Recommendation Algorithms for E-Commerce

Introduction Millions of products are sold over

the web. Choosing among so many options is proving challenging for consumers. Recommender systems have emerged has a response to this problem.

Collaborative filtering(Cf) Works by building a database of

preferences for products by customers. A new customer is matched to discover ‘neighbors’, who have the same tastes. Then the products that these customers bought are recommended to the new users.

Problems with Cf The two conflicting issues related to this

technique are scalability and accuracy. While these techniques are good enough

for neighbor discovery in databases where the number of customers is in some thousands, they however fail, when the sizes reach to the order of millions. Also as the size of each record increases with more data points to be considered, the problem is increased.

Problems with Cf continued… Another issue is of accuracy. In the

context of predictions we have 2 error conditions, one a false positive, where the system recommends a product the user eventually doesn’t like, and a false negative, where the system assumes that a product will not be liked by a user when it not so in reality. It is much more dangerous to have a false positive, because that will lead angry customers!

Problems with Cf continued… These two issues are conflicting in

the sense that in order to be fast a recommender system may not search exhaustively through the database and thus increase the chances for an error.

Recommender Systems based on Cf There are three phases of operation: Representation: In a typical CF-based

recommender system, the input data is a collection of historical purchasing transactions of n customers on m products. It is usually represented as an mxn customer-product matrix, R, such that ri;j is one if the ith customer has purchased the jth product, and zero, otherwise. We term this m n representation of the input data set as original representation.

Recommender Systems based on Cf Neighborhood formation:The most

important step in CF-based recommender systems is that of computing the similarity between customers as it is used to form a proximity based neighborhood between a target customer and a number of like-minded customers. The neighborhood formation process is in fact the model-building or learning process for a recommender system algorithm.

The third step is recommendation generation