Why do you have to make personalized recommendations? Is the product manager too busy to do anything? Is it the end of the road for traffic? Is user experience sacrificed for the sake of OKR beauty? .

Hello everyone, I am the product manager of rare earth Mining community, and I am currently in charge of the transformation of personalized recommendation of information flow. Now the first stage of work has been completed and launched. In view of your feedback, I hereby come to share with you.

The transformation project of personalized recommendation of information flow was launched at the end of May this year, which lasted nearly half a year and was fully launched in the middle of October this year. After the launch, all kinds of feedback related to it flooded in, including praise — I felt pleasantly surprised for the exposure of my articles written before. There are unhappy – some old articles in the information stream of digging graves; But more criticism or ridicule of the voice, the core of ridicule is: old outdated and in power, the first new article is snow.

At the start of the feedback from the open experimental observation, he had received, every piece of feedback we have carefully recorded and included in the optimization of the next iteration, and personalized recommendation algorithm optimization is a long and sustained process, our iteration speed did not keep up with everybody’s expectations, cause a lot of digging friends feel sad or disappointed. Some people may feel that, well before, why do you have to do personalized recommendation? Is the product manager too busy to do anything? Is it for traffic and unscrupulous? Are you sacrificing user experience for your OWN OKR beauty?

When I see these doubts, I feel guilty, because the depth of love makes me feel responsible. I believe that every friend who puts forward his or her own ideas is holding great enthusiasm and high expectations for us. It’s also a good time to talk about why we’re doing this…

Why do personalized recommendations

For many old users, especially the creators who have been with us for a long time, they must have had this experience before, their hard work to write the article in just one or two days to get fast exposure, after the harvest of a certain amount of praise in the next time will only sporadic browsing and praise. Therefore, I would complain with my operation classmates whether I could give the author a continuous flow, not the peak of the first release. Indeed, we also recognize that, unlike information, technical content is less time-sensitive, and a lot of classic content has stood the test of time and is of long-term value to many developer audiences and the developer ecosystem.

However, at that time, the mechanism of relying on people to push, on the one hand, in the recommendation information flow of a single strategy, it is difficult to provide continuous flow for high-quality content, resulting in the result of the first is the peak; On the other hand, due to the influence of human factors, the recommended content is relatively single, which is difficult to meet the content consumption demands of a wider range of users. Based on the above background, we decided to introduce algorithm model for personalized recommendation; On the one hand, more high-quality articles can be distributed in a wider range, while obtaining the long tail of traffic exposure, so that these high-quality content has sustained value; On the other hand, it can satisfy diversified users’ personalized content demands.

We want to achieve a win-win situation among the three parties through this, and improve the efficiency of content distribution through personalized recommendation algorithm, so that users can obtain valuable content more efficiently, and creators can gain long-term benefits from their hard work, so as to continuously expand the volume of the platform. That’s why we’re doing this.

Current problems

Although the previous outlook is very good, at present, everything is in the initial stage, and the basic framework and model have just been built. Therefore, there are indeed many poor experiences: for content consumption, a large proportion of old articles in the information flow often have poor experience; For creators, it is difficult for new articles to get enough exposure in a short time, resulting in lack of creative motivation; These are issues that we are working on with high priority right now.

We’re working on it

Combined with the mid – and long-term planning of personalized recommendation and the current problems, we are carrying out the following work, which we would like to share with you here:

New article on cold start support strategies

For new content, due to the lack of browsing and interaction data, it is difficult to get the opportunity to show in the algorithm sorting, which also causes the current situation of new content is difficult to get exposure. Therefore support the strategy of the new content cold start is an essential part of the push, this strategy has been developed, it can introduce the logic behind: we will cut out part of the broader market flow Fielding drills designed specifically to new content, new articles to win after the initial flow, will be involved in the big sort algorithm. At present, this strategy is under experimental observation and will be online soon.

Algorithm side time attenuation strategy

In the ranking model, the strategy of time attenuation is continuously strengthened. The current ranking model is more based on the data of the contents being viewed as the key index to sort. The time factor does not get enough weight, so the time attenuation strategy is also being incorporated into the ranking model.

Creator Center optimization

At present, the Creators Center is developing and testing the statistical function of content exposure, aiming to provide you with a more intuitive content consumption data, so that creators can keep track of their content consumption. Maybe when you see this content, this function has been online.

Growth-oriented algorithms

Let algorithms serve the growth of developers. Continue to explore growth-oriented algorithm model, for everyone to mine for their current growth of high value content.

Community building

Open up more rights to creators. We need to continue to open up and allow more users to participate in the key decisions and governance of the content ecosystem of the community

Write in the last

Prior to joining the community, I am also a loyal users in the community, I know the adjustment to the impact of a large, deep hurt, a lot of people are already with the nuggets old friends for many years, we grew up watching the nuggets, therefore the present of blame, such as needle, every word of the language of a poke in my heart like a sword, in this sincerely apologize to you. With your trust and encouragement, I will humbly listen to the voice of digging friends and adopt. Personally, I am a marathon enthusiast. The biggest lesson I have learned from more than ten years of running experience is that it is appropriate to have a broad vision. It’s hard to run far when you’re fast, but you have a better chance of achieving long-term benefits by consistently exercising and producing consistent output. Algorithm optimization is the same as the technical community. We rely on creators to produce content and provide content for developers, so as to grow together. Only by taking each step well with both sides can we jointly build a more active and better developer ecosystem. I firmly believe that as the gold mining platform grows, it will benefit everyone on the platform and more people in the ecosystem!