Yelp Review Sentiment Prediction using NLP (Glove)

This repo demonstrate a sentiment prediction based on Yelp reviews using natural language processing based on Glove word embeddings in Keras.

How to Use:

1. Clone in local directory

git clone [email protected]:mmortazavi/yelp_review_sentiment_prediction_nlp.git

2. Install requirements

pip install requirements.txt

3. Download glove.6B.100d.txt Glove Embedding

copy glove.6B.100d.txt embedding

4. Navigate into yelp_review_sentiment_prediction_nlp directory, and run:

python setup.py

Results after 20 Epochs on 1 Millions Data Points:

Epoch 1/20
80000/80000 [==============================] - 217s 3ms/step - loss: 0.4156 - acc: 0.8058 - val_loss: 0.3105 - val_acc: 0.8642
Epoch 2/20
80000/80000 [==============================] - 201s 3ms/step - loss: 0.3136 - acc: 0.8650 - val_loss: 0.2947 - val_acc: 0.8739
Epoch 3/20
80000/80000 [==============================] - 192s 2ms/step - loss: 0.2888 - acc: 0.8768 - val_loss: 0.2743 - val_acc: 0.8834
Epoch 4/20
80000/80000 [==============================] - 193s 2ms/step - loss: 0.2726 - acc: 0.8845 - val_loss: 0.2619 - val_acc: 0.8883
Epoch 5/20
80000/80000 [==============================] - 174s 2ms/step - loss: 0.2605 - acc: 0.8907 - val_loss: 0.2559 - val_acc: 0.8930
Epoch 6/20
80000/80000 [==============================] - 172s 2ms/step - loss: 0.2488 - acc: 0.8964 - val_loss: 0.2637 - val_acc: 0.8901
Epoch 7/20
80000/80000 [==============================] - 172s 2ms/step - loss: 0.2408 - acc: 0.9004 - val_loss: 0.2637 - val_acc: 0.8887
Epoch 8/20
80000/80000 [==============================] - 171s 2ms/step - loss: 0.2327 - acc: 0.9036 - val_loss: 0.2615 - val_acc: 0.8901
Epoch 9/20
80000/80000 [==============================] - 203s 3ms/step - loss: 0.2259 - acc: 0.9074 - val_loss: 0.2477 - val_acc: 0.8953
Epoch 10/20
80000/80000 [==============================] - 178s 2ms/step - loss: 0.2169 - acc: 0.9116 - val_loss: 0.2484 - val_acc: 0.8966
Epoch 11/20
80000/80000 [==============================] - 183s 2ms/step - loss: 0.2123 - acc: 0.9141 - val_loss: 0.2486 - val_acc: 0.8981
Epoch 12/20
80000/80000 [==============================] - 183s 2ms/step - loss: 0.2063 - acc: 0.9163 - val_loss: 0.2553 - val_acc: 0.8962
Epoch 13/20
80000/80000 [==============================] - 170s 2ms/step - loss: 0.2020 - acc: 0.9185 - val_loss: 0.2493 - val_acc: 0.8971
Epoch 14/20
80000/80000 [==============================] - 170s 2ms/step - loss: 0.1951 - acc: 0.9223 - val_loss: 0.2522 - val_acc: 0.8960
Epoch 15/20
80000/80000 [==============================] - 164s 2ms/step - loss: 0.1913 - acc: 0.9225 - val_loss: 0.2592 - val_acc: 0.8956
Epoch 16/20
80000/80000 [==============================] - 162s 2ms/step - loss: 0.1856 - acc: 0.9251 - val_loss: 0.2645 - val_acc: 0.8913
Epoch 17/20
80000/80000 [==============================] - 171s 2ms/step - loss: 0.1818 - acc: 0.9280 - val_loss: 0.2587 - val_acc: 0.8949
Epoch 18/20
80000/80000 [==============================] - 200s 3ms/step - loss: 0.1764 - acc: 0.9301 - val_loss: 0.2581 - val_acc: 0.8951
Epoch 19/20
80000/80000 [==============================] - 184s 2ms/step - loss: 0.1732 - acc: 0.9304 - val_loss: 0.2670 - val_acc: 0.8915
Epoch 20/20
80000/80000 [==============================] - 203s 3ms/step - loss: 0.1689 - acc: 0.9327 - val_loss: 0.2754 - val_acc: 0.8923
Training Complete in 3666.44 s.
----------------------------------------------------------------------------------------------------

Score: 89.22

Classification Report:
               precision    recall  f1-score   support

    Negative       0.85      0.82      0.84      6706
    Positive       0.91      0.93      0.92     13294

   micro avg       0.89      0.89      0.89     20000
   macro avg       0.88      0.88      0.88     20000
weighted avg       0.89      0.89      0.89     20000

----------------------------------------------------------------------------------------------------

The Repo Contains:

- data dir: a small fraction of yelp reviews in parquet format
- notebook dir: 
    - 'EDA' exploratory data analysis  on Yelp reviews
    - 'NLP-Glove_SentimentAnalysis' demonstrate a run of the model and results
- resutls dir: figures i.e. model training history
- setup.py, main file for running the project. 
  important: please note some variables like DATA_PATH, DATA_FILE, RESULT_PATH, EMBEDDING_PATH, EMBEDDING_VEC, EMBEDDING_DIM need to be modified.
- yelp dir: list of all python modules

Easy Approach:Run the Docker Container

1. Pull the Image from DockerHub:

docker pull majidmortazavi/yelp_review_sentiment_prediction_nlp

2. Run the Container locally:

docker run -t yelp_review_sentiment_prediction_nlp