Recommendation framework: Improving automated collaborative filtering by trusted category

Abstract

The key idea behind user user automated collaborative filtering is that it predicts item's rating based on similar users who share the same taste by rating item similarly. Automated collaborative filtering (acf) is proposed on a hypothesis that users with similar rating will also have similar rating in everything. People often agree on an idea, or on a group of ideas but it is very rare that agree on everything. So in this thesis we suggest that this is one of the main causes behind huge noises while working with a large number of neighbors in acf. In this thesis we are proposing a method where at the time of calculating acf we will only consider a subset of products based on their category that we trust based on users previous ratings. Usually when a developer has to build a recommendation system he has to start writing code from scratch like in the early days of web development, which consumes a lot of time and development energy so we tried to build a framework for our updated acf that provides the building blocks for a recommendation system as APIs. By describing models of their recommender in simple JSON format, within a few commands and using provided API support, programmers can easily get a customizable generic service running. All the input and output communications are done using RESTful APIs so that the system can communicate with any part of the whole system.