Program Name: Word Retrofitting Tool

Overview:

The Word Retrofitting Tool is a jupyter-notebook-based program that performs retrofitting on word vectors using a lexicon of word relations. Retrofitting aims to improve the quality and semantic coherence of word vectors by incorporating information from the lexicon.

Prerequisites:

• Python 3.x
• Required Python packages: numpy, pandas

Requirements:

In order to run the Word Retrofitting Tool, make sure you have the following models and libraries downloaded:

For French:

Pre-trained Word Vectors:

• Download vecs100-linear-frwiki.zip or vecs50-linear-frwiki.bz2 file for French word vectors trained on a Wikipedia dump.
  • Training on a dump of Wikipedia (frwiki-20140804-corpus.xml.bz2 downloaded here)
  • Preprocessing:
    • Use xml2txt.pl to remove XML tags.
    • Tokenize the text using sxpipe-light (e.g., perl ~/installation/melt-2.0b7/sxpipe-melt/segmenteur.pl < frwiki_raw.txt > frwiki_tokenized.txt).
  • The word vectors are trained using vecs100 or vecs50 with the following command:
    • ./word2vec -train frwiki_tokenized.txt -output vecs50 -threads 2 -min-count 100 -cbow 0 -negative 10
• Alternatively, you can use other word embeddings such as the cc.fr.300.vec embeddings. Refer to fastText documentation for more details.

Semantic Resources:

• Use WordNet for semantic relations.

Evaluation Dataset:

• For lexical similarity evaluation, use the rg65_french.txt file.

For English:

Pre-trained Word Vectors:

• Download vectors_datatxt_250_sg_w10_i5_c500_gensim_clean.tar.bz2 file for English word vectors.
• The word vectors are trained using the gensim library with the following configuration:
  • gensim.models.Word2Vec(size=250, min_count=500, window=8, sample=1e-3, workers=8, sg=1, hs=0, negative=10, iter=5)

Semantic Lexicons:

• Use WordNet for semantic relations.

Evaluation Dataset:

• For lexical similarity evaluation, use the ws353.txt file.
• For sentiment analysis evaluation, use the stanford_sentiment_analysis.tar.gz file from Stanford NLP.

Usage:

1. Open the jupyter notebook files
2. Run the program by pressing the button “run”. The program will start executing and perform retrofitting on the word vectors using all predetermined parameters and lexicon.
3. Once the process is completed, the retrofitted word vectors will be saved in the specified output file.

Notes:

• Ensure that the input file, lexicon file, and output file paths are correctly specified and accessible.
• The program may take some time to complete retrofitting, depending on the size of the word vectors and the lexicon.

Authors:

• Nina alias NinaNusb
• Deeksha alias deekode99
• Iness alias IAiness