Research And Implementation Of Personalized Advertising Click-Through Rate Prediction

Online advertising as the Internet’s most profitable business model develop rapidly. Display advertising in recent years, the share of the online advertising is growing, especially NGD (Non-guaranteed Display Advertising), personalized advertising can significantly improve economic benefits.Behavioral targeting personalized advertising is the better one, but far from thorough explored. CTR (Click-through rate) is the most commonly used evaluation about advertising effectiveness, advertising click-through rate prediction play a crucial role to enhance advertising revenue and user experience. However, traditional behavioral targeting build model between user behavior and click ad directly, facing the problem of cold start,sparse data, slow speed of the model update and ad CTR prediction real-time insufficient.In response to these issues, this paper use ad text information, build a preference model between users and advertisers topic to obtain the relevant features of users and advertisers, used to build efficient personalized CTR model, thereby reducing sparse data and ad clicks cold start problem. Make a more comprehensive comparison and improvement about distributed convex optimization methods deal with model training. Propose an effective advertising pre-filtering algorithm to reduce the set of candidate ads with level category information, thereby Enhance the efficiency of click-through rate prediction.