What this Algorithm is Used For
Use this algorithm to make recommendations on products, services, etc. to users based on their similarity to previous users. This is an easy way to improve the effectiveness of an online commerce presence.
Common Use Cases
Ecommerce - Recommending new products or services for e-commerce
Online Gaming - Recommending virtual goods for online gaming
Finance - Recommend new financial products and services
User based recommenders are a type of collaborative filtering algorithm. They make recommendations based on similarity of users. They use a defined statistical method (Log Likelihood, Pearson Correlation, etc.) to determine the similarity of a new user to previous users based on a set of characteristics or "features". To get a recommendation for a new user, it will look at what items are preferred by similar users and make a recommendation for the new user accordingly (If user A and B are similar, and B likes item 1 from a set of 10 items, then user A would receive a recommendation for 1 over other items in the set). Every time a new user buys a product or service, that information is fed back into the algorithm which increases its predictive accuracy over time.
Note: User based recommenders work better for smaller data sets ( < 10 million records). For larger data sets or faster recommendations consider using item based recommendations.
Data Input and Output
Data Formats Accepted:
Columns Required in Data File:
1. Sample data set with Table of users with at least two characteristics defined per user. This data is used to train the algorithm.
2. New data set with user profiles
Collaborative Filtering, Recommendations, Machine Learning