NetAb

This repo hosts the code for paper "Learning with Noisy Labels for Sentence-level Sentiment Classification" (in EMNLP-2019).

Python Development Environment

• Python 3.6.7
• MongoDb
• pymongo
• tensorflow-gpu 1.9.0

Installation

1. Download the project NetAb;
2. Unzip the downloaded project;

Then the project is organized as follows

├── .idea                 <- IntelliJ’s project specific settings files
├── Data
│   ├── TestSens          <- Clean-labeled test sentences
│   ├── TrainingSens      <- Noisy-labeled train sentences
│   ├── ValSens           <- Clean-labeled validation sentences
│   └── word2id           <- Word to index
│
├── model                 <- Network functions
├── config.py             <- Configuration information
├── create_w2v_mongo.py   <- To create a word2vectors with mongodb
├── data_helper.py        <- Utilities
├── main.py               <- Main function
├── README.md             <- Guide for user(s) to perform this project.

Usage

1. Download the pre-trained word vectors GloVe.840B.300d; and then place it to the folder ./data/;
2. Run create_w2v_mongo.py to create a mongodb version for the GloVe.840B.300d;
3. Run main.py to produce the sentention classification results on each dataset, (e.g., python -m main -dataset 'movie').

Any questions, please let me know. Thanks!

Hao WANG Email: cshaowang@gmail.com