Research introduces the theory behind machine learning by defining its structure and processes, problems that it deals with, and its subfields. Following the purposes of the study, the research then continues to define specific machine learning algorithms best suited for Natural Language Processing tasks. During the practical part of the work, author deals with a semi-structured dataset of more than 260 000 objects, which is a collection of professional vehicles placed on an online platform for remarketing purposes. Practical work consists of creation of appropriate classification algorithm and further improvement of its quality through exploration of unclassified data using JavaScript programming language and word2vec deep learning technique. Author has achieved the initial coverage of 80% while keeping the accuracy close to 100%. Upon improvement, the coverage of the model increased to 85%, without significant drops to accuracy. Recommendations for further improvement of the model are then provided.

Keywords

Big data, unstructured textual data, machine learning, deep learning, natural language processing, word2vec, data modelling

Jazyk práce: English

Datum vytvoření / odevzdání či podání práce: 29. 11. 2018

Obhajoba závěrečné práce

• Vedúci: Ing. Tomáš Hlavsa, Ph.D.

Czech University of Life Sciences Prague

Faculty of Economics and Management

Master programme / odbor:
Economics and Management / Economics and Management