Shulou(Shulou.com)06/01 Report--

Basic Flow and Algorithm of DSP Release

DSP usually follows the following process when launching:

Step 1: Identify the user:

DSP systems typically place a detection point at the time an ad is displayed, so that when an Internet user visits an advertiser's website for the first time, a cookie is planted so that DSP can track the behavior of the user on the advertiser's website. DSP also collaborates with media outlets and third-party DMP's to perform Cookie Mapping to identify the user before bidding.

Step 2: Audience Selection:

For each advertiser campaign, develop a model that takes as positive the users who convert on the advertiser website (conversion behavior can be registration, click, purchase, download, etc.) and as negative the users who do not convert. After the model is established, the conversion probability p(c) is estimated for all users.| u), that is, how likely is the user u to convert on the advertiser's website (c represents conversion), remove most users with very low conversion probability, and classify the target users into different delivery plans according to the conversion probability. This way we find a lot of target users for each campaign, and these users are placed in different delivery schedules according to their quality.

Step 3: Conduct real-time bidding:

When adx(Ad Exchange) sends a request, DSP gets the following information: information about the current ad slot, cookies of the current user, and terminal information. DSP needs to decide whether to buy this presentation, which compaign ad to place, how much bidding, and return bidding information to Ad exchange within a specified time (usually within 100ms) according to its own bidding algorithm based on the analysis of the current user and the current ad slot. If the DSP does not respond after a period of time, the exchange defaults to the DSP giving up the bid.

Step 4: Display advertising:

If a presentation is won, the DSP returns the creative and the user sees it in that ad slot.

Step 5: Track the conversion:

DSP buries a dot on the advertiser's website to see if the user has converted after the presentation. According to these data statistics conversion rate, average cost per conversion and other indicators, summarized into reports to advertisers

DSP Audience Selection Algorithm:

1. Low-level Model: The purpose of this model is to do primary selection. All users who convert on the advertiser website corresponding to the campaign are taken as positive examples, and other users are taken as negative examples. The model has only one class of features, which are URLs that users have visited historically. Linear approach to reduce dimensional impact

2. High-level Model: The role of this model is fine selection. The sample of the model is the same as the Low-level Model, and the characteristics are not limited to the URLs visited by the user, but can be various mined attribute tags of the user, including interpretable classification tags, uninterpretable clusters, and association features with advertiser websites. which is how you label users.

DSP bidding algorithm

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.