In the recommendation system, in order to allow researchers to predict the results provide more value for users, we will focus on customer satisfaction. In view of the recommendation system in addition to allowing users to buy more similar products, but also must be "useful" for users, researchers will focus on the interactive experience on consumer and user experience when using the system. Currently, researchers are working to solve this problem by evaluating the different indicators, rather than simply by forecasting accuracy and machine learning techniques.
Recommended system performance should be measured by its value as user-generated. On the recommendation system of assessment, there are many indicators, such as coverage, novelty, diversity, the degree of surprise. These evaluation methods names vary.
Some scholars recommendation system novelty, relevance, such as the degree of surprise called "concept (concept)", some scholars called it "dimension (dimensions)", and some called it assessed as "recommendation system The method (measures of recommender system evaluation) ".
In this article, we will use the term "concept", when referring to evaluate different aspects of the recommendation system. After classification of the existing concept, we will be divided into six categories: practical, novelty, variety, singularity, coverage, and coverage of surprises. But there are some concepts not mentioned, such as: trust, risk, robustness, privacy, flexibility and scalability. For the convenience of readers, we put these big concepts presented in a different space.
Table 1 summarizes the notation used herein in all assessment indicators.
Practicality
There are many practical recommendation system of another name, such as relevance, usefulness, value and recommended user satisfaction. "Recommended System Manual" (Recommender Systems Handbook) believes that practical value on behalf of the user at the time of recommendation obtained. If you like to recommend the project, he / she receives a recommendation is useful. Practicality is further defined as consumption of users in order of preference. If you only consume their favorite things, it is recommended that these projects can help users quickly find the hearts of love, so as to achieve the recommended practicality.
As can be seen, most practical and user-defined desire to consume and user satisfaction hook. In this definition, the practical evaluation recommendation system should focus on the recommendation system generates the predicted user do react. We can assess the ratings by users after a given consumer goods, in order to measure the recommended system availability. If the recommendation results in order to bring the value of this approach appears to be desirable, but it comes to online assessment. And when it comes to off-line assessment, some scholars have suggested that the use of indicators to evaluate the accuracy based.
In this article, we use the notation