User satisfaction
Describe the user satisfaction with the Recommendation results, which is the most important indicator of the recommendation system. Generally, users are obtained by questionnaire or online behavior data monitoring.
Prediction Accuracy
Describes the recommendation system's ability to predict user behavior. Generally, it is calculated based on the recommendation list provided by algorithms on offline datasets and the overlap of user behavior. The higher the weight, the higher the accuracy.
Coverage
Describes the recommendation system's ability to discover long tails of items. Generally, the ratio of all recommended items to the total items and the probability distribution of all items recommended are calculated. The larger the ratio, the more even the probability distribution, the higher the coverage rate.
Diversity
Describes whether the recommendation results in a recommendation system can cover different areas of interest. Generally, the similarity between items in the recommendation list is calculated based on the similarity between items. The less similar the items, the better the diversity.
Novelty
If you have never heard of most of the items in the recommendation list, it indicates that the recommendation system has a better novelty. The recommendation results can be obtained through the average popularity of the Recommendation results and the user questionnaire.
Surprise
If the recommendation result is not similar to the user's historical interests, but the user is satisfied, it can be said that this is a pleasant recommendation for the user. It can be measured by the similarity between recommendation results and users' historical interests and user satisfaction.
In shortA good recommendation system is based on the accuracy of recommendations, the recommended items for all users should be as extensive as possible (Mining Long tails), and the recommended items for a single user should cover as many categories as possible, at the same time, do not recommend too many hot items to users. The most amazing thing is to let users see that there is a feeling of "seeing each other" after recommendations.
Classification of Recommendation Systems