Browse Prior Art Database

Context-Dependent E-Mail Classification

IP.com Disclosure Number: IPCOM000013641D
Original Publication Date: 2001-Mar-01
Included in the Prior Art Database: 2003-Jun-18

Publishing Venue

IBM

Abstract

Current e-mail classification mechanisms use incremental machine learning techniques for e-mail classification. A known e-mail classification system uses a modified decision tree algorithm which uses word frequency for e-mail classification. Such system work well if the task is, for example, to route incoming customer requests to the right customer representative and to automatically display a short list of mail-templates for answering the request. The reason for this is that the request and the responses to these requests are not context-dependent and do not change much over time. The situation is quite different when the goal is to sort the mail of office workers engaged in project-oriented work. Classification here will be based not only on the content and the sender of the massage, but as well on the status of the projects the receiver is involved in. For example, the same message from the secretary "Here are your flight and hotel reservation details" will be filed under very different projects each time a trip is done for a different project. The disclosed mechanism assumes that each project an office-workers is working has a set of filing agents (or other computer interpreted records) which listen for the expected incoming mails for that project. So, for example, by sending a request for trip planning to the secretary in the context of a project a user automatically generates a filing agent that will listen for incoming messages from the secretary with respect to trip planning. Once the filing agent detects such a message it will file it into the associated project. Furthermore the agent might also be set up in such a way that it issues a warning if a matching message did not arrive within a specified time frame. Thus, as a by-product of working on a project and issuing requests to other users filing agents for context dependent e-mail classification are set up. In addition to these implicit setup mechanism filing agent can also be set up explicitly for each project by specifying expected future events. The proposal disclosed here does not try to replace current mechanisms for e-mail classification, but to augment it. Filtering agents as discussed in the examples above will use "classical" classification mechanisms to determine their applicability to a given e-mail.