Yelp Prediction

Team members: Cecily Liu, Tingting Liang, Richard Yuan

Description

This project is aimed to predict Yelp ratings based on 1.5 million train data. We use a L2-regularized logistic classfier, which only spends less than 3 minutes on training a model and yields a RMSE 0.65.

Code

Code for data cleaning and image generating.

• data_clean_punc.py & data_clean_text.py: data cleaning
• logit-saga.py: model fitting
• word cloud plot.R: generating word cloud for categories
• data_clean_example.py & print_picture.py: examples
• word2vec.py: convert words to vectors
• process_fasttext_result.py: process result of fastText

Image

• Character Length(star*).png: character length distribution from 1-star to 5-star ratings
• Negative_Category.png & Positive_Category.png: ratings distribution based on categories
• Punctuation().png: average ratings vs count of a certain punctuation
• Wordcloud().png: word tag plots of the most frequent words of four categories.
• wordcloud.png: overall word importance
• wordcloud_packages_neg.png & wordcloud_packages_pos.png: word importance with largest negative/positive coefficients

Slides

Two slides for two parts of this project

Yelp.ipynb

Summary of our whole project.