A smarter interface for e-mail
Context: Many people and businesses rely almost entirely on e-mail to manage diverse transactions. But e-mail programs are optimized to manage messages, not to-do lists. Technologies to make e-mail more useful have tried aggregating messages that have a common header or tagging messages as associated with specific, predefined tasks. A new way to free workers from sifting through copious messages—a machine-learning algorithm that automatically keeps track of tasks, and which e-mails are associated with them—hails from Nicholas Kushmerick at University College Dublin and Tessa Lau at IBM.
Methods and Results: First, the algorithm groups e-mails according to the transactions they’re part of, which it deduces by identifying, say, order numbers or clients’ names. Next, messages are grouped by the events they represent, such as shipping notifications or order confirmations. Combining these two perspectives, the algorithm looks for common patterns, or workflows, that recur across a given set of transactions. For example, e-commerce transactions typically involve order notification, shipping notification, and messages about delayed or modified orders.
On the basis of such patterns, the algorithm automatically determines the status of a given transaction. Without requiring user input or manually labeled examples, the algorithm correctly identified the transaction stages represented by 101 out of 111 messages in an e-commerce test set representing 39 transactions by six vendors.
Why it Matters: A 2003 survey of major industries found that more than 90 percent of organizations use e-mail to respond to customer inquiries, and about 70 percent use e-mail for invoicing and contract negotiation. Kushmerick and Lau envision their algorithm as the core of an interface that automatically organizes e-mail by task as easily as by date or sender. By learning workflows, the algorithm can facilitate even specialized processes, which gives it an advantage over techniques that rely on message headers or preformatted content. Eventually, this technique and others like it should help convert cluttered in-boxes into a set of well-oiled workflows.
Source: Kushmerick, N., and T. Lau. 2005. Automated e-mail activity management: an unsupervised learning approach. Proceedings of the 10th International Conference on Intelligent User Interfaces, pp. 67–74.