Design And Implementation Of Data - Intensive Advertising Click Rate Forecasting System Based On Probabilistic Graph Model

With the rapid development and widespread usage of Web2.0, Internet industry and Electronic Commerce, advertising becomes the most profitable way of Internet companies. Meanwhile, as an important standard of evaluating success advertising, Click-through Rate being predicted accurately is in favor of increasing the revenue of Internet enterprises and improving users’experience.There exist similarities of some activities such as browsing and searching between online users, which have uncertainty. Then, Bayesian Network is an important method of reasoning for uncertain data as an important Probabilistic Graphical Model. Thus, we predict Click-through Rate of target advertising by using uncertainty expression and inference method of Bayesian Network modeling and computing the relationship of dependency between users.In this thesis, including data preprocessing, Bayesian Network construction and advertising Click-through Rate prediction based on Bayesian Network inference. The details are as follows.· We process massive data of users’searching advertising log based on MapReduce framework. We store massive data in distributed file system HDFS. And support subsequent Bayesian Network Directed Acyclic Graph structure building by using MapReduce framework distributed program to read massive users searching data and extracting searching key words as users features which willed be stored in distributed database HBase.· We construct Bayesian Network based on MapReduce framework. It counts conditional probability tables of each nodes in Bayesian Network which willed be stored in HBase by reading and processing data in HBase parallelly, building DAG structure of Bayesian Network efficiently and processing data in HBase by using distributed infrastructure MapReduce framework parallelly.· We predict Click-through Rate via Bayesian Network inference based on MapReduce framework. System could find out the collection of similar user quickly and efficiently by using distributed framework. Then, system predicts Click-through Rate via using similarities between users.Finally, we develop the advertising Click-through Rate predicting system which based on Hadoop platform can process the massive data based on above algorithm. The search engine could get the Click-through Rate of the target user by simple invoke the interface which provided by Advertising Click-through Rate Predicting System.