Skip to content

A chrome extension to get the authenticity score of any amazon product

Notifications You must be signed in to change notification settings

yuvisidhu19/DTekt

Repository files navigation

Description

DTekt is a chrome extension that can identify counterfeit products on Amazon using sentiment analysis of the product reviews by only "verified purchase" buyers i.e. the review of the customers who have actually bought the product. Sentiment analysis applies a mix of statistics, natural language processing (NLP), and machine learning to identify and extract subjective information from text files, for instance, a reviewer’s judgments and assessments about a particular product. This extension can scrape the website and collect the necessary data which is further fed to a pre-trained ML (Bert) model, which in turn, returns a rating from 1 to 100. The higher the rating, the higher is the product's authenticity.

Demo: https://www.youtube.com/watch?v=xn-f6er939U&ab_channel=JashandeepSingh

Methodology

Two types of dataset are used to create models. One dataset (original.txt and fake.txt) is used to classify whether the given text implies if the product is a counterfeit or not. But due to lack of data, another dataset (positive.txt and negative.txt) was used which classifies if the given text implies a positive emotion or a negative emotion. Sometimes, a negative comment can imply that the customer is not satisfied with a product or the product is of poor quality, which might be due to the fact that the product is a counterfeit. So, two bert models have been trained to produce a negativeness score and a fakeness score and both are combined in a way, providing more weight to fakeness score and less weight to negativeness score, to generate a final score which implies the authenticity of the product.

Formula used

counterfeit_score = fakeness * 0.7 + negativeness * 0.3

authenticity_score =  round((1 - counterfeit_score)*100, 2)

Screenshot

WhatsApp Image 2022-07-31 at 7 07 31 PM

Releases

No releases published

Packages

No packages published

Languages