How to recommend content: 5 errors of Russian media

    image

    In addition to the b2c Surfingbird project , our company has a b2b Relap product . This is a media recommendation system. You could see our technology in action on Lifehacker , AdMe , RIA , COUB and other sites that you use every day. We help clients engage the audience in the consumption of content, using algorithms that have been developed for several years.

    We often encounter misconceptions and myths about what recommendations should be around the main content that the user comes for. In the article, we talk about the most common mistakes that media make when designing interfaces and how to do it correctly.

    Recommend by tags


    The biggest and most popular misconception. Most often, the media make recommendations at the end of the article on tags. This is what Look At Me and RBC do, for example. There is material with tags: tractor, Putin, cheese. Texts about a tractor, about Putin and cheese are displayed to him. At first glance, it’s logical:

    image
    image

    A similar recommendation mechanic in real life would look like this. You go to the grocery store. And put the butter in the basket. A consultant comes up to you with palms sweaty from excitement and says: “Oh, I see that you took the oil and that means you need oil. Take another five types of creamy village and sunflower and goat butter. ” The maximum that can happen out of the ordinary - you will be offered a transmission if you read something about cars. And it will already be considered rocket science.

    The world is a little more complicated. In fact, everything works like this: the user comes to the site, watches the dollar exchange rate, the cube from Mad Max and a video where someone throws chebureks. So, another person who has already looked at the dollar and Mad Max is likely to want to look at the flying cheburek, no matter how strange it sounds. This is called collaborative filtering. There are clusters of users and patterns by which you can maximize reader engagement in your content.

    According to the results of A / B tests, collaborative filtering without additional settings gives 20-30% more clicks than a selection by tags. And this means that no one should compose “Read Also” blocks based on tags.

    Recommend content from the same section


    The second incarnation of the previous fallacy is to segment recommendations by sections.

    Imagine that a consultant in the butter department takes you hostage with the words: “Today you will buy only butter, oh yes you will buy a lot of butter!” You scream open the door and never return to this store.

    We showed one group of users segmented recommendations: a person reads the news from the "Society" section - we recommend him articles only from the "Society" section. Another group received recommendations from the entire site (cross-segment recommendations). CTR of the widget with recommendations excluding sections is 2 times higher, the percentage of rejection is lower by 16% and the time spent on the site was 23% higher. It makes no sense to limit the reader to a single section. Be diverse in your recommendations.

    Recommend popular


    To recommend popular news is to recommend unnecessary news that everyone has already seen. Popular is content that has been watched by many people. That is how the popular becomes popular. Every time you want to make such a block on your website, remember this offer, because that’s how it works. This is the news that everyone has seen.

    In A / B tests, we compared popular and collaborative filtering. CTR widget with collaborative filtering is 7 times higher. This does not mean at all that our algorithms are so cool. This means that the block popular on the site sucks and is not needed at all by anyone. We understand that the Popular block on the site is a must have for most media, but it's time to say goodbye to this misconception.

    Do not set a time limit


    It is difficult to understand why news publications recommend news of six months ago. This approach gives rise to wild examples like this:

    image
    image

    Media thinks the user wants to know what happened on this topic before. No, he doesn’t want to! He is not interested, and nobody is interested.

    During tests on the news media, we found out the most optimal age for news that should be recommended to visitors. Three options were tested: 72, 48 and 24 hours from the date of publication. The test sample was 2.7 million readers and a month was spent. Most of our colleagues bet on 24 hours, because it seemed to them that the news became very outdated very quickly, and no one would read yesterday’s news. Slightly fewer people believed in 48 hours. Apparently, because not everyone has time to read the relevant news for the day and, most likely, they missed something yesterday and want to catch up. No one believed in 72 hours. Yes, 72 hours won. In this range, users find the most interesting materials and the blocks collected from such news are clickable 4.2% more than 48 hours and 10.9% more than 24 hours. Probably, this is because people do not have time (or simply do not want) to consume the entire amount of information that the media generate. Therefore, the news released the day before yesterday is still relevant for them. The exception is breaking news.

    Make recommendations with your hands


    “An algorithm cannot be smarter than a person and know the needs of an audience better than a professional” - this is the average argument of those who collect recommendations with their hands.

    We do not agree. People have learned to fly to the moon, land a satellite on a comet flying in space, electric cars, supercomputers, robots, a hadron collider and all that. What hints at the triumph of technology around us. And so no, the algorithm knows better. So that the story does not turn into dystopia, we’ll tell you how it works.

    Each user does a little editor work. One or two times a day goes to the resource, reads the news, and performs two or three internal transitions. The algorithm understands what the user likes and finds recommendations for him and for people with similar interests.

    We conducted an A / B test, where we compared the work of native media widgets with recommendations. In the first case, editors gathered recommendations. In the second, our automated widget worked. As a result, Relap increased the block CTR by 2 times.

    We are very glad if this article helped in the work of your editorial staff and in the design of interfaces. If you have questions, crazy hypotheses and ideas that you want to implement with us, write to lab@relap.io.

    If you want to use our technologies, engage readers and generally automate the recommendations of content on the site, we will be happy to help with this. Email hello@relap.io.

    Also popular now: