What and why is VKontakte experimenting in the feed?

2024 Author: Malcolm Clapton | [email protected]. Last modified: 2023-12-17 03:44

How tests and research work in the social network team, what they are working on now and what happened to the likes.

Let's start with the important and figure out why you need a news feed and why something is changing in it again. In the feed, we communicate with friends and a variety of authors: entertainment public, business, media, bloggers, thematic media. The feed helps to find out the news, express opinions, discuss events and keep abreast of everything that is happening right now.

We have very ambitious tasks for the development of the tape:

• make an ideal service for communication between readers and authors;
• highlight the most important among the huge amount of information and show the news in the correct order;
• to create a convenient and fast interface for viewing texts and media, as well as communicating on all platforms.

There are two key sub-tasks: interesting content and user-friendly interface.

Experiments with content

We have hundreds of different models of smart feed and recommendations running in parallel - not all experiments are noticeable to users. Sometimes they differ in little things: we show 1% more content from friends than from publishers, and see how this affects the use of the news feed. Sometimes the tests are more radical: we rank situational content and what is always relevant in fundamentally different ways. This greatly changes the composition of the feed for readers. Some posts are relevant only for a few hours (media news, announcements, sports results), while other posts may be interesting for many days (reviews, articles, vacation photos).

Some experiments can be evaluated in a couple of days - then updates are quickly turned on for all users. Other tests, on the other hand, can take more than a month. User habits do not change so quickly, and it is important for us to evaluate the results in a few weeks.

In a complex product, it is impossible to immediately identify the ideal formula; this is the path of iterative improvements. We test hundreds of hypotheses step by step. For example:

• posts with a lot of comments are more interesting;
• reading a post for a long time is a more important signal than a like;
• posts added to bookmarks are as interesting as possible;
• longreads and short text posts should be shown at different times of the day for each user;
• posts commented by friends are more interesting than others.

We test all these hypotheses first of all on ourselves: we implement a new feature in the algorithm and start using it ourselves. If we like, we include users on the test group. We look at objective metrics, evaluate the statistical significance of the experiment. We draw conclusions: either we roll it out at all, or we refine it further. And so for several years since the launch of the smart tape. Thousands of different hypotheses and experiments. As a result of each, new thoughts and assumptions appear, there is still a huge list of ideas ahead.

Work on the interface

It seems that the news feed is nothing complicated and nothing to invent in terms of interfaces? But in fact, this is also a constant search for beauty.

This is how the post display has changed over the past years:

At the same time, posts in recommendations look fundamentally different. Why? There are completely different consumption patterns and accents in terms of perception. In the feed, we subscribed to specific authors, it is important for us who wrote the post, when, whether friends are discussing it. And is it possible to quickly respond, like, write a comment.

And in the recommendations, we do not know the author. Most likely, the content is important to us, so we generally represent posts in a different way:

The recommendations also included experiments. We showed the author of the post in different ways, put the like button in the compact mode of the feed, and so on. Even now, on iOS and Android, the interfaces are slightly different, as the experiments do not end there.

We look at what tasks users most often solve in the feed, and try to make the interface ideal for these tasks. And this is what we are working on now.

1. Ribbon without unnecessary counters

In the feed, you want to read the content itself as conveniently as possible and be less distracted by different counters. Therefore, within the framework of one experiment, we do not show the number of likes and reposts, we leave only comments. There is a hypothesis that this can positively affect the engagement and the amount of activity in the feed in general, and over time make the likes themselves more meaningful.

Why comments were left? It's not just a counter. Comments are also content, information that complements the post. The comment counter helps you figure out whether it's worth going to a separate post screen to read the discussion. Sometimes they are more interesting than the publication itself.

How will we evaluate the experiment? Based on objective metrics and user feedback. Sometimes we meet complaints, which is unusual, but there are also many positive reviews, which is really convenient when the feed is more concise and there is no overload of unnecessary information. But the interface was launched recently, it is too early to evaluate objective product metrics. There is always an addictive period at the beginning.

We plan to check other versions of the interface, which lack counters, but will focus on content. If objective metrics show an increase in activity in the feed (time, number of sessions, engagement, and so on) and the feedback from readers and authors is rather positive, then perhaps this version will be rolled out to everyone at some point.

2. Easy bookmarking

Another idea currently being tested is to bring the bookmark button to the ribbon screen. In communicating with users, we have heard the idea many times that likes can be used as a way to save a post "for later." For this, VKontakte has a more convenient tool than like - a bookmarking service. Therefore, in the experiment, we replace the post views counter in the feed with the add to bookmarks button. This test has already been launched, we are collecting opinions and metrics.

3. Compact display of posts

In addition to counters and action buttons, we have been dealing with the compactness of posts for quite some time. Unfortunately, there are many examples when a tape recording can take up two or three phone screens, which is objectively inconvenient. Therefore, we launched an experiment where we show not all attachments, but only key ones.

This approach has many advantages: the tape becomes visually pleasing and more convenient. But it severely breaks a number of familiar patterns. For example, it is convenient to take a survey and two photos to choose the best one. And if for such a post you hide the photo and leave only the survey, then all the convenience is lost. Therefore, the development of this experiment continues.

4. Speed of work

In addition to visually noticeable experiments, we are now actively working on the task of speeding up the operation of the interface itself. We are talking about the speed of loading the feed, media content, opening a screen with comments.

Our servers process tens of thousands of tape requests per second. Therefore, fast loading of the feed and the section of recommendations requires no less creativity and effort than an interface that is ideal in terms of the arrangement of elements. To improve such a plan, we have a separate team that creates special databases that are maximally optimized for our tasks.

In addition, we will implement a clever logic of multi-stage information caching. It includes a cache on the mobile client - to instantly download data received in previous sessions. And also - various caches on the server, which allow you to do a lot of calculations in advance.

5. Convenient comments

Much has changed in the comments over the past year. We made them two-tier - this was sorely lacking. But there is still a huge plan of obvious improvements ahead.

Discussions with dozens of different threads and hundreds of participants in the conversation are already difficult to view from the phone. Not so long ago, we launched a convenient ranking of comment threads: you can see the most interesting or the newest. But in general, we are still at the beginning of the path. The plans are to significantly improve the algorithms for identifying interesting comments and display them above, to implement updating and receiving comments and responses in real time. In addition, to make the interface as convenient as possible for communication in a dialogue-like format, and to implement notifications and interfaces that will keep abreast of the most interesting discussions for the user.

Conclusion

The devil is in the details. We work a lot on coverage, different formats, monetization opportunities, and more to make the content system as convenient as possible for authors and readers. But at the same time, we believe that in parallel with the overall strategy and fundamental launches, it is critically important to work through all the smallest details of the news feed and recommendations. Expect more experimentation and updates.