Poppins - A lightweight document classifier

Poppins is a text classifier, which means that you can use it to classify text with diverse criteria. It all depends on how you train the software. Training the software means showing it examples of documents that pertain to any arbitrary set of categories, plus one set of documents to be classified.
It is actually pretty simple: say you have a bunch of texts about cooking recipes and another of instruction manuals for tape recorders. These are two categories of documents with same examples. Imagine then you have a third set of different documents that mix the previous two categories. These are the texts you want to classify.
You can use this program to classify documents in any language and you can have any number of categories.
Some users reported success in classifying documents by topic, by language, by genre, by author. You are encouraged to try new ideas. No ones knows what other criteria might be to classify text.

History

Poppins has been online since 2005. It was created by Rogelio Nazar as a final assignment in an NLP course in Universitat Politècnica de Catalunya, in Barcelona, and it has been used by few people over the years. The graphic design was done back then by Sebastián Márques and Alejandro Bevaqua. Now (13 October, 2023), with the help of Nicolás Acosta, we inaugurate a new version, with a more friendly interface compared to the original, which is still online, by the way, but we would like to make it disappear as soon as possible.

References

The project is described in the following papers:

In English:

• Nazar, R. & Sánchez Pol, M. (2007). An Extremely Simple Authorship Attribution System. In M.T. Turell, J. Cicres, & M. S. Spassova (Eds.), Proceedings of the 2nd European IAFL Conference on Forensic Linguistics / Language and the Law 2006. Barcelona: Documenta Universitaria. (PDF)

In Catalan / en català:

• Nazar, R. (2010). Invitació a l'estudi estadístic del llenguatge. La terminologia i la documentació: relacions i sinergies: actes de la VII Jornada SCATERM: Terminologia i documentació, Barcelona: Societat Catalana de Terminologia, pp. 47-73. (PDF)

In Spanish / en castellano:

• Nazar, R. (2018). Invitación al estudio estadístico del lenguaje. ArXiv:1804.07349, [stat.AP]. (PDF)

Here we show some examples.

Example with the Federalist Papers

The first experiment we made was with the Federalist Papers, a famous case of disputed authorship (learn more about the case in Wikipedia).
If you would like to repeat the experiment yourself, you can download the corpus from our server. It's a single zip file, but it contains other zips for every category (in this case, author) plus one called test, which contains the disputed essays to be classified.
Training...

	hamilton.txt	hammad.txt	jay.txt	madison.txt
Class hamilton
Class hammad
Class jay
Class madison

Classifying...

	unknown1.txt	unknown10.txt	unknown11.txt	unknown12.txt	unknown2.txt	unknown3.txt	unknown4.txt	unknown5.txt	unknown6.txt	unknown7.txt	unknown8.txt	unknown9.txt
Class hamilton
Class hammad
Class jay
Class madison

Finished.

---

Another example with authorship attribution

The following is another case of authorship attribution, and is the result of a collaboration with Marta Sánchez Pol. We experimented with a dataset she compiled for her thesis, which consisted of many short texts by a variety of authors.
Training...

	1_01.txt	1_010.txt	1_02.txt	1_03.txt	1_04.txt	1_05.txt	1_06.txt	1_07.txt	1_08.txt	1_09.txt	2_01.txt	2_010.txt	2_02.txt	2_03.txt	2_04.txt	2_05.txt	2_06.txt	2_07.txt	2_08.txt	2_09.txt	3_01.txt	3_010.txt	3_02.txt	3_03.txt	3_04.txt	3_05.txt	3_06.txt	3_07.txt	3_08.txt	3_09.txt	5_01.txt	5_010.txt	5_02.txt	5_03.txt	5_04.txt	5_05.txt	5_06.txt	5_07.txt	5_09.txt	6_01.txt	6_010.txt	6_02.txt	6_03.txt	6_04.txt	6_05.txt	6_06.txt	6_07.txt	6_08.txt	6_09.txt
Class 1
Class 2
Class 3
Class 5
Class 6

Classifying...

	1_011.txt	1_012.txt	1_013.txt	1_014.txt	1_015.txt	1_016.txt	1_017.txt	1_018.txt	1_019.txt	1_020.txt	2_011.txt	2_012.txt	2_013.txt	2_014.txt	2_015.txt	2_016.txt	2_017.txt	2_018.txt	2_019.txt	2_020.txt	3_011.txt	3_012.txt	3_013.txt	3_014.txt	3_015.txt	3_016.txt	3_017.txt	3_018.txt	3_019.txt	3_020.txt	5_011.txt	5_012.txt	5_013.txt	5_014.txt	5_015.txt	5_016.txt	5_017.txt	5_018.txt	5_019.txt	5_20.txt	6_011.txt	6_012.txt	6_013.txt	6_14.txt	6_15.txt	6_16.txt	6_17.txt	6_18.txt	6_19.txt	6_20.txt
Class 1
Class 2
Class 3
Class 5
Class 6

Finished.

A third example classifying by topic

In this case we classify texts by topic: economy, medicine and computer science. Again, the performance is pretty good.

Training...

	economia1.txt	economia11.txt	economia12.txt	economia13.txt	economia2.txt	economia3.txt	informatica1.txt	informatica10.txt	informatica12.txt	informatica14.txt	informatica2.txt	informatica3.txt	medicina1.txt	medicina10.txt	medicina100.txt	medicina101.txt	medicina103.txt	medicina104.txt	medicina105.txt	medicina11.txt
Class economia
Class informatica
Class medicina

Classifying...

	economia14.txt	economia16.txt	economia17.txt	economia18.txt	informatica16.txt	informatica17.txt	informatica18.txt	informatica19.txt	informatica23.txt	informatica24.txt	informatica25.txt	medicina105.txt	medicina106.txt	medicina107.txt	medicina109.txt	medicina114.txt
Class economia
Class informatica
Class medicina

Finished.