GuidedNMF

This is the repository for the paper

On a Guided Nonnegative Matrix Factorization
by Joshua Vendrow, Jamie Haddock, Elizaveta Rebrova, and Deanna Needell, published in IEEE ICASSP 2021.

Abstract: Fully unsupervised topic models have found fantastic success in document clustering and classification. However, these models often suffer from the tendency to learn less-than-meaningful or even redundant topics when the data is biased towards a set of features. For this reason, we propose an approach based upon the nonnegative matrix factorization (NMF) model, deemed \textit{Guided NMF}, that incorporates user-designed seed word supervision. Our experimental results demonstrate the promise of this model and illustrate that it is competitive with other methods of this ilk with only very little supervision information.

Included Files

The format of the repository is as follows:
-Newsgroup.ipynb contains the experiments on the 20 Newsgroup data set.
-Twitter.ipynb contains the experiments on a Twitter political data set.
-Ablation.ipynb contains the comparison between GuidedNMF and SeededLDA.
-seeded_lda is a folder containing helper scripts for running SeededLDA, as well as the results of the ablation experiment.
-read_tweets.py contains a script to generate the twitter data set (access to this data requires a Twitter Developer account).

Instructions for Running SeededLDA.

The authors of the SeededLDA method have an official repository for the paper at https://github.com/bsou/cl2_project/tree/master/SeededLDA. To run SeededLDA, clone this repository and follow these instructions in addition tho those provided by the authors:

1. Prior to any runs, change the file paths for any provided scripts in seeded_lda, as well as lines 1028 and 1029 of src/SeededLDA_M1.cpp.

2. To compile, run c++ -I../libc SeededLDA_M1.cpp inside of the src folder.

3. Place seeded_lda/script.sh into the SeededLDA/src directory, and place all the remaining foles from seeded_lda into the SeededLDA/data directory.

4. Inside of the SeededLDA/data directory, run python3 create_seeds.py to create the a seperate file for each amount of with seeds words.

5. To run the experiments, use the command ./scipt.sh. You may need to first run chmod u+x script.sh.

6. This script runs experiments at a given rank. To change the rank, change line 4 in 20news_config to "iNoTopics N" where N is the desired rank and change line 7 of script.sh, setting the name "SeededLDA_docTopicDist_N_$i.txt" where N is the desired rank.

Quick Dependency Note

The codebase for Guided NMF depends on verssion version 0.0.2 of the ssnmf package. We make this requirement explicit in the requirements.txt file and restate it here due to issues raised by others using this repository. In order to install the correct version of ssnmf using pip, run the following command:

$ pip install ssnmf==0.0.2

Alternatively, all of the requirements for the repository can be installed with the following command:

$ pip install -r requirements.txt