Paper Info
Title
A Case Study in Engineering a Knowledge Base for an Intelligent Personal Assistant
Abstract
We present a case study in engineering a large knowledge base to meet the requirements of a personal assistant. The agent is designed to function as part of a semantic desktop application with the goal of helping a user manage and organize his information as well as support the user in performing day today tasks. We discuss our development methodology and the knowledge engineering challenges we faced in the process. The use of ontologies, metadata annotations, and semantic web protocols on desktop computers will allow the integration of desktop applications and the web, enabling a much more focused and integrated personal information management as well as focused information distribution and collaboration on the Web beyond sending emails. In this paper, we present our experience in constructing a large knowledge base (KB) designed specifically to support a personal assistant called CALO (Cognitive Assistant that Learns and Organizes). CALO is a multidisciplinary project funded by DARPA to create cognitive software systems that can reason, learn from experience, be told what to do, explain what they are doing, reflect on their experience, and respond to surprises. CALO KB uses an upper ontology called Component Library (CLIB) (Barker, Porter et al. 2001) and off-the-shelf standards such as iCalendar, as starting points and extends them to meet the requirements of CALO. The primary contribution of this paper is in providing a comprehensive description of the process of engineering a large knowledge base. CALO offers unique functionality by integrating an impressive array of AI technologies (more than 100 major software components spanning machine learning, planning, and reasoning written in about 10 different programming languages spanning Lisp, Prolog, and Java), and the effort of pulling them together into a semantic whole is unprecedented. The CALO KB effort as presented here has been a key ingredient in accomplishing the goal of semantic integration in CALO, and much can be learned from the description of this experience by others interested in undertaking efforts to construct large knowledge bases. We begin this paper by identifying the knowledge requirements of CALO and then describe how we developed the KB. We then give an overview of the knowledge content, and discuss three ontological challenges in some detail. We give a review of tools that we used in the process and conclude with a comparison to related work and by identifying research issues suggested by this experience. 2 Knowledge Requirements in Project CALO CALO's role is to know and do things for its user, and therefore it must have knowledge about the environment in which the user operates and the user's tasks. It is best to understand the knowledge requirements of CALO in terms of the major six functions it performs. Details of implementing the functions are not the focus of the present paper. 2.1 Organize and Manage Information From a user's information about emails, contacts, calendar, files, and to-do lists, CALO learns an underlying relational model of the user's world that provides the basis for higher-level learning. The relational model contains information such as the projects a user works on, which project is associated with which email, the people a user works with, and in what capacity. Providing such functionality requires vocabulary to relate information across the email, contact records, calendar entries, information files, to-do lists, and people. Since much of the learned information is probabilistic, there is a need to provide a way to absorb and maintain this knowledge with the symbolic knowledge.
Year
Venue
Keywords
2006
SemDesk
ontologies,intelligent assistants,semantic desktop,case study,software systems,semantic web,machine learning,knowledge base,functional requirement,personal information management,semantic integration,relational model,knowledge engineering,software component,programming language
Field
DocType
Citations 
Knowledge integration,Semantic desktop,Computer science,Personal knowledge management,Knowledge management,Knowledge-based systems,Knowledge engineering,Knowledge base
Conference
6
PageRank 
References 
Authors
0.55
19
6
Name
Order
Citations
PageRank
Vinay K. Chaudhri1587246.49
Adam Cheyer2527119.15
Richard Guili360.55
Bill Jarrold460.55
Karen L. Myers5833114.14
John Niekarsz660.55