search

DARIAH-DE / TopicsExplorer

Explore your own text collection with a topic model – without prior knowledge.
https://dariah-de.github.io/TopicsExplorer
Apache License 2.0
62 stars 10 forks source link
digital-humanities flask-application gui lda pyqt5 standalone-executables topic-modeling topics-explorer

readme

Explore your own text collection with a topic model – without prior knowledge.

The text mining technique topic modeling has become a popular procedure for clustering documents into semantic groups. This application introduces a user-friendly workflow which leads from raw text data to an interactive visualization of the topic model. All you need is a text corpus and a little time.

Topic modeling algorithms are statistical methods that analyze the words of the original texts to discover the themes that run through them, how those themes are connected to each other, and how they change over time. – David M. Blei

Overview

Getting started

Windows, macOS and Linux users do not have to install additional software. The application itself is portable.

  1. Go to the release-section and download the ZIP archive for your OS.
  2. Extract the content of the archive.
  3. Run the app by double-clicking the file DARIAH Topics Explorer.

If you are on a Mac and get an error message saying that the file is from an “unidentified developer”, you can override it by holding control while double-clicking. The error message will still appear, but you will be given an option to run the file anyway. You will have to do this for two executables. First, “topics-explorer-webengine” in the folder resources/webengine, afterwards “DARIAH Topics Explorer”.

With the release of macOS 10.15 Catalina, Apple has introduced new security features which result in issues with starting the TopicsExplorer. As a temporary workaround follow the instructions below to install the application from source. (You need Python 3.6)

You can also use the source code (if you are not on Windows):

  1. Download this repository.
  2. Unzip the archive, e.g. using unzip via the command-line.
  3. Make sure you have Poetry installed.
  4. Run poetry install to set up a virtual environment and install dependencies.
  5. To start the application, type poetry run python application.py, and press enter.

If you wish to use the sample corpus, you have to clone the repository with Git. See also section Sample corpus. If you download one of the archives (except the source code) from the release section, the corpus is included.

The application

DARIAH Topics Explorer

This application is designed to introduce topic modeling particularly gently (e.g. for educational purpose). If you have a very large text corpus, you may wish to use more powerful tools like MALLET, which is written in Java and can be completely controlled from the command-line. The topic modeling algorithm used in this application, latent Dirichlet allocation, was implemented by Allen B. Riddell using collapsed Gibbs sampling as described in Pritchard et al. (2000).

You might want to check out some Jupyter notebooks for topic modeling in Python – experimenting with an example corpus on Binder does not require any software on your local machine.

The sample corpus

We provide a sample corpus (10 British novels) in this project. If you use Git, you can include the corpus, which is actually a submodule in this repository, by writing:

$ git clone --recursive https://github.com/DARIAH-DE/TopicsExplorer.git

or if you have already cloned the repository:

$ cd data/british-fiction-corpus
$ git submodule init
$ git submodule update

Example visualizations

The following visualizations display the topic model output of 10 novels (written by Charles Dickens, George Eliot, Joseph Fielding, William Thackeray and Anthony Trollope).

Topics Explorer’s visualiztaions are interactive. You will be able to navigate through topics and documents, get similar topics and documents displayed, read excerpts from the original texts, and inspect the document-topic distributions in a heatmap.

Topics are probability distributions over the whole vocabulary of a text corpus. One value is assigned to each word, which indicates how relevant the word is to that topic (to be exact, how likely one word is to be found in a topic). After sorting those values in descending order, the first n words represent a topic.

Below the topics are ranked by their numerical dominance in the sample corpus; each bar displays a topic’s dominance score.

Each document consists to a certain extent of each topic, which is one of the theoretical assumptions of topic models. Although some values are too small to be visualized here (and have therefore been rounded to zero), they are actually greater than zero.

Visualizing the document-topic proportions in a heatmap displays the kind of information that is probably most useful. Going beyond pure exploration, it can be used to show thematic developments over a set of texts, akin to a dynamic topic model.

Developing

As a very first step, git clone this repository and install the dependencies.

Dependencies

Pipenv automatically creates and manages a virtualenv for this project. Install the tool as usual:

$ pip install pipenv

This application requires Python 3.6 – it is highly recommended to use pyenv for managing Python versions. Pipenv and pyenv works hand-in-hand.

To install the project’s dependencies:

$ pipenv install

Running the application

After spawning a shell within the virtual environment, using pipenv shell, you can run the application with:

$ python application.py

If you wish to access the application through your web browser, use the following command:

$ python application.py --browser

Simulating a frozen app and just starting the backend, run:

$ python application.py --frozen

Freezing the backend

This can be really hard, starting with the fact that you have to create an executable on the operating system you want it to run on.

