|
Jian Zhang Language Technologies Institute Carnegie Mellon University Sepetember 2003 |
Advanced IR Lab,
11-743 Instructors: Jamie Callan, Yiming Yang |
| This project is to fulfill the course requirement of Advanced IR Lab. As emails become more and more important in people's daily life, email prioritization emerges to be a problem. That is, how to rank the emails of a user based on his judgment about importance. In order to evaluate the proposed method, we need to collect some emails from some users, and label them based on some criteria. We use two evaluation measures: 1) recommendation evaluation measure used in Collaborative Filtering; 2) Kendall's \tau. Preliminary results should be presented in the end of this report. |
|
Email prioritization is the task of ranking emails of a user for emails he received by not read yet. The ranking of emails should be based on the their importance, which is primarily predicted based on that user's previous ranking information, as well as others. Though can be simply treated as a chunk of text, emails have more information than usual documents in the following ways: ...... Besides, the task of ranking emails is different from existing classification algorithms. That is, the input of the model should be several ranking lists, the system is going to learn a model which can predict a ranking list for a given set of emails. If we discretize the importance ranking of emails and put them in several categories (say, "very important", "important", "mediocore", "ad", "junk"), then it is still inappropriate to use traditional classifiers for the following reasons: the relations between two classes are different. To be more specific, there is a linear relationship between different classes, which should be utilized in the classification scheme. In this project, I propose the following way to handle this problem: |
|
This is the algorithm: |
|
Algorithm ...... |
|
Results ...... |
| [1] J. S. Breese, D. Heckerman and C. Kadie, Empirical
Analysis of Predictive Algorithms for Collaborative Filtering,
Uncertainty in Artificial Intelligence, 1998. [2] D. M. Pennock, E. Horvitz, S. Lawrence and C. L. Giles. Collaborative Filtering by Personality Diagnosis: A Hybrid Memory and Model-Based approach. Uncertainty in Artificial Intelligence, 2000. [3] C. M. Kadie, C. Meek and D. Heckerman. CFW: A Collaborative Filtering System Using Posteriors Over Weights of Evidence. Uncertainty in Artificial Intelligence, 2002. [4] D. Heckerman, D. M. Chickering, C. Meek, R. Rounthwaite, C. Kadie. Dependency Networks for Inference, Collaborative Filtering, and Data Visualization. Journal of Machine Learning Research 1, 2000. |