This is a microservice that can run in a docker container and perform SeaNMF topic detection when queried by a REST call.
See swagger.yaml for details. The tool at https://editor.swagger.io/ can be used to render the swagger file.
alpha - Factorization weight for word-semantic correlations, higher alpha may increase coherence, but reduce interpretability of topics. Can be in (0,1] range. Default: 0.1
beta - Sparsity factor, increase beta (eg. 0.1) for SparseSeaNMF (SSeaNMF), for normal SeaNMF this parameter is not needed (=0). Can be in [0,1] range. Default: 0
n_topics - The number of topics that shall be detected. Higher topic coherence indicates a better n_topics. Can be any number > 0. Default: 10
max_iter - Maximum number of iterations that will be performed. Default: 500
max_err - Error threshold. The processing will stop when the error is smaller than max_err or max_iter is reached. Default: 0.1
fix_random - Set to true to fix random seed to 0. This will make the results reproducible. Default: false
vocab_min_count - Only words that occur more than vocab_min_count times will be added to vocabulary, if the dataset is small and the vocab_min_count to high, the processing will fail. Default: 3
This repository is based on the SeaNMF implementation of the paper
- Tian Shi, Kyeongpil Kang, Jaegul Choo and Chandan K. Reddy, "Short-Text Topic Modeling via Non-negative Matrix Factorization Enriched with Local Word-Context Correlations", In Proceedings of the International Conference on World Wide Web (WWW), Lyon, France, April 2018. PDF
Original Repository: (https://github.com/tshi04/SeaNMF)
The code has been modified to be dockerized and to offer a REST API via Flask.
Free use of this software is granted under the terms of the GPL version 3 (GPL 3.0).