What is the typical use of a correlation matrix in recommender systems?

• A. To assign class labels
• B. To generate random recommendations
• C. To compute average price
• D. To measure similarity between items

Answer: (D)

The main idea being tested is how a correlation matrix is used to measure similarity between items in a recommender system. A correlation matrix holds pairwise correlation coefficients between items (or between users), showing how much their interactions tend to move together. This similarity information is exactly what you leverage in item-based collaborative filtering: when a user likes or interacts with one item, you look for other items with high positive correlation and recommend them because they tend to be liked by the same users.

High positive correlations indicate items that often appear together or are preferred by the same audience; negative correlations suggest items that tend not to be liked together, and values near zero imply little linear relationship. This helps build a set of neighboring items that are likely to appeal to the user.

The other choices don’t fit because assigning class labels is a supervised learning task, computing an average price is just a basic statistic, and generating random recommendations ignores the learned relationships captured by the correlations.