A recommender system is an information filtering system that predicts the ratings or preferences that users assign to entities like books, songs, movies, and other products.
The goal of a recommender system is to recommendation to users items or products that might interest them. The typical recommendation task is to predict the rating a user would give to an item. The Teradata Aster recommender functions are based on Collaborative Filtering (CF), which relies only on historical rankings of products by users to identify similarities between users and between products, and thus to identify products that are new to a particular user that the user would rate highly.
Teradata Aster provides these recommendation functions:
- The WSRecommender function is an item-based, collaborative filtering function that uses a weighted-sum algorithm to make recommendations. (For information about such algorithms, see "Item-Based Collaborative Filtering recommendation Algorithms." Badrul Sarwar, George Karypis, Joseph Konstan and John Riedl.) The function predicts the rating a user would give to an item by calculating the average of ratings the user has given similar items, weighted by a similarity score between the items.
- KNNRecommenderTrain and KNNRecommenderPredict take a similar approach to WSRecommender, but attempt to increase prediction accuracy by adjusting for systematic biases and replacing heuristic calculations of similarity coefficients with a global optimization that simultaneously estimates all weights. (For more information, see "Improved Neighborhood-based Collaborative Filtering." Robert M. Bell and Yehuda Koren.)