Information overload is a far greater problem today than it ever was in the past. The goal of ICARUS is to help alleviate this problem by providing users with personalized information relevant to the context of their current activity.

icarus logo The relevance is determined based on a multi-tiered, fine-grained user model built from information contained in multiple sources such as email data, organization structure, and enterprise online social network.

Currently there are two applications of ICARUS in the email domain: personalized email prioritization and automated contextual information retrieval (IR). These two applications use the same user model which encodes granular information about the user's pair-wise interaction with different people on different topics, and the user’s relationship (e.g. direct-report, team member) with each of his/her email contacts in an enterprise environment.

icarus priority inbox screenshot

The email prioritization application automatically identifies high-priority emails in a user’s inbox based on the message metadata, content, and relevant information from the user model, and displays high-priority messages in a separate section from other emails in the interface, with the goal of assisting the user in his/her daily triage of incoming messages. It dynamically combines a global priority classifier (established from labeled training data of multiple users) with a user-specific classifier built from ongoing feedback training to achieve the balance between reliability and adaptability.

icarus contextual IR screenshot

The contextual IR application uses the user model, and the information from the current email message, to determine what content is relevant (e.g. documents from hard drive or online resources, information about sender). The content provided varies based on who the sender is, what the previous interaction has been between the sender and the user, and the topic of the current message. The retrieved information is provided "at the user's fingertips" within the email tool so that the user doesn't need to leave his/her current context to gather necessary information.


For more information, please contact Jennifer Lai (jlai@us.ibm.com) or Jie Lu (jielu@us.ibm.com).