Using PageRank to Recommend Products
In addition to searching, PageRank can also be used for recommendation systems. When someone visits a website that is selling some sort of product there is usually a recommendation section. Rather than just giving related products, PageRank can be used to suggest new products for the user based on other factors. The notes use Netflix as an example where the user is someone using the website and the products are the movies. Their method was to say that “a movie is relevant for me if other similar people liked it and a person is similar to me if they like movies that are relevant to me.” This way of recommending items can also be seen on other websites like Amazon. Amazon has a section for each item that is called “Customers Who Bought This Item Also Bought.” This section on Amazon has the same concept as the approach for recommending movies in the Netflix example.
This is similar to the way we ranked webpages in class. The relationship between movies and users is circular just like the relationship between authority scores and hub scores. For ranking webpages, a webpage’s authority score was the sum of the hub scores pointing to it and the webpage’s hub score was the sum of the authority scores of the webpages it is pointing to. For the Netflix example, a user’s relevant movie score depends on the sum of the similar user scores of people who also like that movie. The similar user score depends on the sum of the relevant movie scores for movies the user likes. After calculating the scores they are then normalized. Because Netflix is trying to sell their products (movies) and is not a social networking website, the relevant movie scores can be used to recommend other movies and the similar user scores could come into use if the website wants to add in a friends feature. PageRank provides a good way to recommend products and also conveniently provides scores that can be used to recommend friends.
Sources: http://web.stanford.edu/class/msande233/handouts/lecture8.pdf