With the rapid growth of programmatic purchase scale, DSP has evolved from a conceptualization stage to a market standard product service with standardized assessment targets. However, the current DSP market is mixed, and each DSP can standardize the docking traffic trading platform (Exchange). In the case of advertising, the algorithms and data become the core competitiveness of DSP. They essentially constitute the same. Under the resource constraints, DSP achieves the logical support point of better KPI.
In fact, algorithms and data are the underlying structures that DSP can't see. Their value is mainly reflected in three aspects: one is to accurately identify and reach users for advertisers, and the other is to accurately estimate the value of users. The third is to choose the marketing information that impresses TA according to the stage of cognition of demand. By solving these three problems, the advertising budget can be reasonably distributed to optimize the performance of the advertising.
Algorithm and data scenario application
In general, the algorithms and data in the DSP are mainly used in the following scenarios:
First, find the target user (TA) in the vast Internet resources market. Take the maternal and child industry as an example. There are two difficulties in the marketing process. One is to use credible logic to confirm that the “mother and baby†users you find are indeed “maternal and infant†people, and the second is that “mother and baby†people are in different channels. Unified recognition capabilities between different media. Ideally, each tag population cluster is a continuous process in which the DSP algorithm builds a model based on the user's Internet behavior and then verifies the iterative optimization model through the trusted Panel library. Therefore, DSP based on complete user behavior and trusted panel library, establish a credible demand judgment model, is the key point for building a "trusted" label user.
The most practical Panel library is the customer's first-party data, including the offline member information or the online conversion information collected (in theory, the customer's purchase user group on each e-commerce platform should be their own assets). For example, the YOYI label system is divided into two groups of interest levels: interest tags and purchase intentions. Correspondingly, we will refer to the user's behavior characteristics in the official website to determine the different levels of user needs as one of the modeling test standards. .
At the same time, YOYI accesses the traffic of almost all trading platforms and mainstream media in the domestic market, covering a large area of ​​user Internet behavior; on the other hand, we integrate the user's natural behavior and advertising behavior, from the user's point of view, as long as it happens The actual browsing and click behavior is an embodiment of the user's interest in the information, and does not strictly distinguish whether the information is natural information or advertising information. The above two points allow YOYI to fully restore the user's Internet behavior track.
The complete user behavior trajectory guarantees that the model can be used with complete features and accurate feature values. The rich first-party Panel data provides a real and effective training and verification set for model training. These two points together ensure the validity of the user judgment model. In actual operation, feature classification and feature extraction are performed according to the behavior of the user's behavior, media, time, frequency, browsing duration, search, click advertisement, browsing advertisement, etc., and the user needs phased behavior model is trained to perform label modeling. . In this way we can make more accurate judgments on the user's needs and requirements hierarchy.
At the same time, multi-channel unified identification of target users is another problem that technology needs to solve. At present, there are opportunities to complete large-scale Internet PC and mobile terminal unified ID identification. Only giant companies with large mobile and PC users can let DSP It is easier to cross this issue. However, there is currently no large-scale unified multi-screen ID standardization service in China, so DSP needs to build its own multi-screen unified ID system. YOYI's approach is to build a multi-screen user identification algorithm, and then use the market-standard third-party multi-screen ID as the training set for optimization verification. For example, the same person on the PC and mobile, wifi access, geographic location, behavioral trajectory, and catalyst habits will have some similarities, which are the basis for building a unified ID algorithm. The combination of machine learning algorithms and manual rules for large-scale data can deconstruct data and recognize data from many dimensions, thus solving the problem of inconspicuous rules and inconsistent behavior. The user extracts and models the intersection dimensions of PC, mobile, and PC and mobile, and trains the unified ID model. The standard third-party multi-screen ID is used as the evaluation standard, and the accuracy rate is available.
Secondly, to analyze the value of the target users, not every absolutely accurate "mother and baby" crowd is "just right" to buy milk powder when encountering in each scene. User-level conversion rate estimates and click-through rate estimates are key issues for performance advertising. Estimating the clickthrough rate and conversion rate of a particular ad for a particular user at a particular location is a typical machine learning problem for large-scale data. We constructed a user feature system, an advertisement feedback feature system, a traffic feature system, and a cross-feature system for each dimension. The classic LR was used as the predictive model, and the GBDT was used as the high-dimensional feature extraction model to predict the click-through rate and conversion rate. Estimated, offline evaluation and online results have a good performance. After estimating the clickthrough rate and conversion rate, we can calculate the KPI-oriented CPM based on marketing goals. Take the automobile Leads as an example. Ecpm=CPA* conversion rate, according to different users and different traffic, CPM bidding, in order to achieve the best budget within a limited budget.
In addition, there is a strategy that is often overlooked but crucial to prevent cheating. How to identify the data of the input model and eliminate false traffic, clicks, and even conversions is a core topic that needs to be developed by a separate article. Make a record here to remind it of its importance. The conversion rate after data purification is very important. On the one hand, it is an important factor in the advertisement screening order in the delivery, which determines which of the most relevant advertisements the user sees in the current scene, and on the other hand, in the authorization of the advertiser. The algorithm can automatically bid for the target user to ensure that the appropriate range jumps out of the CPA's bid limit to track the core user base, avoiding being "robbed" by competitors to show opportunities.
Finally, after finding the person and knowing the focus of the TA and the corresponding demand stage, it is necessary to consider using appropriate communication methods to impress the TA. At present, limited by the richness of current advertisers' creativity, production ability and media review cycle, our algorithm application in smart creativity is still in the early stage of development, and there is still a lot of room for the explosion of creative combination utility. However, it has been seen that some third-party creative companies in the market have been working hard in this direction, and a creative revolution in Internet advertising is just around the corner.
Industry-specific applications for algorithms and data
Take the algorithm and data application of the automotive industry as an example to illustrate that the general algorithm and data framework need to be optimized by the industry to achieve the best effect. In the actual advertising of cars, we found an interesting phenomenon: the time dimension of the user's car-related behavior is a very useful feature for the last sale of Leads. We analyzed that some car hobby people, although long-term attention to car forums or browsing car knowledge, but they do not have car purchase demand in the short term, so there is not much effect on the sale of Leads. However, for those who have a need to buy a car in the short term, their timeliness will be very obvious. Therefore, based on the general model, we have differentiated the modeling factors of key industries, so that certain explicit features in some industries play a unique role in user demand identification, and become an effective weapon to meet different marketing targets of advertisers.
In summary, the algorithm and data core application in DSP are in three aspects, one is the identification of different stages of the user's various interests, one is the judgment of the value of the user in different media scenarios, and the other is the choice of displaying effective information to the user. Around these issues, the user interest and effect value algorithm system constitutes the core algorithm system of YOYI, which is continuously improved in practice to help advertisers achieve marketing goals.
China leading manufacturers and suppliers of FOR Mitsubishi Oil Filter,Oil Filter For Mitsubishi, and we are specialize in FOR Mitsubishi Car Oil Filter,FOR Mitsubishi Auto Oil Filter, etc.
Mitsubishi Oil Filter Replacement,Oil Filter For Mitsubishi,Mitsubishi Car Oil Filter,Mitsubishi Auto Oil Filter
Zhoushan Shenying Filter Manufacture Co., Ltd. , https://www.renkenfilter.com