Knowledge-Based Agents for Advanced Information Access

Research project P 13871 Knowledge-Based Agents for Advanced Information Access Thomas EITER 11.10.1999 The importance of accessing data and information scattered over a number of different sites, which are connected through wide area networks such as the Internet, has been rapidly increasing. The World Wide Web (WWW), for example, provides a huge wealth of vastly unstructured, heterogeneous data which is difficult to search. There are basically two approaches to tackle this problem: one is to use software robots which traverse autonomously across the Web, the other is to employ dedicated information agents embedded in a given multi- agent environment. The first approach is in general not very effective because search results may contain too much irrelevant information; popular search engines like Alta-Vista or Yahoo belong to that category. Multi-agent environments provide a better approach to access information from heterogenous information repositories. A multi-agent architecture contains different kinds of agents, each of which has a special purpose, and where the agents have the ability to cooperate with each other in order to perform their special duties. Moreover, one of the distinguishing properties of an agent is its capability of reasoning. This capability is in particular needed during the decision-making process, when the agent has to determine which actions are appropriate in response to a given stimulus. An agent should be able to come to plausible conclusions given less than certain information. In order to predict and understand the behaviour of agents, it is important to have suitable principles and formal methods which describe the agents. Current systems do not pay sufficient attention to these points. In this project, we plan to research methods and techniques for intelligent information access in a multi-agent environment, which are based on formal grounds and equipped with a well-defined semantics. We shall investigate to what extent existing principles and methods from the area of knowledge representation are suitable for this purpose, and which may be exploited for the construction of intelligent information agents, possessing advanced reasoning capabilities, including an appropriate treatment of incomplete information, default behavior, and the handling of preferences. The particular needs of information agents will request for adaptations and domain- specific methods, though, which will be developed along with suitable algorithms for that methods. The results of these investigations will be examined on a concrete application, which will be built in the context of the IMPACT agent environment.

We developed knowledge representation and reasoning components for intelligent information agents, which are based on formal methods and allow for intelligent knowledge base updates and information site selection. At present, the World Wide Web faces several problems regarding the search for specific information, arising, on the one hand, from the vast number of information sources available, and, on the other hand, from their intrinsic heterogeneity. A promising approach for solving the complex problems emerging in this context is the use of information agents in a multi-agent environment, which cooperatively solve advanced information-retrieval problems. An intelligent information agent provides advanced capabilities resorting to some form of logical reasoning, based on ad hoc knowledge about the task in question and on background knowledge of the domain, suitably represented in a knowledge base. In this project, our interest was in the role which declarative methods from the field of logic programming can play in the realization of reasoning capabilities for intelligent information agents. We first considered the task of updating a knowledge base, since, in order to ensure adaptivity, an agent`s knowledge base is subject to change. To this end, we developed update agents, which perform knowledge base updates following a formal policy and which are implemented in the IMPACT agent environment. These update agents adhere to a clear semantics and are able to deal with incomplete or inconsistent information in an appropriate way. Furthermore, we considered a particular problem of information agents, namely information source selection, and developed an intelligent site-selection agent. The problem here is, given a query by a user, which out of a collection of information sources should be selected for answering the query, such that the utility of the answer, in terms of quality of the result and other criteria (e.g., costs), is as large as possible for the user? We developed a knowledge-based approach to this problem, which formally analyses the query and uses background knowledge about the query domain to infer the most promising information site to be queried. If, for example, we are looking for the release date of the movie `Arsenic and Old Lace` and we can infer from the background knowledge that Cary Grant acted in it and if there exists a site with detailed information concerning Cary Grant, then of course we expect this site to be queried, although Cary Grant is not explicitly mentioned in the query. To allow for such a decision process to be performed automatedly by an intelligent information agent, a site selection component has been implemented on top of the DLV KR system and its PLP front-end for prioritized logic programs. We report experimental results for this implementation, obtained using a representative example from a movie domain.