Topic-Modeling- and Subject-Classification-Analyses of Articles from the EURASIP Journal on Advances in Signal Processing
Loading...
Files
Date
2019-11-26
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Description
This data set contains the results of topic-modeling- and subject-
classification-analyses of the abstracts of 87 articles from the EURASIP
Journal on Advances in Signal Processing (ISSN: 1687-6180). All of the
selected articles had in common that they were assigned the keyword “OFDM”
(Orthogonal Frequency-Division Multiplexing) by the authors or the publisher.
The topic modeling analyses were carried out with the program GibbsLDA++
(<http://gibbslda.sourceforge.net>) once with and once without stemming
(model-final.twords_w_stemming.txt and model-final.twords_wo_stemming.txt,
respectively). The program parameters were set to: src/lda -est -alpha 0.5
-beta 0.1 -ntopics 10 -niters 1000 -savestep 100 -twords 20
The subject classification analyses were carried out with the web-application
Annif.org (<http://annif.org/>), which offers different algorithms for the
classification. The following algorithms were used (the name of the
corresponding result file is given in brackets): Annif prototype API English
(Annif.png), fastText English (fastText.png), Maui English (Maui.png), TF-IDF
English (TF-IDF.png), YSO ensemble English (YSO.png).
A list with the DOIs of the articles can be found in the file
"DOIs_analyzed_articles.txt" and the analyzed abstracts of these articles in
the zip archive "Abstracts_EURASIPJAdvSignalProcess.zip".
_
We noticed that at least 7 of the 87 abstracts analyzed had been incomplete in
the first version of our data set. These abstracts had in common, that they
contained numbers that were embedded as inline-graphics within an inline-
formula-element. Due to an error in the preprocessing of the texts, the
abstracts were cut off behind these elements. For this reason, we repeated the
analysis with the complete abstracts. Please notice, that the graphical
numbers themselves are still omitted, so that "complete" only refers to the
plain text.
DOIs of the articles that were affected:
* 10.1155/S1110865704403102
* 10.1155/S1110865704401140
* 10.1155/S1110865704311054
* 10.1155/S1110865703309060
* 10.1155/S1110865702000884
* 10.1155/ASP.2005.2730
* 10.1155/ASP.2005.525
_
Citation
Endorsement
Project(s)
Faculty
Collections
License
Except where otherwise noted, this license is described as CC BY 4.0 - Attribution 4.0 International
Version History
You are currently viewing version no. 5 of the item. This is the most recent version.
1 - 2 of 2
Version | Date | Summary |
---|---|---|
5* | 2019-11-26 13:11:16 | We repeated the TDM analyses because we had noticed that at least 7 of the 87 abstracts analyzed had been incomplete in the first version of the dataset. A more detailed error description can be found in the description section of the new version. |
2019-10-09 14:51:54 |
* Selected version