Lei Feng network press: The author Xie Defu (micro-signal: beancurd191), more than 9 years of experience in the Internet industry, more than 6 years of personalized recommendations, big data-related experience. After graduating with Huawei for 6 years, he started his career in 2013, focusing on the application of big data technology in the fields of film and television, culture, video, and entertainment.
A few days ago rumors of being bought by Tencent should be no stranger to many people today, I am also a loyal user of today's headlines, why do I think today's headlines do better than other news / information clients, precisely because it shows to me content All I want to see is that with more and more behaviors in the above, the more accurate the content it shows to me, the core technology used in today’s headlines is personalized recommendation technology.
With the rise of the mobile Internet, many of the user's behavior has gradually shifted from the PC to the mobile. People spend more and more time on their mobile phones. People are using mobile phones anytime, anywhere, while using your mobile phone while you are in the car, when you go to the toilet, when you eat, and even when you are walking. Compared with the PC, the mobile terminal is characterized by a narrow screen, and the user's time is fragmented. At the same time, with the increasing amount of information, it is difficult for people to quickly find what they want from a large amount of information. This experience is very poor. If you are a product manager, if you are faced with the same problem, I hope that the following content will help you.
What is the recommendation engine?If you bought a book on Amazon, you may encounter this situation. When you choose to put a book into the shopping basket, it will automatically recommend other books for you. For example: The person who has purchased the book has also bought XXXX, guess that you may also like XXXX, combo recommendations, purchase the book and several other book combinations can enjoy a favorable combination price. These are all recommendation systems. In simple terms, the recommendation system is to study all user behaviors on the platform, perform portraits of users, and research content/products on the platform. At the same time the process of matching users and products.
The scope of application of the recommendation engine?The recommendation system has a wide range of applications in various fields, such as e-commerce sites, video sites, video broadcast platforms, news clients, literary websites, music websites, and so on. The following figure shows some application cases and application effects of the recommendation system on famous e-commerce websites and video websites.
Why is the recommendation system widely used in various fields?
1. Searching for desired content through a directory or search may require multiple screen scrolling on a small screen of the mobile terminal . The cost of finding interesting content is high, and the user experience is poor.
2. The content presented to the user through the recommendation system is of interest to users, and each user sees something different. Amazon’s CEO Bezos said that 1000 users who visit Amazon should see 1000. Different Amazon.
3, the current user's choice is very many, the choice of diversity and fragmentation of time , the user opens the phone, if not quickly find the content of interest, will soon leave.
4. Personalized recommendation technology uses algorithms to accurately recommend content that users are interested in. It helps users quickly find interesting content. When you finish reading a content, it will immediately recommend related things to you, which can increase user stickiness .
5, to help users find more high-quality long-tail content , the general platform users only limited access to the popular 10% of the content, a lot of content will never sink in the database no one found.
6. Help balance the ecology of the platform and avoid the Matthew effect . Popular content is always more exposed. Unpopular content has never been given the chance to be noticed, so that content production is ecologically polarized.
Recommended system architecture and core algorithmsThe following is an example of a product I have done before to explain, in the architecture, may be different when each is doing, but some of the core algorithms used, we should be similar. Specifically how to achieve, product managers do not need to pay attention to such fine, only need to understand the principle of it.
The commonly used algorithms in the recommender system include user preference algorithm, collaborative filter algorithm (item_base, user_base), association rule algorithm, clustering algorithm, content similarity algorithm (content_base) and some other supplementary algorithms. The final analysis results are the following:
1. According to the user preference algorithm, the content/product that the user is interested in is calculated and then recommended to the user.
2, according to the association rules algorithm, calculate the support and confidence between the items. The most common application is a combination purchase, and beer and diapers are very classic examples.
3. The item_base calculates the similarity between items based on collective user behavior, and then recommends the most similar item that the user has seen or purchased.
4, clustering algorithm can be based on user clustering , but also on the product clustering. After clustering, you can recommend for large categories or continue to calculate the relationship between user and product categories.
5, content_base is based on the properties of the item itself to perform correlation calculations , calculate the similarity between items, the most common application is the same type of recommendation.
6, user_base is based on the collective behavior of computing similarity between users , such as A and B are very similar to calculate, you can put B like the content, but A has not seen, recommended to A.
Common application scenariosHome guess you like to recommend
Since the screen of the mobile terminal is small and the content displayed on one screen is small, the user needs to turn down the screen one by one to find the content he is interested in. In this place, personalized recommended content can be quickly captured by the user.
Finding content in the section that is of interest to the user can surprise the user.
Associated/relevant recommendations for content details page: Users can be recommended content similar to the current content on the content details page.
End of Reading/End of Video Playout/End of Live Streaming Recommendation: Content similar to the current content is recommended.
Search page recommendation: When there is no result in the search, users can recommend content that they are interested in.
Personalized recommendation system is a very complex system, which involves not only the flexibility of data processing algorithms and system architecture, but also the problems of system robustness, data sparsity, cold start problems, system accuracy and diversity.
1. Garbage data processing : For the abnormal data and garbage data generated by the system, a set of cleaning rules is required for the business characteristics.
2, cold start problem : As new users do not have data precipitation, it is difficult to recommend according to user behavior, the most common method is to provide interest labels guide page when new users log in for the first time, guide the user to set, combined with other recommendations algorithm. Another more ideal method is to use the user's social data on other platforms.
3, data sparseness problem : You can use clustering algorithm to perform dimensionality-increasing operations, combined with other algorithms for combination recommendation.
4. The accuracy and diversity of the recommendation results : a combination of multiple algorithms is recommended. Ensure the accuracy and diversity of the recommended result set.
The above is a basic knowledge of the personalized recommendation system and I hope to help everyone.
This article is issued by Lei Feng network (search "Lei Feng network" public number attention) that everyone is a product manager. Reproduced without permission!