Recommendation Algorithms

As explained in the features list published earlier, the Recommendation System will be a very important feature for Shopster. One of the kinds of recommendations that it will offer, will be offers indirectly recommended by a user’s top friends. To find a user’s top friends, we are planning to introduce a measure of similarity for calculating top similar friends. This can be done using the Pearson Correlation coefficient which is a neighbourhood based algorithm for comparing past ratings on the same offers by both users and producing a value between -1 and +1, which would be indicative of the strength of their relationship. After that, the top rated offers of each top friend will be recommended to the current user, considering that he/she has not already rated them. These recommendations will be available to the user when he clicks on the “Recommendations” tab in the Search screen, as shown in this mock-up screen.

If the user, however, does not have any friends yet or is a new user to the system, the recommendations offered in that tab will be the top ratings of the most trusted users in the system. A user will generally have a high trust rating if his ratings on offers are similar to the actual average rating of an offer but also on the amount of helpful comments he/she has posted in combination with the ratings he/she has received on offers posted by that user.

Another simpler type of recommendation will be the top rated recommendations of all users and this is also the 1st tab shown in the Search screen mock-up. Offers from shops that a user has “liked” (or in other words, subscribed to) can also be accessed in the 3rd tab of the search screen.

The next also very important recommendation algorithm will be the recommendations that a user will receive when viewing a specific offer about a product. We can use the Item-to-Item Collaborative Filtering approach for recommending similar or related products to the products described in the offer currently being viewed (just like Amazon’s item-to-item recommendation – users who bought this also bought that, etc).

Leave a Reply