The media’s algorithm is a black core. We don’t know what it is, and even if we were told, we wouldn’t understand it. But we can look at some basic things, like doing homework at school, listening to lectures in class, previewing before class, and reviewing after class. Everyone’s specific learning methods are different, but these basic logics are common.
I have collected some definitions of quality from Toutiao, Tencent, and Kuaishou, and sorted out some possible influencing factors. You may know these things, but you may not be so sure or know them so comprehensively. I think it is worth a look for optimizers.
We mainly focus on headlines, but other media can also be used as reference. The account levels are named according to the headlines of “Account-Group-Plan-Creativity”.
1. Money
The algorithm decides to whom the quantity is distributed, and the core of who the quantity is distributed to is money . But the money will be divided into two parts: one is direct advertising revenue , and the other is user experience , which is future money. Only when the user experience is good and users are willing to stay on the APP can the media make money for a long time.
Therefore, when the media distributes traffic, user experience must be an important consideration. User experience can be divided into 3 parts:Interactive behaviors : like, thumbs up, comment, like, shareViewing behavior : clicks, reading percentage, completion, video playback time, effective playback rate, completion rateComplaint behavior : dislike, report If users give more positive feedback (positive comments, longer viewing time, etc.), then the traffic allocation for the advertisement will definitely be increased; if users have negative behavior, it will also suppress the advertisement. So "Will complaints about ads affect the volume?"
Yes, the algorithm must have weights for the complaint rate and the cumulative number of complaints.
2. Simplified process of advertising Advertising is the process of media selling traffic, which can be simply divided into three steps: ① Confirm you want to buy ② Depends on how much money you give ③ Comprehensive consideration of user experience Everything is OK, and it can be displayed to the user. Corresponding to the advertising system, each step corresponds to some settings:
Let’s look at it in detail:
1. Confirm the investment status To confirm the investment status, you need to check several items: First, the switch must be turned on, and then the date and time period must include the current time point, and whether there is available budget and balance. There is a saying about available budget and remaining balance: take the minimum. a) The account budget, group budget and planned budget take the minimum value; b) Minimum budget - Amount consumed = Available budget c) Minimum available budget and account balance After confirming that the "minimum value of available budget and account balance" is consumable, you can enter the investing state. If the available budget or balance is about to be exhausted, the ad will be phased out. Several main factors to confirm the investment status:2. Estimate eCPM To estimate eCPM, go back to the formula for calculating eCPM: Bid by click: eCPM = (click bid) × (impression-click rate) Bid by conversion: eCPM = (conversion bid) × (impression-click rate) × (click-conversion rate) To estimate ecpm, we will use the concept of "quality". Let's take a look at the factors that influence the quality of each media: quick worker Tencent Headlines The influencing factors can be divided into several categories:a) Check account : historical performance of the account, credit of the promoted merchantb) Creativity : click-through rate, cover click-through rate, effective play rate (mainly), creative history click-through rate (mainly), freshnessc) Landing page : loading speed of the landing page, duration of stay on the landing page, relevance of the landing paged) Targeting : Relevance (Tencent refers to the relevance between creativity and users) The factors mentioned here are almost at the level of "doing homework at school, listening to lectures in class, previewing before class, and reviewing after class". They are very basic, and the media may not completely overlap, but the ones that can be put here are the basic points and are worth referring to. Let’s take a closer look at each one in turn.a) Check household registration Before allocating traffic, you must first "check your account" to see how much you consumed yesterday and whether you have received any complaints before. So “Do accounts have weights?” There must be. The account's historical performance is good and has an impact on the new plan testing plan.b) Creativity If the creative is a video cover, the click-through rate has weight, and the historical performance of the creative is very important. The estimation of an idea will be largely based on historical data. Tencent mentioned the concept of freshness, which I have rarely seen on Toutiao. It may also encourage "new" creativity.c) Landing Page Next is the landing page. Tencent does not mention the landing page in this picture, but its explanation of landing page optimization also includes these factors: opening speed, relevance to creativity, etc. Tencent landing page optimization instructions: For landing pages and targeting, Toutiao’s plan diagnosis has more detailed instructions: The factors that affect the loading speed of the landing page are mainly the total size of the web page and the number and size of images. Therefore, the landing page cannot be too large, it should be "less and smaller" - put less content and make the webpage smaller. Toutiao landing page optimization instructions:d) Orientation Targeting can be broken down into coverage, precision and blue ocean. They refer to: the scope of the population covered, the accuracy of the targeting, and the intensity of competition within the