This article mainly introduces Taobao's online recommendation strategy, including how to collect and predict user behavior to achieve more accurate recommendations, and the importance of this strategy to improve user experience and increase conversion rate.
Today, I will briefly talk to you about a Taobao strategy.
I don't know if you have found such a case when you usually visit Taobao: when you enter the Taobao homepage, click on a product from the product feed flow on the homepage, and enter the product details page to browse for a certain time.
When you return to the homepage again, you will find that a card will move up at the bottom of the product slot just now, and the content in this card is very "similar" to the product you just browsed.
You can experience it for yourself.
In addition to Taobao, JD.com, and Meituan all have similar front-end interactions. What is this?
In fact, it is a typical real-time recommendation.
We all know that the core of recommendations is to show users results based on their interests, but don't lose sight of the fact that users' interests can change over time.
I used to give an example: a guy and his girlfriend have a quarrel, so he buys a women's bag in order to make his girlfriend happy, this kind of scene is all too common.
Therefore, how to quickly and accurately capture the near-"real-time" needs of users is very important, and it is also the core means to improve the referral transaction rate.
Obviously, the traditional recommendation based on offline data is obviously not satisfied, and this is actually the quasi-real-time recommendation based on the user's current session behavior.
How? I've drawn a business process to illustrate the solution to this problem
So, you can actually see the key points according to the above process:
The collection of user behavior, of course, is more about predicting the user's real-time preferences, for example, it can be predicted according to the attributes of the products (category, brand, price, etc.) that are currently browsed;
Based on real-time preference prediction, calculate the set of goods that may be of interest, which can be calculated in real time or generated based on offline similar item sets;
The end is rearranged, and the front end gives users real-time feedback.
Of course, there are many other "real-time" recommendations, which are to add the behavioral data that occurred when you visited the app once or several times into the recommendation system into the policy logic of the recommendation system, and the impact of these behaviors on the recommendation results will be reflected in the next user visiting the app.
This kind of recommendation is sometimes defined as real-time recommendation, but the timeliness is obviously weaker than the scheme mentioned above, so it is sometimes called this kind of quasi-real-time recommendation.
This article was written by Everyone is a Product Manager Author [Xia Bluff People], WeChat public account: [Strategic Product Xia Master], original / authorized Published in Everyone is a product manager, without permission, it is forbidden to reprint.
Image from Unsplash, based on the CC0 license.