This is the website of our group project "comment grading for recommendation" for Google ML Camp 2020. This repository contains all the code we used for data cleaning, neural network architecture, and result analysis. You can easily download and deploy it on your local computer for testing ^_^
Poster: poster.pdf
ppt： Turiss_pre.pdf
Dataset: https://www.kaggle.com/snap/amazon-fine-food-reviews
BERT pretrained model: https://tfhub.dev/tensorflow/bert_en_uncased_L-12_H-768_A-12/1
The folder ./SingleModal contains DataClean.ipynb, Simple_LSTM.py, and Simple_BERT.py. These baseline models are only based on comments embedding.
The folder ./Multi_Input contains dic_XXX.pkl, which are used for dropping out the rare IDs(user, product, and time) and remapping the rest to numbers.
The folder ./MultiModal contains run.py. You can use it to test the BERTMulMod model and the BERTMulModMulTask model. The usage is run.py [--epoch X] [--big 1] [--model choice['SingleLSTM', 'SingleBERT', 'BERTMulMod' , 'BERTMulModMulTask']]. The logs and result files are also in this folder.
The folder ./Stat contains the statistic-related codes used for the analysis of recommendation systems.
The folder ./img_folder contains the images that visualized the performance of our neural networks and ranking systems.