A Product Recommendation Algorithm Based on the Expectation of a Not Scored Item by the Prospective User

With its unique advantages, the traditional collaborative filtering algorithm has been one of the original and most successful methods in product recommendation. However, the intensified sparsity of the matrix has caused a decreasing precision of particular recommendations, which the traditional CF algorithm is of little help. By making statistical analysis about users´ selected items, this paper will estimate the distribution of scores they give to a product from two aspects: the probability of choosing the item that has not been scored as well as the preference of that item. We establish a product recommendation algorithm based on the expectation of an item not been scored by the prospective user. By our algorithm, a recommended set of items will be generated when the score expectations of all items are calculated and ordered. Then empirical analysis on the data set of Movie-Lens recommendation system shows that the algorithm developed in this paper greatly improve the precision of the recommendation while the amount of calculation is not much expanded.