Research Method: Topic Modeling
Workflow: Topic Model Runs | Facilitate Interpretation

    Topic Modeling Tools (see DH Toychest for fuller list)

• MALLET (command-line version; download and install in a folder called "mallet" directly in root directory of your computer)
• Java GUI version of Mallet (aka "GUI Topic Modeling Tool"; download the .jar file and run from your computer)
• LDAvis ("R package for interactive topic model visualization) (example of use)
• MALLET-to-Gephi Data Stacker (online tool that takes "the '--output-doc-topics' output from MALLET and reorganize it into a format that Gephi understands")
• [for other tools related to text-preparation and workflow for topic modeling, see below]
  

 

   Tutorials for MALLET Topic Modeling

• The Programming Historian MALLET Tutorial (includes instructions for installing MALLET)
• Instructions for install
• "Topic Modeling" page on MALLET site
  
• Matt Burton, "Mallet Options Grouped"
• Text Analysis with Topic Models for the Humanities and Social Sciences (Python tutorial; see especially the Visualizing Topic Models page)

 

    Initial Proof of Concept for small sample of documents from WhatEvery1Says corpus

• “The Heart of the Matter” Topic-Modeled (A Preliminary Experiment)
• “The Heart of the Matter” Visualized

 

Participants in the project should all perform experimental topic modeling runs on samples from the corpus as they work on particular components of the text-preparation processes outlined above.  Ideally, there will be an iterative relation between tweaks to those processes and tweaks to the topic modeling.

 

• Log of Topic Model Runs (and MALLET commands): collective log of our topic model runs with a record by date of each run and the MALLET commands used, plus commentary on or links to the results (in progress)
• Alan's topic model run of 1 May 2014 (using MALLET): results

1. Experiment with methods for facilitating the interpretation of topic models. Examples:
1. Andrew Goldstone's interface
1. Shawn Graham, "Using Goldstone’s Topic Modeling R package with JSTOR’s Data for Research"
2. Jeffrey M. Binder and Collin Jennings's interface
3. Lexos word clouds for topic models (in progress: Scott Kleinman)
4. LDAvis ("R package for interactive topic model visualization) (example of use) 
5. MALLET-to-Gephi Data Stacker (online tool that takes "the '--output-doc-topics' output from MALLET and reorganize it into a format that Gephi understands")
2. Create a web site with a good front-end interface for exploring the WhatEvery1Says topic model (to do)

 

Other Tools and Research on the state of the art in interpreting topic model results:

• Jacob Eisenstein, Iris Sun, and Lauren F Klein, Exploratory Thematic Analysis for Historical Ne.pdf  (from Conference Abstracts of the DH 2014 conference, Lausanne, 7-12 July 2014)