targeted population (Tencent's "relevance" refers to the accuracy of the targeting) Toutiao Targeted Optimization Instructions: It is easy to understand, conversion volume = conversion rate * population range, conversion rate and population range correspond to accuracy and coverage respectively. However, finding these people does not mean that you can convert them. There are other advertisers competing with you. For example, if everyone uses a certain industry’s audience package, then the competition among this group of people must be fierce, and you may not be able to convert them even if you have targeted them. The main factors affecting eCPM estimates for each media:3. Frequency control Although eCPM is a very important traffic distribution indicator, we cannot only look at eCPM. Controlling the frequency of ads displayed to users is also a very important strategy. Frequency control is mainly divided into two aspects: filtering out content that users don’t like and removing duplicates . A possible strategy for filtering dislikes is: if a user chooses not to like an ad, not only the ad, but also the “industry, product, source, landing page, ad group” related to the ad will not be shown to the user. Another aspect is deduplication. Deduplication is a basic strategy for all media (including small DSP platforms). It is not only based on user experience, but also from the perspective of advertising effectiveness. If users always see repeated ads, the possibility of conversion will be relatively low. Therefore, the frequency of repeated display of ads to the same user should be controlled. Moreover, when you click "×" on various media, whether it is content or advertising, the media will provide an option for reasons for not liking it: duplicate content. This shows that people often choose not to like content because they see repetitive content. The reasons provided by each media after clicking "×" all include this one There are several dimensions for deduplication:
Advertisers
Installing Packages
Landing page link
Account
source
Group
plan
Creativity
You should first know that these are all dimensions that may be frequency controlled. The media will consider these items as a level, but whether these levels will be frequency controlled varies from media to media. So “Is there really competition between the same account and the same group?” real. All advertisements are originally in competition with each other. Because there is frequency control at the account and group levels, only one advertisement may be shown to a user, an account or a group, and the competition will be more obvious. At the same time, different media will have different focuses. For example, some media will strictly control repeated creativity, while others will not be so strict. Even some early media may not have frequency control. Guangdiantong emphasizes that "duplicate ads" are the reason for low exposure. Duplicate ads will only be displayed once, but Toutiao's excellent advertising system will automatically build a similar plan for you. In addition to deduplication, there will also be restrictions on advertising formats: large images and downloads. In the past, Toutiao clearly stated that large images and downloads were subject to frequency control, but it was no longer mentioned later, so some media may have this and some may not; and new users are generally protected and do not display advertisements, or display very few advertisements. Possible dimensions of frequency control To sum up, we can simplify advertising into three parts, each of which corresponds to some settings. You can think of it as a map and know that these are possible influencing factors. When the advertising effect is not good, you can try to change these things.3. A Possible Algorithm We know that there are so many possible influencing factors when it comes to media estimates, but how are they estimated specifically? This involves the "algorithm" - the "algorithm" is a method of estimating the probability of conversion based on certain characteristics of the advertisement and the user (this probability can ultimately be converted into ecpm). In order to help everyone understand the "algorithm" intuitively, here is a simple algorithm as an example. There is an algorithm like this. Each item has a different weight. After an advertisement comes in, it is scored from various dimensions, and then a total score is calculated. For example:
Video duration 0.003 minutes
Music popularity 0.03 points
Video model appearance 0.02 points
Cover color 0.001 points
Creative historical performance 0.2 points
Landing page loading time 0.04 points
Number of landing page images 0.01 points
Landing page length 0.01 points
…
Get a total eCPM score, rank the total score, and then allocate traffic according to the ranking. Those who work on algorithms have excellent technology and are very smart people. The smarter the algorithm, the closer the estimated results will be to the actual data.
IV. Conclusion User experience is a factor that must be considered in the algorithm. User interaction, viewing and complaint behaviors will have an impact on the weight of the advertisement.Interactive behaviors : like, thumbs up, comment, like, shareViewing behavior : clicks, reading percentage, completion, video playback time, effective playback rate, completion rateComplaint behavior : dislike, report We have sorted out some public explanations of quality indicators from several media outlets and found that these are possible influencing factors that can be used as optimization points in the long term. P.S. To what extent can the video be analyzed? Very fine degree. These analyses are technically feasible and may serve as influencing factors for estimation. In Toutiao, the music, shooting techniques, and the way the video is reversed may all be estimated influencing factors.
The above, welcome to leave a message for discussion.
