Design principles for long‐lived Internet agents

Prior research on intelligent Internet agents has failed to address the needs of long‐lived data‐collecting agents, focusing instead on short‐lived transaction agents. Transaction agents typically run for a few seconds and retrieve information for a single transaction. With the staggering growth of electronic commerce, researchers and practitioners will want to design long‐lived data‐collecting agents that intelligently search for, retrieve, interpret, categorize, and store vast amounts of related information each time that they run. Such agents can run over the course of days rather than seconds and can be used by practitioners for decision support applications or by researchers as part of an empirical research methodology. This paper proposes a framework for agent sophistication, and emphasizes a number of design concepts for long‐lived Internet agents, including intelligence, validation, concurrency, recovery, monitoring, and interactivity. These concepts are used in the development of an illustrative tool called Electronic Data Retrieval LexicaL Agent (eDRILL), an object‐oriented data‐collecting agent. eDRILL is designed using the Unified Modeling Language (UML) and is written in Java. It gathers research data from an online auction. © 2000 John Wiley & Sons, Ltd.