Introduction

Amazon.com is world wide e-commerce retailer that sells a huge range of products online. Many ranking show amazon is second biggest retailer in US in 2022, how can amazon become so successful? People all know amazon brings superior customer service, with blazing fast delivery and risk free return option, customer service is always amazon’s first priority. But what else make amazon so different besides its service? There are a couple of differenct perspectives here. Recommendation system and ads definitely plays an important part there.

In 2015, an experiment was conducted that performing recommendation A visual dataset of styles and substitutes in amazon store. (https://cseweb.ucsd.edu/~jmcauley/pdfs/sigir15.pdf). In 2016, Ruining and Julian developed models to Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering. (https://cseweb.ucsd.edu/~jmcauley/pdfs/www16a.pdf). Attach below is a sample review crawlered from amazon website, showing related product info, customer’s rating, review text details.

Clustering and ARM are common ways to categorize data and find the strengh of the relationship between different products. Clustering in a retail store can be helpful for helping customers find products efficiently, also help to boost the store saling. While ARM can tell the common products customer buy together, so that store manager can put these stuffs in same place, and kindly remind the consumer to bring them at the same time. However, these stragey looks simple, to apply and make good use of them usually takes time and efforts.

\

Naive Bayes and decision trees are efficient machine learning classifiers based on Bayes’ thoerem, by applying algorithm on product rating, the store can predict customers satisfication with exisitng and upcoming products, and use that as the base for promotion. On the other hand, they can also be applied in evaluation systems, to decide how goodness the product is and whether to keep them.

SVMs are widely used classification methods in both industry and academic, it is particularly effective when dealing with huge dataset with large number of features, by applying different kernel functions, SVM can create complex and non-linear hyperplane, and robust to noise. However SVM also has essential drawbacks for sensitive to outliers and certain, the effectness of model needs to be evaluated.

10 questions to answer about Retail

• How to gather retail data
• How to process the missing / incorrect data
• How to do data normalization
• How to find the production relation
• How to classify the product based on category
• How to cluster the products for saleing purpose
• How to personally recommend the product to the user
• How to apply svm to retail
• How to find common trend in transcation
• How to label the product for saling purpose