• For text classification and information retrieval tasks, text data has to be represented as a fixed dimension vector.
• We propose simple feature construction technique named P-SIF: Document Embeddings using Partition Averaging, accepted to appear at AAAI 2020.
• We demonstrate our method through experiments on multi-class classification on 20newsGroup dataset, multi-label text classification on Reuters-21578 dataset, Semantic Textual Similarity Tasks (STS 12-16) and other classification tasks.

There are 3 folders named 20newsGroup, Reuters, STS and other_datasets which contains code related to multi-class classification on 20newsGroup dataset, multi-label classification on Reuters dataset, Semantic Texual Similarity Task (STS) on 27 datasets, and multi-class classification on several datasets such as 20newsgroup, BBC sports, Amazon, Twitter, Classic, Reuters, and Recipe-L.

Change directory to 20newsGroup for experimenting on 20newsGroup dataset and create train and test tsv files as follows:

$ cd 20newsGroup
$ python create_tsv.py

Get word vectors for all words in vocabulary:

$ python Word2Vec.py 200
# Word2Vec.py takes word vector dimension as an argument. We took it as 200.

Get Sparse Document Vectors (SCDV) for documents in train and test set and accuracy of prediction on test set:

$ python psif.py 200 40
# ksvd_sif.py takes word vector dimension and number of partitions as arguments. We took word vector dimension as 200 and number of partitions as 60.

Change directory to Reuters for experimenting on Reuters-21578 dataset. As reuters data is in SGML format, parsing data and creating pickle file of parsed data can be done as follows:

$ python create_data.py
# We don't save train and test files locally. We split data into train and test whenever needed.

Get word vectors for all words in vocabulary:

$ python Word2Vec.py 200
# Word2Vec.py takes word vector dimension as an argument. We took it as 200.

Get Sparse Document Vectors (SCDV) for documents in train and test set:

$ python psif.py 200 40
# ksvd_sif.py takes word vector dimension and number of partitions as arguments. We took word vector dimension as 200 and number of partitions as 60.

Get performance metrics on test set:

$ python metrics.py 200 40
# metrics.py takes word vector dimension and number of partitions as arguments. We took word vector dimension as 200 and number of partitions as 60.

Change directory to STS for experimenting on STS dataset. First download paragram_sl999_small.txt from John Wieting's github and keep it in STS/data folder dataset is inside SentEval folder for gmm based data partioning, parameters for cluster, weightage etc is stored in parameters_gmm.csv Create word topic vector for each word by using wordvectors from paragram_sl999_small.txt

$ python create_word_topic_gmm.py

Get similarity score for each sts dataset

$ python psif_main_gmm.py
# it will output each dataset similarity score and corresponding parameters.

for ksvd based data partioning, parameters for cluster, weightage etc is stored in parameters_ksvd.csv Create word topic vector for each word by using wordvectors from paragram_sl999_small.txt

$ python create_word_topic_ksvd.py

Get similarity score for each sts dataset

$ python psif_main_ksvd.py
# it will output each dataset similarity score and corresponding parameters.

For running P-SIF on rest of the 7 datasets, go to Other_Datasets folder. Inside other_datasets folder, each dataset has a folders with the dataset name. Follow the readme.md has been included for running the P-SIF. You have to download google embedding from here and placed in the Other_Dataset folder.

Minimum requirements:

• Python 2.7+
• NumPy 1.8+
• Scikit-learn
• Pandas
• Gensim

@inproceedings{gupta2020psif,
  title={P-SIF: Document Embeddings using Partition Averaging},
  author={Gupta, Vivek and Saw, Ankit and Nokhiz, Pegah and Netrapalli, Praneeth and Rai, Piyush and Talukdar, Partha},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2020}
}

Note: You neednot download 20newsGroup or Reuters-21578 dataset. All datasets are present in their respective directories. We used SGML parser for parsing Reuters-21578 dataset from here)