Tingshao Zhu, Russ Greiner, Gerald Häubl; "Learning a Model of a Web User's Interests" wins the 2003 James Chen Best Student Paper Award at UM'2003.

• Predicting Web Information Content
  our method learns to predict which words will be in IC-pages ...
• Learning a Model of a Web User's Interests
  a novel method for learning user browsing model from a set of annotated web logs ...
• An Effective Complete-Web Recommender System
  Our goal is a system that can recommend IC-pages from anywhere on the web ...
• Predicting Where a Web User Wants to Go
  introducing our on-going research that predict which web pages she wants to visit ...

For a more extensive list of publications, please visit Tingshao Zhu's publications, or search in AICML PapersDB!

In this paper, we propose a novel method to infer the web user's Information Content(IC), i.e., the information that she must examine to complete her task. In particular, our method tries to predict which words will be in the web page that the user must examine to finish her task --- i.e., IC-page. We use page-content information extracted from the user's clickstream to train a classifier to predict what kind of words will be in the IC-page, i.e., the IC. The classifier is trained on generalized information to indicate how the user treats the information that she has visited, that is, browsing behavior. The classifier can be used to predict the IC of the web user with any given obeservable page sequence, thus it can be used in totally new environment, and to build an effective personalized system. The results indicate that our method can predict web users' IC fairly well.

There are many recommender systems that are designed to help users find relevant information on the web. To produce recommendations that are relevant to an individual user, many of these systems first attempt to learn a model of the user's browsing behavior. This paper presents a novel method for learning such a model from a set of annotated web logs --- i.e. web logs that are augmented with the user's assessment of whether each webpage is an information content (IC) page (i.e., contains the information required to complete her task). Our systems use this to learn what properties of a webpage, within a sequence, identify such IC-pages, and similarly what "browsing properties" characterize the words on such pages ("IC-words"). As these methods deal with properties of webpages (or of words), rather than specific URLs (words), they can be used anywhere throughout the web; i.e., they are not specific to a particular website, or a particular task. This paper also describes the enhanced browser, AIE, that we designed and implemented for collecting these annotated web logs, and an empirical study we conducted to investigate the effectiveness of our approach. This empirical evidence shows that our approach, and our algorithms, work effectively.

There are a number of recommendation systems that can suggest the webpages, within a single website, that other (purportedly similar) users have visited. By contrast, our goal is a system that can recommend "information content" (IC) pages --- i.e. pages that contain information relevant to the user --- from anywhere in the web. This paper describes how we addressed this challenge, We first collected a number of annotated user sessions, whose pages each include a bit indicating whether it was IC. Our system, ICPageFinder, then used this collection to learn the characteristics of words that appear in such IC-pages, in terms of the word's "browsing features" (e.g. did the user follow links whose anchor included this word, etc.). This paper describes the ICPageFinder system, as well as a tool (AIE) we developed to help users annotate their sessions, and a study we performed to collect these annotated sessions. We also present empirical data that validate the effectiveness of this approach.

In this paper, we introduce our on-going research that uses the content of the user's observed clickstream to predict which web pages she wants to visit. Our method first identifies which words will be in these "information content" pages, and then uses these words to construct search queries to retrieve the relevant webpages. We present empirical evidence that this approach can work effectively.