Implementation of personalized search in active web intermediaries

个性化信息服务是当前研究的热点。在Web搜索领域，个性化的研究重点是如何根据用户的内容偏好来调整搜索结果。在内容交付领域，个性化的重点是如何适应网页根据用户的性能导向的偏好。如果能够同时考虑内容和性能来实现个性化，那么对最终用户来说将是非常有益的。这是这项工作的目的，利用这两个领域的进展，实现搜索个性化。我们的目标是实现个性化的搜索服务，在网络中介的基础上，用户的要求，内容的相关性，以及对服务质量。目前，中介主要是指代理服务器或边缘服务器。从长远来看，中介的概念可以推广到任何位于客户端和服务器之间的信息服务代理，搜索目标不仅限于网页，也可以是Web服务。主要包括以下几个步骤：1）定义与Web搜索密切相关的服务质量因素：2）设计并实现个性化搜索的内容自适应框架; 3）设计并实现了一种能综合与性能相关的服务质量因素的重排序算法（例如，获取页面的延迟、用户的连接速度）到传统的信息检索模型中。该方案的主要特点是：1）搜索个性化作为一种内容自适应服务在Web中介中实现，可以充分利用其日志中保存的丰富的用户数据及其关于页面交付性能和客户端性能的信息; 2）所提出的重新排序算法可以扩充现有的排序模型，这在普适计算环境中特别有用; 3）我们将培养高素质的人才，以深入了解网络搜索，自适应内容交付和个性化。他们将有具体的经验，系统的建设和系统的实验方法，这将有利于他们未来的职业生涯。

Personalized information service is an active area in many research communities. In the web search area, the personalization focuses on how to adapt the search result according to user's content-oriented preferences. In the content delivery area, the personalization focuses on how to adapt the web page according to users' performance-oriented preferences. If the personalization could be implemented considering both content and performance, it would be really beneficial to the end user. It is the purpose of this work to leverage the advances in both areas to implement the search personalization. Our objective is to implement the personalized search service in web intermediaries based on users' requirements on content relevancy as well as on quality of service. Currently, the intermediary mainly refers to the proxy server or edge server. In the long term, the concept of the intermediary could be generalized to any information service agent sitting between clients and servers, and the search target is not limited to the web page, it could also be a web service. There are several major steps involved: 1) to define quality of service factors which are closely related to the web search; 2) to design and implement a content adaptation framework for the personalized search; 3) to design and implement a re-ranking algorithm which could integrate the performance-related quality of service factors (e.g. latency of fetching a page, user's connection speed) into the traditional information retrieval model. The key features of this proposal would be: 1) search personalization is implemented as a content adaptation service in a web intermediary, which could fully utilize the rich user data saved in its logs and its information concerning the page delivery performance and the client capabilities; 2) the proposed re-ranking algorithm could augment the existing ranking models, which is especially useful in a ubiquitous computing environment; 3) we will train the highly qualified personnel to gain a deep understanding on web search, adaptive content delivery, and personalization. They will have concrete experiences on system building and systematic experimental methods, which would benefit their future career.