To maximize profits for an e-commerce company, by provided purchasing suggestions to users.
Easy-to-use Machine-Learning (ML) applications that gives Data Driven suggestions to users by analyzing there average purchase cycles for previously bought products. A simple "Buy-it-again" design.
People make decisions by minimizing costs. While cost are mostly thought of as monetary, behavioral economic research shows that effort is a driving force in decision-making. People prefer to take the 'path of least resistance'. This application is design to oblige to the user's low effort preference.
Example: A person buys paper towels every week. My model picks up on this trend, and can be formatted to remind the user that she needs more towels after one week.
Below is the general architecture of the wsb-scrapper.
located at ncr_project/src/app/services/model_BasedOnPastPurchases.py.
The model takes in two parameters, account_number_hash and current_date.
Gets transactions and items in those transactions based of an account_number_id. Then for each unique item
it finds the average days between the reoccurring purchases, called avy_cycle. If the distance between the current date
and the last-date-of purchase is equal to to greater than the average_cycle, then it returns
that item ID.
The returned item ids are the application's suggestions.
for mac/linux users, clone repo then run
sh build.shThe given data was on for a month, I believe this application has far greater value when there is more time periods. One reason for this is it will increase degrees-of-freedom (df). With more dfs we can start running regressions to predict the next cycle, instead of just calculating average_cycle. An example model would be average_cycle on quantity of good purchased(t-1) and month of year.
Due to the time constraint, the algo is not at optimal speed. I want to decrease runtime. This is especially important if more data is given.
Also, I want to explore unsupervised/supervised learning to find better categories for each item. The categories given, had certain issues. Once I have better categories, I can start to develop a Neural Network to suggest new items. I have experience in PyTorch and wanted to use it, but that is a very time-consuming method.