基于增量学习的精准广告投放系统研究

发布时间：2018-03-09 09:10

本文选题：增量学习　切入点：精准广告　出处：《山西财经大学》2010年硕士论文　论文类型：学位论文

【摘要】： 网络技术的飞速发展,广告成为网络盈利的一个主要手段。网络广告为越来越多的企业和机构所了解,并且大部分企业和机构都进行了网络广告的投放。但是,网络广告形式多样,具有动态性。网络广告投放的随意性和泛滥性,使网络用户应接不暇,产生了厌烦的情绪,使网络广告的投放达不到预期的效果。针对这种情况,出现了精准广告投放这一概念。精准广告投放即向网络用户投放其感兴趣的,并真正能够及时供用户所需的广告信息。而通过对用户的行为分析,对用户进行智能地分类,成为实现精准广告投放的有效方法,那么对网络上的海量用户进行智能分类就成为进行精准广告投放的重要研究内容。由于海量的网络用户信息,采用数据挖掘方法,可以有效地对用户进行智能分类。然而由于所获得信息的实时性与在线性,即在互联网下的用户的行为信息是不断变化的,致使这些海量的信息并不能一次性全部获得。而对于分批获得数据,一般的分类算法,需要不断地重新的更新分类模型,而耗费了大量的时间。增量学习则是对训练数据集的学习过程逐步展开,后续的学习结果是建立在先前学习结果的基础上的。同时,鉴于贝叶斯分类方法能够充分利用先验知识学习这一特点,在一定程度上解决了先验信息传递的问题。因此,本文提出一种基于贝叶斯的增量学习算法用于对精准广告投放系统中用户的分类,它是一个利用样本知识来修正当前知识的连续的、动态的过程。通过对在线用户的行为特征不断进行分析,根据获得的数据信息利用贝叶斯增量学习对用户进行分类,不断地更新分类模型,以达到更好的分类效果,从而更有效地实现精准广告的投放。本文选择了Book-Crossing(BX)数据集作为实验研究对象,它包括278858个用户(匿名但有人口统计信息),提供了对于271379本图书的1149780评分信息。通过利用SQL Server 2005处理成实验所需的数据格式。研究结果表明,增量学习对于在线的实时学习能够解决其他分类器所带来的时间与精力的耗费,并能获得较好的分类效果,从而精确地、及时地对网络用户的分类,达到精准广告投放的目的。本文对精准广告投放系统进行了设计,从系统分析到系统功能的实现都进行了论述,并提出主要是将用户推荐模块作为一个通用接口,不仅在本系统中可以应用,在其他的网站中,只要将代码嵌入并做相应的调整,就同样可以实现对当下网站的用户的分类。本文结论部分对整个文章进行了总结,并提出下一步工作。
[Abstract]:With the rapid development of network technology, advertising has become one of the main means of making profits on the Internet. More and more enterprises and institutions have learned about online advertising, and most of them have carried out online advertising. However, Network advertisement has various forms and is dynamic. The randomness and flood of network advertisement put in, make the network user be overwhelmed, produce weariness mood, make the network advertisement put in can't reach the expected effect. In view of this kind of situation, The concept of precision advertising has emerged. Precision advertising, that is, delivering information that is of interest to users on the Internet and can really provide users with the advertising information they need in a timely manner, can be intelligently classified by analyzing the behavior of users. It has become an effective method to achieve precision advertising, so the intelligent classification of mass users on the network becomes an important research content of precision advertising. Because of the huge amount of network user information, using data mining method, users can be classified intelligently. However, because of the real-time and linearity of the information obtained, that is, the behavior information of users under the Internet is constantly changing. As a result, these huge amounts of information can not be obtained all at once. However, for batch data, general classification algorithms need to constantly update the classification model. Incremental learning is the gradual expansion of the learning process of the training data set, and the subsequent learning results are based on the previous learning results. At the same time, In view of the fact that Bayesian classification method can make full use of the characteristics of prior knowledge learning, the problem of prior information transmission is solved to a certain extent. Therefore, an incremental learning algorithm based on Bayesian is proposed to classify the users in the precision advertisement delivery system, which uses sample knowledge to modify the continuity of current knowledge. Dynamic process. Through the continuous analysis of the behavior characteristics of online users, according to the obtained data information, Bayesian incremental learning to classify users, constantly update the classification model, in order to achieve a better classification effect, In this paper, the Book-Crossing BX data set is selected as the experimental research object. It includes 278,858 users (anonymous but demographically available, provides 1149780 rating information for 271379 books). It is processed into experimental data formats using SQL Server 2005. Incremental learning for online real-time learning can solve the cost of time and energy brought by other classifiers, and can obtain better classification effect, thus accurately and timely classification of network users. In this paper, the precision advertising delivery system is designed, from the system analysis to the realization of the system functions are discussed, and it is proposed that the user recommendation module as a general interface, Not only can it be applied in this system, but also in other websites, as long as the code is embedded and adjusted accordingly, it can also realize the classification of the users of the current website. And put forward the next work.
【学位授予单位】：山西财经大学
【学位级别】：硕士
【学位授予年份】：2010
【分类号】：TP311.13

【参考文献】