Bayesian Network Forecasting Method Based On CTR And Implementation

With the development of the Internet industry, Internet advertising has gradually become the important force to promote the sound development of the Internet industry, the prediction of CTR (click-through rate) is to provide a basis for the accuracy of AdServing, but also improve users’satisfaction of Ad (advertisement) in order to prompting the user to click the Ad that most interested in. This would not only improve the income of advertisers and advertising media, but also promote the development of third-party billing pattern and the Internet industry. Taking into account the requirements of accuracy and personalized, accurate recommendation of the Ad for users is needed. For users without historical record, the recommendation of advertising and the prediction of CTR is still important. Thus it is a crucial issue of the research of Computation Advertising.In this thesis, we discovered user behavior similarity from historical data as the starting point, based on the similar relationship found between users, for the users that are lack of data predicted CTR. Uncertainty due to the user’s behavior, a framework is required to express the uncertainty in user behavior. The Bayesian network (BN, Bayesian Network) as found similar user behavior model uncertainty in the basic framework of knowledge representation and inferring.We build a Bayesian network through the analysis of historical data to reflect of directly similar relationship between the user and the uncertainty of the similarity relation, and then based on Bayesian network inference mechanism mining user similar indirect relationship between user history Click on a record to predict its click-through rate on the ad in order to predict the click-through rate of ad for the user without historical record.The main work and contributions of this thesis can be summarized as follows:■We constructed the model that reflect the users’similarity in advertising search behavior, called SBN (Similarity Bayesian Network). As is known that the construction of DAG (Directed Acyclic Graph) is critical and difficult in construction a BN. We gave out a algorithm by performing statistical calculations through users’historical record in order to construct the network structure of BN, and then found the direct similar users.■We used Bayesian probabilistic inference mechanism, given the the SBN inferring based on Gibbs sampling algorithm to find the SBN indirectly similar relationship with the user efficiently. Then use the similar relationship between users, given the algorithm of predicted CTR.■We tested the effectiveness of the approach based on the KDD Cup2012Track2training data set and developed "Bayesian network-based precision advertising launch simulation software" based on this method.