This repository is a collection of code snippets for working with formal contexts. Not only methods for reading, displaying, manipulating and saving formal contexts are provided, but also a novel technique of topic aggregation via formal concept analysis (FCA) is introduced. The idea extends "The Geometric Structure of Topic Models", Johannes Hirth and Tom Hanika (2024), where the authors compute document-topic incidences and apply FCA as means to visualize the information. We extend this idea by computing directory-topic incidences. Hence, we can aggregate topics on a higher level, which is particularly useful for the exploration of large corpora. The workflow is as follows:
The result for the dataset EYNTKE is shown below:
When running certain methods of conexp-clj I ran into compatibility problems: While the highest supported Java version is 21, the library requires a Java 23/ openjdk-23-jdk. It is possible to overcome this issue by rather using a standalone jar than the default web repository from maven/ clojars. In order to install the standalone jar with java 21 use the following command:
mvn install:install-file -Dfile=/home/kgutekunst/IdeaProjects/conexp-clj/builds/uberjar/conexp-clj-2.6.0-standalone.jar -DgroupId=conexp-clj -DartifactId=conexp-clj -Dversion=2.6.0+java21 -Dpackaging=jarIf you want to create the formal context across multiple directories, you can use the following code snippet:
lein run -m clj-exploration-leaks.explorationIf you don't want the metadata, file tree and statistics functionality,
you have to comment out the respective lines (up to the last letblock) in the code.
Be aware that the code assumes that the data is stored under certain directory names which are defined in the let block
and have to be adjusted to fit your needs.
Moreover, the data merged into the directory-topic context has to be aquired beforehand.
I have a second repository for this purpose, which your can find here:
text_topic.