Design And Implementation Of Click Through Rating System Based On Logistic Regresion Model

Now the internet advertising has become the major source of revenue for companies like Google, Yahoo and Facebook. Not only these companies mainly rely on advertising revenue, but have been coming ahead of the Internet advertising industry. So how to study advertising realization and how to increase advertising revenue are the main domestic of the foreign internet research and the major hotspots of competition for talent. Among them, one of the internet advertising: mobile contextual advertising bases on analysis of user information and page information in a different context, which is essentially different from the traditional advertising in advertising fees and advertising display.The main job of contextual advertising is how to select the appropriate advertising from the larger network traffic and millions of ad libraries in an accurate delivery for users. But how to sort advertising in such a complex and expense of high-performance environment, as well as how to design and implement such a candidate advertising to be calculated for the CTR system is the main work of this paper.In fact, what presented above is based on the environment which the user is triggered by, and use the user’s own information and pages information on relevant advertising to compute the probability which advertising the user may click, thus to sort advertising according to the clicked-probability, and select the largest clicked-probability advertising present to the user, which is called the CTR(click-Through rate, CTR) estimates. Show advertising to users with the highest CTR which means to a win-win-win. For advertisers, precise advertising delivery enables to more click and display, which leads to more potential customers; For companies, more click means more money; For users, users are happy to click which improves their experience. Comparing to the traditional advertising sorting and delivery, the contextual advertising with more comprehensive information, more complex environments, and more accurate estimates and delivery.This paper mainly studies how to design and implement of the CTR estimates system, the feature selection has been carried out in detail, the algorithm module has been optimized and experimental compared, concrete steps are divided into as follows: Firstly, proposed the overall design of CTR estimates, including functional and non-functional requirements analysis; then, proposed technical solutions and evaluation CTR estimates system implementation; secondly, the CTR estimate system is divided into on-line and off-line two modules which carry out to achieve separately; Finally, the feature selection module joins the time decay factor and join APP’s classified information first time, and propose which based on logistic regression algorithms, naive Bayes algorithms, support vector machine algorithm, fusion(gradient boosting decision tree plus logistic regression) algorithm CTR estimates, and through experimental comparison optimization and off-line testing, so that CTR estimate system has been improved.