However, Python applications can be frozen with PyInstaller. Competitor products like cx_Freeze or py2exe/py2app are no longer maintained, or are used much less often (which makes troubleshooting hard).

The whole workflow consists of freezing backend and frontend separately. This is because there have been enormous problems with the packaging of PyQt5 (the Python-based frontend), and we are using a JavaScript-based frontend, Electron, instead. To run Topics Explorer from source code, we still use PyQt5 because it works pretty well and we stick with one programming language.

The backend is a Flask application that is started with the script application.py and the parameter --frozen (or in the actually frozen application also without this parameter). The home page will be available at http://localhost:5001.

It is recommended to install the current develop branch from PyInstaller on GitHub. If you encounter problems, you should try different versions of the application’s dependencies.

Opening files with relative paths in the frozen app does not work. You will have to join paths with sys._MEIPASS, e.g. pathlib.Path(sys._MEIPASS, "file.txt"). In the code, however, this is already implemented in the corresponding places.

PyInstaller creates a *.spec file in which the whole freezing process is configured. For example, if data needs to be copied (in this project: schema.sql), it is configured there. In general, you have the choice between a "closed" single executable, and a directory where the executable and resources are located. The application opens, obviously, in case of a directory much faster. It is also easier to copy data afterwards.

It is strongly recommended to create a new, clean virtualenv from which the application will be frozen. Other packages that are not needed by the application itself may be packaged by PyInstaller, for whatever reason, which unnecessarily inflates the whole thing. It is also strongly recommended to delete the gui.py module, otherwise PyQt5 (~ 100 MB) will be frozen in any case.

To freeze the backend, merge master (or whatever branch you want to freeze) into the freeze-backend branch:

$ git checkout freeze-backend
$ git rebase master

Now you can run the PyInstaller command:

$ pyinstaller topicsexplorer.spec

You can find the executable and all resources in dist/topics-explorer-backend.

Freezing the frontend

This is really easy. Checkout the freeze-frontend branch and follow the instructions in the README.

Putting it all together

You now have a topics-explorer-backend folder containing the frozen Flask application in the form of an executable with all other resources, and a topics-explorer-frontend folder containing an executable with all resources. You could already deploy the application now, since you start the backend first, then the frontend. The convenience-wrapper branch contains a script that does just that for you. It starts a subprocess to start the backend, waits until http://localhost:5001 is available, and then starts the frontend.

Checkout the branch convenience-wrapper, and follow the instructions in the README. If your frozen application adheres to the following structure and everything works, congratulations, zip it and release it:

dariah-topics-explorer-{version}-{os}/
├── DARIAH Topics Explorer
├── LICENSE
├── README.md
├── sample-data/
|   ├── british-fiction-corpus/
|   └── stopwords/
└── resources/
    ├── frontend/
    |   ├── topics-explorer-frontend
    |   └── ...
    └── backend/
        ├── topics-explorer-backend
        └── ...

Troubleshooting

In general:

Regarding the standalone executable:

Regarding the source code:

What is Topic Modeling?

What is DARIAH-DE?

DARIAH-DE supports research in the humanities and cultural sciences with digital methods and procedures. The research infrastructure of DARIAH-DE consists of four pillars: teaching, research, research data and technical components. As a partner in DARIAH-EU, DARIAH-DE helps to bundle and network state-of-the-art activities of the digital humanities. Scientists use DARIAH, for example, to make research data available across Europe. The exchange of knowledge and expertise is thus promoted across disciplines and the possibility of discovering new scientific discourses is encouraged.

This application is developed with support from the DARIAH-DE initiative, the German branch of DARIAH-EU, the European Digital Research Infrastructure for the Arts and Humanities consortium. Funding has been provided by the German Federal Ministry for Research and Education (BMBF) under the identifier 01UG1610A to J.

Citing

To cite the DARIAH-DE TopicsExplorer, please refer to: Severin Simmler, Thorsten Vitt and Steffen Pielström, Topic Modeling with Interactive Visualizations in a GUI Tool, in: Proceedings of the Digital Humanities Conference 2019.

License

This project is licensed under Apache 2.0. You can do what you like with the source code, as long as you include the original copyright, the full text of the Apache 2.0 license, and state significant changes. You cannot charge DARIAH-DE for damages, or use any of its trademarks like name or logos.