10 lessons from the recommended system

1. Make sure you really need a recommendation system . The recommendation system is only necessary when the user encounters information overload . If you don't have too many items on your website , Or users' interests are relatively single , Then maybe you don't need a recommendation system . So don't dwell on the word recommender , Don't make a recommendation system just to make a recommendation system , It should be from the user's point of view , Design a system that can really help users discover content , Whether the system algorithm is complex or not , As long as it can really help users , Is a good system .
2. Determine the relationship between business objectives and customer satisfaction . A good recommendation system for users does not represent a commercially useful recommendation system , Therefore, we should first determine the gap between the user satisfied recommendation system and the commercial demand . Generally speaking , Sometimes user satisfaction does not coincide with business needs . But in general , Customer satisfaction is always in the long-term interests of the enterprise , Therefore, the main point of this article is to balance the relationship between the long-term interests and short-term interests of the enterprise .
3. Choose the right developer . Generally speaking , If it's a big company , You should hire your own developers to specialize in the development of recommendation systems .
4. Forget the cold start problem . Keep innovating , There is any data you want on the Internet . As long as users like your product product , They will continue to contribute new data .
5. Balance the relationship between data and algorithm . Using the right user data is crucial to the recommendation system . A deep understanding of user behavior data is a necessary condition for designing a good recommendation system , Therefore, the analysis of data is the most important part of the design system . Data analysis determines how the model is designed , The algorithm only determines how to optimize the model .
6. It's easy to find relevant items , But it is difficult to show them to users when and how . Don't recommend for the sake of recommendation .
7. Don't waste time calculating users with similar interests , You can use social network data directly .
8. It is necessary to continuously improve the scalability of the algorithm .
9. Select the appropriate user feedback method .
10. Design a reasonable evaluation system , Always pay attention to the performance of the recommendation system .