Customer-Review-Sentiment-Analysis

A sentiment analysis model for product reviews by customers.

Review Dataset

The dataset consists of reviews written by customers for various food products. The reviews, originally given on a 5 point scale, have been adjusted to a +1 or -1 scale, representing a positive or negative review, respectively. Here are two examples from the dataset:

Review 1 – Label -1 (Negative sentiment)

Nasty No flavor. The candy is just red, No flavor. Just plan and chewy. I would never buy them again

Review 2 – Label 1 (Positive sentiment)

YUMMY! You would never guess that they’re sugar-free and it’s so great that you can eat 1 them pretty much guilt free! i was so impressed that i’ve ordered some for myself (w dark chocolate) to take to the office. These are just EXCELLENT

Requirements

There are some general library requirements for the project and some which are specific to individual methods. The general requirements are as follows.

• numpy
• scikit-learn
• nltk

The library requirements specific to some methods are:

• Pytorch

Note: It is recommended to use Anaconda distribution of Python.

Approach

Feature Extraction

• GloVe Embeddings
• Sequence Padding

Model Architecture

Three different model architecutres were used to develop different classifier;

• Embedding Bag Model
• Average Embedding Model
• Bidirectional Long Short-Term Memory Network (BiLSTM)

The performance (accuracy) of the three architectures were compared and BiLSTM had the highest performance - the test accuracy is 0.858.

Information about files

• data/reviews_train.tsv: Train set of reviews
• data/reviews_val.tsv: validation set of reviews
• data/reviews_test.tsv: test set of reviews