Shoes By Robots

Problem Statement

    I’m a big fan of Nike shoes, and when I was looking to buy some new shoes I decided to do so online, and I was stunned by how much variety existed. Now my personality is that I will spend hours pouring over shoes, color schemes, and reviews until I’m confident I get exactly what I want, even for a premium.
As such, for the past few months I’ve been working on a project to better utilize shoe reviews listed online. For now, I’ve nicknamed it ’Shoes By Robots’ as the goal is to use machine learning and existing data to provide better value for a business and improve customer experience through buying shoes.
    With some shoes having more than 1700 reviews (i.e. the Nike Air Force 1) I wanted to research how one could better provide business analytics on trends occurring in the reviews. I developed a basic proof of concept web application that can take a link, read the reviews, and provide basic sentiment analysis on the first 3 reviews. I believe this concept could be developed further to provide a better user experience on what characteristics a shoe has, provide recommendations to other similar shoes, and help better understand customer feedback on varying shoes.

Pain Points

How to provide better value to customers, and generate better company insights

Powerful Analytics

With Natural Language Processing and Word Embeddings, customers could be presented with a comprehensive summary of a shoes characteristics and review sentiments. Stars only tell a fraction of the story.

Uncover Trends

Which shoe line typically runs small? Do certain sizes tend to give more negative reviews than others? Leveraging already existing customer data will allow businesses to better understand where their products fall short of customer expectations, or exceed them.

Intelligent Reccomendations

Paired with customer purchasing data, build profiles of customers and be able to make custom shoe recommendations based on characteristics uncovered in reviews. Build a better online user shopping experience.

Front End:

Back End:

Future Steps

Data is the oil of the digital age…

What next? Unfortunately I wasn’t quite able to figure out how to scrape sufficient shoe reviews to collect a comprehensive enough dataset to really be useful for a recommendation system. However with access to shoe reviews, one could develop a comprehensive recommendation system using methods such as clustering, dimensionality reduction, and  non-negative matrix factorization to identify similar shoes. Paired with a customer profile and purchasing habits, one could build a recommender system to help customers leverage hundreds of reviews they certainly couldn’t leverage themself. Business insights could also gleam what common themes exist in shoes, both positive and negative, to better inform stakeholders.

If you’re interested in recommendation systems and don’t feel like skirting around robot.html files and collecting your own dataset, the Jester Dataset is a great place to start in building your own recommendation system.

Thanks to Demitri Haddad who helped me with some of the visual web design aspects

His personal website can be found here