The state-of-the-art platform for topic modeling.
What is BigARTM?
BigARTM is a powerful tool for topic modeling based on a novel technique called Additive Regularization of Topic Models. This technique effectively builds multi-objective models by adding the weighted sums of regularizers to the optimization criterion. BigARTM is known to combine well very different objectives, including sparsing, smoothing, topics decorrelation and many others. Such combination of regularizers significantly improves several quality measures at once almost without any loss of the perplexity.
- Vorontsov K., Frei O., Apishev M., Romov P., Dudarenko M. BigARTM: Open Source Library for Regularized Multimodal Topic Modeling of Large Collections // Analysis of Images, Social Networks and Texts. 2015.
- Vorontsov K., Frei O., Apishev M., Romov P., Dudarenko M., Yanina A. Non-Bayesian Additive Regularization for Multimodal Topic Modeling of Large Collections // Proceedings of the 2015 Workshop on Topic Models: Post-Processing and Applications, October 19, 2015 – pp. 29-37.
- Vorontsov K., Potapenko A., Plavin A. Additive Regularization of Topic Models for Topic Selection and Sparse Factorization. // Statistical Learning and Data Sciences. 2015 — pp. 193-202.
- Vorontsov K. V., Potapenko A. A. Additive Regularization of Topic Models // Machine Learning Journal, Special Issue “Data Analysis and Intelligent Optimization”, Springer, 2014.
- More publications can be found in our wiki page.
Related Software Packages
- TopicNet is a high-level interface for BigARTM which is helpful for rapid solution prototyping and for exploring the topics of finished ARTM models.
- David Blei’s List of Open Source topic modeling software
- MALLET: Java-based toolkit for language processing with topic modeling package
- Gensim: Python topic modeling library
- Vowpal Wabbit has an implementation of Online-LDA algorithm
Installing with pip (Linux only)
We have a PyPi release for Linux:
$ pip install bigartm
$ pip install bigartm10
Installing on Windows
We suggest using pre-build binaries.
It is also possible to compile C++ code on Windows you want the latest development version.
Installing on Linux / MacOS
Download binary release or build from source using cmake:
$ mkdir build && cd build $ cmake .. $ make install
See here for detailed instructions.
How to Use
Check out documentation for
- Basic model (20 topics, outputed to CSV-file, inferred in 10 passes)
bigartm.exe -d docword.kos.txt -v vocab.kos.txt --write-model-readable model.txt --passes 10 --batch-size 50 --topics 20
- Basic model with less tokens (filtered extreme values based on token’s frequency)
bigartm.exe -d docword.kos.txt -v vocab.kos.txt --dictionary-max-df 50% --dictionary-min-df 2 --passes 10 --batch-size 50 --topics 20 --write-model-readable model.txt
- Simple regularized model (increase sparsity up to 60-70%)
bigartm.exe -d docword.kos.txt -v vocab.kos.txt --dictionary-max-df 50% --dictionary-min-df 2 --passes 10 --batch-size 50 --topics 20 --write-model-readable model.txt --regularizer "0.05 SparsePhi" "0.05 SparseTheta"
- More advanced regularize model, with 10 sparse objective topics, and 2 smooth background topics
bigartm.exe -d docword.kos.txt -v vocab.kos.txt --dictionary-max-df 50% --dictionary-min-df 2 --passes 10 --batch-size 50 --topics obj:10;background:2 --write-model-readable model.txt --regularizer "0.05 SparsePhi #obj" --regularizer "0.05 SparseTheta #obj" --regularizer "0.25 SmoothPhi #background" --regularizer "0.25 SmoothTheta #background"
Interactive Python interface
BigARTM supports full-featured and clear Python API (see Installation to configure Python API for your OS).
import artm # Prepare data # Case 1: data in CountVectorizer format from sklearn.feature_extraction.text import CountVectorizer from sklearn.datasets import fetch_20newsgroups from numpy import array cv = CountVectorizer(max_features=1000, stop_words='english') n_wd = array(cv.fit_transform(fetch_20newsgroups().data).todense()).T vocabulary = cv.get_feature_names() bv = artm.BatchVectorizer(data_format='bow_n_wd', n_wd=n_wd, vocabulary=vocabulary) # Case 2: data in UCI format (https://archive.ics.uci.edu/ml/datasets/Bag+of+Words) bv = artm.BatchVectorizer(data_format='bow_uci', collection_name='kos', target_folder='kos_batches') # Learn simple LDA model (or you can use advanced artm.ARTM) model = artm.LDA(num_topics=15, dictionary=bv.dictionary) model.fit_offline(bv, num_collection_passes=20) # Print results model.get_top_tokens()
BigARTM is released under New BSD License that allowes unlimited redistribution for any purpose (even for commercial use) as long as its copyright notices and the license’s disclaimers of warranty are maintained.