Title: | Deep Neural Networks for Selected Natural Language Processing Tasks |
Authors: | Martínek, Jiří |
Issue Date: | 2019 |
Publisher: | Západočeská univerzita v Plzni |
Document type: | report zpráva |
URI: | http://www.kiv.zcu.cz/cz/vyzkum/publikace/technicke-zpravy/ http://hdl.handle.net/11025/35992 |
Keywords: | zpracování přirozeného jazyka;optické rozpoznávání znaků;kategorizace textu;rozpoznání dialogu |
Keywords in different language: | natural language processing;optical character recognition;text categorization;dialogue act recognition |
Abstract in different language: | This report presents research in several tasks of the natural language processing, namely optical character recognition, text categorization and dialogue act recognition. The report is focused on modern deep neural network classifiers, which are first introduced theoretically with the support of relevant publications and then they are experimentally verified on suitable datasets. For text categorization and dialogue act recognition tasks, multi- and cross-lingual approaches are presented and evaluated. Since the importance of multi-lingual text processing is significantly growing, it is crucial to be able to process information in multiple languages. Unfortunately, there is often a lack of annotated data in some not very widespread languages, so the cross-lingual approaches are requested. Methods for optical character recognition in historical documents are presented with the main emphasis on classifiers which achieve excellent results in natural language processing tasks. Finally, the aims of the future doctoral thesis are set with regard to the recent research. |
Rights: | © Západočeská univerzita v Plzni |
Appears in Collections: | Zprávy / Reports (KIV) |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
Martínek.pdf | Plný text | 3,46 MB | Adobe PDF | View/Open |
Please use this identifier to cite or link to this item:
http://hdl.handle.net/11025/35992
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.