vyraun / text-classification-wv

Text Classification Using Word Vectors.

WordVectors

Text Classification with Word Vectors

Download Pre-trained Vectors

• cd into Repository
• wget http://nlp.stanford.edu/data/glove.6B.zip
• unzip glove.6B.zip

Install

• pip install pandas sklearn gensim tensorflow keras bs4
• nltk.download('punkt')
• nltk.download('stopwords')

Get Reduced Vectors from Pre-trained vectors

• python reduction_algo.py [embedding_file] [reduced_dimensions] (e.g. python reduction_algo glove.300d.txt 150)
• e.g. python reduction_algo glove.300d.txt 150 --> the reduced embeddings will be saved in reduced_embeddings_150.txt

20Newsgroup on Pretrained-Glove

• svc.py creates the document vectors + gives results
• e.g. python svc.py glove.300d.txt 300

Reuters on Pretrained Glove

• svc_reuters.py creates and evaluates the document vectors
• e.g. python svc_reuters.py glove.300d.txt 300

Train Vectors Using Word2Vec Model

• Run the Word2VecModel_on_Newsgroup.py and Word2VecModel_on_Reuters.py files
• Embedding files will be created, use them just as pre-trained vectors for evaluation
• e.g Word2VecModel_Newsgroup.py 400 -> created embedding_on_newsgroup_400.txt

Evaluation Table

Embedding	20Newsgroup	Reuters
Glove-300D	60
Glove-200D	53
Glove-100D	50
Glove-Reduced-150D	51
Glove-Reduced-100D	42
Glove-Reduced-50D	36
Fasttext-300D
Fasttext-Reduced-150D
Word2Vec-300D
Word2Vec-Reduced-150D
W2V-Newsgroup-300D	73 (0.7379182156133829)	x
W2V-Newsgroup-200D	0.6736590546999469	x
W2V-Newsgroup-400D	0.7124269782262347	x
W2V-Newsgroup-Reduced-150D	60 (0.6023632501327668)	x
W2V-Newsgroup-Reduced-100D		x
W2V-Newsgroup-Reduced-200D	0.6427243759957515	x
W2V-Reuters-300D	x	41 (0.4121083377588954)
W2V-Reuters-200D	x
W2V-Reuters-100D	x
W2V-Reuters-Reduced-150D	x	32 (0.3252788104089219)
W2V-Reuters-Reduced-100D	x
W2V-Reuters-Reduced-50D	x