editor's introduction: In the process of information dissemination, the algorithm and recommendation are monopolistic and unstable, and can easily create an information cage for ordinary users. Different people have different attitudes towards algorithm recommendations when dealing with thousands of people. Some people are happy and some are worried. Where will algorithmic recommendation go in the future? Let’s take a look!
"It is indeed a luxury for human beings to remain sane forever." In " The Wandering Earth ", MOSS didn't understand why astronaut Liu Peiqiang was willing to die until his destruction.
MOSS is a planetary engine that humans have been able to create to propel the navigation of the earth. Technology has reached a new level of advanced artificial intelligence, but it is still far away from artificial intelligence with independent intelligence like 007. In essence, it is different from the current through data The algorithm recommendations that bring optimal solutions to users are the same across layers, algorithm layers, and application layers.
01 What is algorithm recommendation
When talking about algorithm recommendation, the first word that may flash in most people's minds is " today's headlines ". Indeed, products such as Toutiao, , Douyin, and owned by ByteDance, empowered by recommendation algorithms, have left a deep impression on the public.
In fact, in addition to ByteDance, Alibaba has widely used recommendation algorithms in Tmall , Taobao, NetEase in NetEase Cloud Music, and Bilibili video recommendation streams. It is no exaggeration to say that in the Internet industry, as long as it is not direct information that users actively seek, there are even recommendation algorithms behind advertisements.
In 1994, the GroupLens research group of the University of Minnesota launched the first automated recommendation system, GroupLens, and proposed collaborative filtering as an important technology for the recommendation system. It was also one of the earliest automated collaborative filtering recommendation systems.
In layman’s terms, algorithm recommendation means that APP uses big data to scientifically “tell fortunes” to users.
First, the back-end system will collect and collect statistics on all information through the registration information and user click behavior of the application layer; then it will classify the information through the relevant algorithms of the strategy layer and outline the user portrait; finally, the strategy layer will compare it with other information on the platform. Match the user portrait and present the results to the user's eyes at the application layer.
This process is like fortune telling. You tell the fortune teller your birth date and the fortune teller , and then the fortune teller will give you the answer you want based on your birth date and related laws.
Data source: amazon According to different trigger conditions, there are currently two types of algorithm recommendation systems:
- One type is passively triggered, requiring the user to delineate certain restrictions, and the system will recommend the optimal solution to you;
- The other type is actively triggered , as soon as you open the APP, the system will automatically recommend content to you, without the user setting conditions. Among the passively triggered algorithm recommendations of
, 58 same-city can be regarded as a classic case.
For example, when renting a house, 58.com will make statistics at the data layer based on the user's purchase conditions, such as price, location, apartment type, etc. Then the strategy layer will classify the characteristics of the housing and assign a weight to each characteristic, and then through User data and property characteristics are combined to form recall data. Finally, the recall data gives priority to the recall data with high weight at the application layer until it completely violates the user's requirements.
Data source: "The Evolution and Practice of 58 City Intelligent Recommendation System" actively triggers algorithm recommendations. Toutiao-based Douyin must have a name.
Douyin is different from the multi-category and multi-level complex information flow of 58.com. The purpose of Douyin is to let information find people, and people are the traffic pool of the platform.
Therefore, Douyin will first count the videos uploaded by users, and then classify the videos into the content traffic pool through keywords. At the same time, the system background will count and classify the user's behavioral keywords to outline the user Portrait, and then match the user portrait with the content of the video traffic pool, and finally distribute content that the user is more interested in. In this process, there is almost no need for users to actively filter.
02 Why algorithm recommendation easily creates an information cage
Chapter 58 of "Laozi" says: Misfortunes are where blessings depend; blessings are where misfortunes lie. As mentioned earlier, whether it is a passively triggered algorithm recommendation or an actively triggered algorithm recommendation, its main purpose is to reduce the efficiency of users in obtaining information.
For example, when we watched long videos before, we either looked for videos by categories such as movies, TV series, variety shows, etc., or we directly searched for the title of the drama and watched it directly. Anyway, it was very troublesome to find a video that suits our taste. In the era of short videos dominated by algorithm recommendations, watching videos all the time has become the norm.
But while algorithm recommendation improves the efficiency of information acquisition, it also puts us in an information cage.
