ITR: Bayesian Models Linking Web Site Structure and Usage

摘要PI：Alan Karr提案编号：ITR-0218759尽管万维网无处不在，但人们对其了解甚少。因此，许多网站难以导航，难以使用，结构混乱，以至于用户可能无法找到内容并放弃网站。基本需要是将用户行为与网址结构联系起来;比较不同时间或不同类别用户对网址的使用情况;对会话进行分段;量化各网页之间的相互关系;预测用户行为，包括预测促销活动的经济影响等。最终的影响是更高效的网站，更有效地为用户服务。这项研究将创建一组四个日益复杂，但可扩展的贝叶斯模型，这些模型将网站的使用（特别是用户页面转换）与其结构联系起来。它将应用、验证和改进这些模式，并使用来自四个性质不同的网站的真实的数据，这四个网站是一个电子商务网站、一个由大型金融机构经营的网站、一个内容网站和一个信息网站。这些模型是可伸缩的，因为来自给定页面的目标是反映站点树结构的页面类，而不是单个页面。例如页面的父级、子级和同级。所有四个模型都假设每个页面的转换都是Dirichlet先验分布。前三个采用非常聚合的转换类，并根据转换分布和先验是否对所有页面都相同而有所不同。第四个模型分解了“子”和“兄弟”目的地。后验分布的计算难度各不相同：有些是以封闭形式提供的，而另一些则需要密集的马尔可夫链蒙特卡罗计算。此外，严格的模型评估将提供洞察什么样的聚合水平是适当的网络数据的分析。

AbstractPI: Alan KarrProposal Number: ITR-0218759Despite its ubiquity, the World Wide Web is poorly understood. As a consequence, many sites are difficult to navigate, hard to use and have confusing structure, to the extent that users may be unable to find content and abandon the site. Essential needs are to relate user behavior to Web site structure; to compare site usage at different times, or for different classes of users; segmentation of sessions; quantification of inter-relationships among pages; and prediction of user behavior, including forecasts, for example, of the economic impact of promotional campaigns. The ultimate impact is more efficient Web sites that serve users more effectively. This research will create a set of four increasingly complex, but scalable, Bayesian models that relate the usage (specifically, user page transitions) of a Web site to its structure. It will apply, validate and refine the models and use real data from four qualitatively different Web sites, an E-commerce site, a site operated by a large financial institution, a content site and an information site. The models are scalable because the destinations from a given page are classes of pages that mirror the tree structure of the site, rather than individual pages. Examples are the parent, children and siblings of a page. All four models assume Dirichlet prior distributions for transitions from each page. The first three employ very aggregated classes of transitions, and differ according to whether the transition distributions and the priors are the same for all pages. The fourth model disaggregates the "child" and "sibling" destinations. Calculation of posterior distributions varies in difficulty: some are available in closed form, while others require intensive Markov chain Monte Carlo computation. In addition rigorous model assessment will provide insight into what level of aggregation is appropriate to which analyses of Web data.