suhasagg/sif

Sentence embedding using Smooth Inverse Frequency weighting scheme

A useful tutorial

https://www.kaggle.com/procode/sif-embeddings-got-69-accuracy

Sentence embedding using Smooth Inverse Frequency weighting scheme

This is the implementation of SIF prposed in "A Simple but Tough-to-Beat Baseline for Sentence Embeddings" https://openreview.net/forum?id=SyK00v5xx.

This code heavily reuses the implementation available at https://github.com/PrincetonML/SIF

The changes I have done are to make it fast and scalabeld and fix few issues. The code focuses on generating the embeddings using SIF which can be used in the downstream tasks.

Check the test.py file to understand how to generate the embeddings.

Download and copy the glove file glove.840B.300d.txt to the working directory and run python test.py

test.py uses a sample test.csv file and generated embeddings for the column called question1. To show how good the embeddings are, cosine similarity is used to show the top 5 similar questions for each question and the result is saved in result.csv

test.csv is a sampled file from test.csv of quora challenge.