PURRS, a profile-based email management system that provides personalized email prioritization and adapts to users' changing behaviors.
The proposed system learns and leverages behavioral and social-networking information to build profiles of email senders. Users will be able to identify important email quickly based on senders' ranking. Therefore, they can handle email more efficiently as the system mitigates their frustration and negative impacts from email cluttering.
Although standard email systems allow users to define rules for email prioritization, they require the users to update the rules manually and regularly as user behaviors and contacts change over time. Previous work on automatic email prioritization mostly focused on text categorization using email bodies and subjects. Spammers and/or non-important contacts can craft email to bypass these filters. Our work takes a different approach by mimicking users' perceptions-how they examine senders and subjects to triage email. We hypothesize that users tend to handle email from senders within their social networks first because they already learned of the senders' importance. For this reason, our system transparently learns how users interact with contacts, and then it clusters contacts into Social Network Rings (SNR). Contacts whom the users have paid more attention to are in higher priority groups and placed closer to the center of the SNR. The system prioritizes incoming email by senders' positions in the SNR. Moreover, the system uses information from the users' social networks to prioritize email from unknown senders. The specific contributions of this work are as follows:
1) We use an extensive set of user actions, including reading (and reading delay), deleting (and deleting delay), replying, forwarding, composing, flagging, and archiving email.
2) We introduce the Social Network Rings (SNR) with five rings/layers to represent not only users' social network maps but also relationship strength. The five rings represent five priority groups, which include five relationship strength levels, adapted from a human relationship development model.
3) Since user behaviors and relationships change over time, we propose to apply the time-sliding database and the Slow Adjustment Tuner. Our time-sliding database assigns decaying weights to older user actions, giving more attention to recent ones. The Slow Adjustment Tuner, adapted from the TCP Slow Start with Fast Recovery scheme, regularly updates the system parameters in accordance with user feedback on email priority classifications.
4) Because a user does not have information about unknown contacts, we propose to apply a social network concept to help identify new contacts that could be important to the user. The user's profile is shared among selective contacts, and potentially important new contacts are suggested from the user's social networks. For privacy, the user has control over who receives which pieces of his/her profile information.
PURRS is implemented as a Microsoft Outlook 2007 add-in and we'll conduct a user study to evaluate the system. The evaluation goals are:
1) To verify the project concept, by confirming the actions that indicate the relationship strength with contacts and by studying the user's information sharing behaviors,
2) To measure system performance, including the priority prediction correctness, the system's learning ability, and the correctness of the unknown contact's rank estimation,
3) To measure the typical system resource consumption, and
4) To assess user feedback of the system performance. We will conduct a 12-week user study in which participants use the system, while we automatically collect data weekly. At the end of the study, we will conduct an email-identification study, conduct a user feedback survey, and analyze the collected data. Upon the completion of the work, the system will allow the user to handle a large volume of email more efficiently and effortlessly, reducing frustration and negative impacts from email overload.
In this project, I will be working with Kornchawal Chaipah, the head of project. And the part that I am responsible for is the self-sharing module which will enable our system to share user's data across all email clients.