20-F
BAIDU, INC. filed this Form 20-F on 03/31/2017
Entire Document
 


Table of Contents

from both Baidu Search and Baidu Newsfeed is improved significantly. This not only enhances user experience, but also increases websites’ page visit traffic. Nearly one billion webpages have adopted this technology to optimize user experience in China.

P4P Technology

Our P4P platform serves billions of relevant, targeted sponsored links each day based on search terms users enter or content they view on the web page. Our key P4P technology includes:

P4P Auction System. We use a web-based auction system to enable customers to bid for positions and automatically deliver relevant, targeted promotional links on Baidu’s properties and Baidu Union members’ properties. The system starts by screening the relevance between the sponsored links and a particular query. Our intelligent ranking system takes into consideration the quality factor of a sponsored link for a search query in addition to the price bid on the keyword. The quality factor of a sponsored link for a search query is determined based on the relevance and certain other factors. The relevance is determined based on the analysis of past search and click-through results. Links to customers’ websites are ranked according to a comprehensive ranking index, calculated based on both the quality factor of a sponsored link for a search query and the price bid on that keyword. We employ a dynamic mechanism in determining the minimum bidding price for each keyword.

Phoenix Nest is designed to generate more relevant results. Phoenix Nest helps customers more easily find users’ favorite search terms to bid on, and provides customers with more tools for budget management and more data for the effective measurement of ROI. We have been continually improving our click-through rate, or CTR, estimation technology. For example, we have introduced deep neutral network, or DNN, technology into our CTR estimation. We have also developed a new generation Phoenix Nest deep learning network CTR estimation modeling system, which enables the estimation of clicks on different combinations of advertisement materials and significantly improved the timeliness of the model estimations. In 2016, we were the first to introduce FPGA (Field-Programmable Gate Array) into CTR model-based online services to optimize the click-through rates in online marketing, which has significantly improved users experience.

Generative Triggering Model, which includes query rewrite and auto-trigger technologies, has broken the status quo where traditional index-trigger is unduly restricted by literal meaning, and instead formed a sophisticated set of real-time sequence deep learning system to fully understand the search intent of users, enabling more accurate match between search intents and search results. We continue to enrich our business knowledge database in order to better understand the search intent of our users, and to present dynamic results in order to satisfy user demand, enhancing the realization of long-tail website flow.

P4P Billing System. We record every click and charge customers a fee by multiplying the number of clicks by the cost per click. Our system is designed to detect fraudulent clicks based on factors such as click patterns and timestamps. This system also computes the amount a Baidu Union member or a distributor should be paid. The billing information is integrated with our internal Oracle ERP financial system.

P4P Customer Service System. This system offers data and tools to analyze data for our customers to evaluate and optimize the performance of our online marketing services provided to them. Through this system, our customers can also manage information relating to online marketing services such as their budgets and time periods for the services.

ProTheme Contextual Promotion Technology. Our ProTheme technology employs techniques that consider factors such as theme finding, keyword analysis, word frequency and the overall link structure of the web to analyze the content of individual web pages and to match sponsored links in our P4P platform to the web pages almost instantaneously. With this targeting technology, we can automatically provide contextually relevant promotional links. For example, our technology can provide links offering tickets to fans of a specific sports team or a news story about that team.

 

56