Visual Comparison of NLP Pipelines

Information

• Publication Type: Student Project
• Workgroup(s)/Project(s):
  • Visual Information Foraging on the Desktop
• Date: 2019
• Date (Start): November 2018
• Date (End): April 2019
• Matrikelnummer: 11833154
• First Supervisor: Manuela Waldner

Abstract

Natural Language Processing (NLP) is a sub-field of artificial intelligence (AI). It enables computers to understand, process and analyze large amounts of unstructured natural language data (raw text). Nowadays with the new techniques of machine learning, we got good performance and brings us closer to unfolding the semantic meaning of the text. However, it is far from perfect. Therefore, an alternative approach to helping humans understand a text corpus is to provide a visualization of the content. To generate such a visualization, several NLP steps are necessary to convert the raw text into features, such as weighted keywords or phrases, that can be visualized. The words to be visualized and their weights strongly depend on which NLP steps are performed, in which order, and with which parameters. However, there is currently no standard how to set up such an NLP pipeline and NLP pipeline configurations vary significantly across visualizations and input texts. Our project consists of visualizing high dimensional data with different pre-processing steps with a different order. To compare the results, we choose a well-known and wide-spread overview visualization technique: word clouds. Word clouds are composed of words used in a particular text or subject, in which the size of each word indicates its weight computed in the course of the NLP pipeline.

Additional Files and Images

Weblinks

No further information available.

BibTeX

@studentproject{samoul-2019-cnp,
  title =      "Visual Comparison of NLP Pipelines",
  author =     "Muhammad Samoul",
  year =       "2019",
  abstract =   "Natural Language Processing (NLP) is a sub-field of
               artificial intelligence (AI). It enables computers to
               understand, process and analyze large amounts of
               unstructured natural language data (raw text). Nowadays with
               the new techniques of machine learning, we got good
               performance and brings us closer to unfolding the semantic
               meaning of the text. However, it is far from perfect.
               Therefore, an alternative approach to helping humans
               understand a text corpus is to provide a visualization of
               the content. To generate such a visualization, several NLP
               steps are necessary to convert the raw text into features,
               such as weighted keywords or phrases, that can be
               visualized. The words to be visualized and their weights
               strongly depend on which NLP steps are performed, in which
               order, and with which parameters. However, there is
               currently no standard how to set up such an NLP pipeline and
               NLP pipeline configurations vary significantly across
               visualizations and input texts. Our project consists of
               visualizing high dimensional data with different
               pre-processing steps with a different order. To compare the
               results, we choose a well-known and wide-spread overview
               visualization technique: word clouds. Word clouds are
               composed of words used in a particular text or subject, in
               which the size of each word indicates its weight computed in
               the course of the NLP pipeline.",
  month =      apr,
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2019/samoul-2019-cnp/",
}