Client Sentiment Analyzer

About

Tiny app to check client sentiment based on product review 💸.

Core ML model has been taught using 12 000 client reviews from Amazon US marketplace, category: Toys.

Contents

🧸 Client Sentiment Xcode Project

Client-Sentiment-Analysis.xcodeproj

A simple application written using SwiftUI to demonstrate the operation of a created neural network model. The project contains the learned AI model, which was created using Create ML. App is ready to use on newer versions of MacOS, iPadOS and iOS. Scales well to the screen aspect ratio

✂️ Jupyter Notebook

client_sentiment.ipynb

This notebook contains Python script to prepare records stored in .csv file using Pandas library. Every record then is stored in separate .txt file, to suit Create ML demand.

The data I used is available at Amazon Customer Reviews Dataset webpage.

🏗 Model Generator

ClientSentimentClassifier.mlproj

Apple Create ML project to generate text classifier model.

Model in project has been trained using 12 000 records classified from range 1 - 5, validated with 4 000 records and tested with 4 000 others. At first glance, the model may seem poorly taught, but after a deeper diagnosis, it can be concluded that about 90% of the answers are well classified or classified into an adjacent class. Even a human 💁‍♂️ would have a problem with the correct classification of the review and the person who wrote this review may not have given the number of stars consistent with the opinion.

To sum up, the model works very well - try it!

🔎 Model Checker

checkTraining.playground

A simple playground to check how many % of queries the model has classified correctly and one class more or less.

Usage

1. Clone repository
2. Prepare your own dataset with star_rating and review_body columns.
3. Sieve data using Python script.
4. Generate Core ML model.
5. Import it to the app.
6. Have fun!