Forum Posts

Robin Islam
Apr 10, 2022
In Welcome to the Forum
Generally, there are four situations that trigger the display of cached ad content: Preloading ads: For example, for a homepage list ad, after the screen-opening ad is displayed, the preloading cache is triggered, and the display is triggered when entering the homepage. Hit cached ad: If an ad source bids for the opportunity to display this ad, and the ad already exists in the cache pool, the ad will be displayed. Bottom-up cached ads: If there is no ad source for ad response, the cached bottom-up ads will be displayed first. The request interval is not reached: If the request interval is not reached, the cached ad of the last exposure will be displayed by default. It is worth noting that a request generally counts for a valid impression and click, and a cache-based content display without a request is not counted as a valid impression. Therefore, for the media side, the statistics of the original display and the effective display of the two events can be recorded separately. In addition, the exposure corresponding to one request is charged only once at most. For the needs of evaluating the value of advertising space and CTR estimation, it is possible to provide statistical repeated exposure/click data for reference. For example, repeated exposure of information feed advertisements without re-request, frequent clicks on the same exposure by the same user, etc. 04 Optimization of the mechanism Finally, let's talk about the optimization of ad requests and caching mechanisms. In a sense, the mechanism is a means to achieve an end, and the advertising mechanism is no exception. Therefore, the optimization of the request and caching mechanism behind the advertisement is a dynamic process, which is mainly related to the advertising goals of the current media. For example, the current media volume is small and the traffic quality lacks the core advantage. In this case, the goal can be set to increase the ad fill rate. Then, in the request mechanism, it would be more appropriate to allocate part of the traffic to ad sources with stable fill rates, rather than the "ecpm-only" theory. In the optimization of the advertising mechanism in practice, a sound data burying mechanism is indispensable. Through layer-by-layer filtering and analysis, we can locate the problem points and find corresponding optimization ideas. I hope the following two data funnels can provide some reference: 1. Request filter funnel According to the request funnel, we can combine the overall media traffic, single advertising space traffic and other dimensions to conduct layer-by-layer analysis to locate the key to traffic loss. For example, because a certain type of crowd accounts for too much, the request has been filtered first, and it is recommended to relax certain crowd restrictions. For example, we found that a certain advertising source channel has a serious return Phone Number List timeout phenomenon, and we can suspend the opening of a certain channel first, and then adjust it after the channel is optimized. 2. Display the filter funnel According to the data of the display funnel, we can also optimize the caching mechanism to improve the success rate of ad display. For example, if the proportion of expired traffic from the cache is too high, we can adjust the expiration time range of the cache; for example, if the video resource is often not downloaded, then we can start the preloading caching mechanism for some ad slots involving the resource, etc. Wait. As for what is the standard endpoint for optimization? There is no one-size-fits-all answer to this, it is based on a never-ending product cycle of ideal states. We need to constantly try-verify-optimize to get as close as possible to the ideal state (the index data set in the current period). We've seen "countless" ads, and countless others that never had the chance to show up. For the user experience (for the media to have the opportunity to earn more ¥ ), the request and caching mechanism behind the advertisement display will also be a long way to go, and it will be done and optimized~
0
0
3
 

Robin Islam

More actions