On October 10, 2017, another busy Monday, the " Washington Post " reported a piece of news that made the American people extremely angry. Thousands of ads placed by Facebook during the US presidential election affected the election, and even Also revealed is " colludes with Russia and ".
According to reports, Cambridge Data obtained the data of 50 million Facebook users, used the Five Forces Model of personality to create advertisements, and then used Facebook’s algorithm recommendation to ultimately achieve the purpose of influencing the US election. The algorithm recommendation became an accomplice.
On the one hand, algorithm recommendation occupies a dominant position in information dissemination.
In 135 BC, the king of Yelang Kingdom, who was supposed to promote his country's prestige in front of the Han envoys, became an eternal laughing stock because he was bigger than the Han Dynasty. The reason was not only because the king lived deep in the palace and had no information, but also because Because the peripheral ministers had long recommended " Yelang is the best in the world", the king was convinced.
From the information of CNKI " Recommendation System ", we can intuitively see that the algorithm recommendation is the ministers around King Yelang. Although they are all talented and speak well, the information they convey is very limited, and The information conveyed is relatively simple. For example, if you like watching funny videos, the algorithm recommendation will recommend 7 out of 10 videos to you, and only the rest will be expanded to other videos. This is why Facebook will affect the presidential election.
On the other hand, algorithm recommendation is also unstable when information is disseminated.
algorithm recommendation system is an information distribution system designed by programmer . In the final analysis, it has not escaped the scope of machine distribution. Considering the current development level of artificial intelligence, it is relatively easy to exploit the loopholes of algorithm recommendation. .
In this way, those third parties who have mastered the platform algorithm rules will deliver more private goods. Aren't people surfing in the ocean of spam every day?
After all, now as long as you enter the keywords recommended by a certain platform and algorithm into a search engine, tens of millions of relevant information will appear to help you exploit the loopholes recommended by the algorithm.
03 Where should algorithm recommendations go?
There are a thousand Hamlets in the hearts of a thousand readers. Different people have different attitudes towards algorithm recommendations. In front of algorithm recommendations, some people get bonuses and praise them, while others are addicted to them. , and because survivor stories deify it, superstitious algorithm recommendations are becoming a common phenomenon in the Internet industry.
But as mentioned earlier, algorithm recommendation is monopolistic and unstable in the process of information dissemination, and can easily create an information cage for ordinary users. So where will such an imperfect algorithm recommendation go in the future?
First, algorithm recommendation parallelization
Parallel and serial are two different ways of data communication and transmission. According to the report "In the Big Data Era of Information Overload, How to Build a Big Data Recommendation System, and Where Are the Trends" from the Big Data Journal, the traditional Most of the algorithm recommendation systems adopt a serial transmission method. The advantage of this transmission method is that it is suitable for long-distance transmission, but it can only transmit one data unit at a time, so it limits the algorithm recommendation in obtaining user data and feedback information to the user. The amount of data deepens the possibility of information cage.
As for the parallelization of algorithm recommendation, we can use the feature of parallel transmission of multiple data units at a time to obtain more information and outline a more accurate user portrait. At the same time, more information can be fed back to reduce the information cage formed by algorithm recommendation with enough information. possibility.
Second, the algorithm recommendation gradually introduces new parameters such as humanities and social sciences.
mentioned earlier that because machines do not have human biology and thinking, current algorithm recommendations are faced with the situation of bad information taking advantage of loopholes. The main reason for this situation is that machines lack humanistic emotional judgment standards.
Therefore, in the future, algorithm recommendation needs to implant nodes of "thinking innocently" and "don't do to others what you don't want others to do to you" in the neurons of the recommendation system, collaborative filtering, and finally feedback better information to users.
For example, when a user accidentally sees a "cat abuse video" on a video website that exploits a loophole and enters the traffic pool, then algorithm recommendations without humanities and social sciences as new parameters will continue to mine animal abuse in the traffic pool. videos, so in addition to mathematical parameters such as click-through rate, content tags, and collections, humanities and social science parameters are also added.
In fact, no matter how the algorithm recommendation develops, it is just a condiment to assist humans in obtaining information. Don’t be superstitious and mythical about algorithm recommendations. I don’t want to become a vassal of the machine in 30 years.
This article is published by Everyone is a Product Manager cooperative media @Community Marketing Research Institute and is authorized to be published on Everyone is a Product Manager. Reprinting without permission is prohibited. The
title picture comes from Unsplash and is based on the CC0 license.
