Overview of Recommendation methods used by Amazon and Google

This blog post discusses different recommendation methods, including the recommendation methods used by Amazon and Google. Wikipedia defines a recommendation system as a specific information filtering technology, it is trying to present users with information that may interest users (movies, music, books, news, pictures, webpages, etc ). Wikipedia also pointed out that recommendations are generally based on information projects (Content-based recommendation technology) or user-based social environments (collaborative filtering recommendation technology ). We think there is another way to personalize the technology that Google focuses on.

Xavier Vespa pointed out that Pandora uses "Deep Project Analysis" to generate recommendations. Strands uses "social analysis of projects" to form recommendations. Aggregate knowledge uses "hierarchical analysis and Behavior Analysis of Projects" to generate recommendations.

Two years ago, Alex listed four major recommendation methods he saw. The first is personalized recommendation, which is based on the user's historical behavior. The second is social recommendation, which is based on the historical behavior of the target user's neighbors. The third is content-based recommendation, the recommendation is generated based on the project itself. The fourth is the hybrid recommendation method, that is, the first three methods are used together.

Amazon: King of recommendations

There is no doubt that Amazon is the most authoritative case in the recommendation system field. After analysis, Amazon uses all of the above recommendation methods. Amazon's recommendation system is very complex, but all its recommendations are based on user behavior, coupled with the characteristics of the project and the behavior characteristics of other users. In short, the ultimate goal of the recommendation system is to allow users to add more products to their shopping cart.

However, emerging Internet companies are attempting to focus on a specific recommendation method. For example, Pandora's recommendation system uses in-depth project analysis (using the "Genetics" theory ). Strand has obtained enough funds from VC to build the best social recommendation system on the planet. The recommendation system of aggregate knowledge focuses more on behavior analysis methods.

Google: Focus on Personalized recommendations

There is no doubt that Google is the most successful internet company in this age. It is also using recommendation technology to improve its core search products. Google has two methods.

• Google will provide you with personalized search results based on your location and recent search activities;
• After you log on to your Google account, you will see more useful results related to your search keywords, which are obtained based on your search history.

Therefore, Google uses your location information and search history to generate search results, which is the magic of Personalized Recommendation technology. In fact, personalization has become a popular word for Google in recent years. However, Google also uses two other recommendations for its core search products.

• Google's search algorithm "PageRank" basically relies on Social recommendation. For example, WHO has linked a webpage;
• Google is also using content-based recommendations in the form of "you mean (Did you mean.

It is certain that Google has added other recommendation technologies to its search products. Not to mention products in other fields of Google, such as Google News, igoogle, and its business site Froogle, all of which present recommendation features.

Reposted from Diandian lab