Automating To-Do Lists for Users: Interpretation of To-Dos for Selecting and Tasking Agents

To-do lists have been found to be the most popular personal information management tools, yet there is no automated system to interpret and act upon them when appropriate on behalf of the user. Automating to-do lists is challenging, not only because they are specified as free text but also because most items contain abbreviated tasks, many do not specify an action to be performed, and often refer to unrelated (personal) items. This paper presents our approach and an implemented system to process to-do list entries and map them to tasks that can be automated for the user by a set of agents. Since the format of to-do entries is not very amenable to natural language processing tools that can parse and create a structured interpretation, our approach is to exploit paraphrases of the target tasks that the agents can perform and that specify how the free-text maps to the task arguments. As users manually assign to-do to agents for automation, our system improves its performance by learning new paraphrases. We show an evaluation of our approach in a corpus of 2100 to-do entries collected from users of an office assistant multi-agent system.