Collaborative filtering (CF) recommendation approaches Sample Clauses

Collaborative filtering (CF) recommendation approaches. Collaborative filtering (CF) is a recommendation technique that bases its predictions and recommendations on the past ratings or behavior of other system users (Xxxxxxxx 2011). The assumptions behind this technique are (Bernardes 2015): first, if users agree about the quality or relevance of some items, then they will likely agree about other items; and second, users’ preferences remain stable and consistent over time. The majority of CF algorithms operate by first generating predictions of the user’s preference and then produce their recommendations by ranking candidate items by predicted preferences. 1. A rating matrix R, stating the rating of users over different items. 2. A similarity function computing the similarity between two users (using their ratings), s: UxU → R. Some typical similarity functions are Xxxxxxx, Constrained Xxxxxxx, Xxxxxxxx and Cosine (Xxxxxxxx 2011). 3. A method for using similarities and ratings to generate predictions. In a nutshell, this method uses the similarity function s to compute a neighbourhood N ⊆U of neighbours of a specific user and later it combines the ratings of users in N to generate predictions for the preference for an item i of the current user. This is typically done by computing the weighted average of the neighbouring users’ ratings of item i. Item-item algorithms (e.g., (Verstrepen 2015)) are similar to user-user algorithms, but rather than using similarities between users’ rating behavior to predict preferences, they use similarities between the rating patterns of items. The assumption is that if two items tend to have the same users liking and disliking them, then they are similar and users are expected to have similar preferences for similar items. Similarly to user-user algorithms, item-item algorithms need: 1. A rating matrix. 2. A similarity function. 3. A method for using similarities and ratings to generate predictions. However, in contrast to user-user algorithms, in item-item algorithms the similarity and prediction method are based on items instead of users. In RS where the number of users exceeds the number of available items, item-based approaches are preferred since they provide more accurate recommendations, while being more computationally efficient and requiring less frequent updates (Ricci 2015). Several approaches have been proposed to improve memory-based algorithms: ● Xxxx (2012) uses weighting schemes to improve the similarity measures used in this type of RS and avoiding the bias of ...