Future Directions


One of the keys to making a background knowledge-oriented approach work is support for gathering and maintaining the domain knowledge. We have directly or indirectly been involved in using knowledge representation techniques to support reasoning and retrieval in a number of systems (e.g.,  [13, 7, 2, 3]) only to see the knowledge acquisition task become the overwhelming part of the project. Our goal is to provide a collaborative topic-building environment which allows domain experts to input information about ways that the topic set should be expanded or otherwise modified. We are experimenting with an interface in our community site which allows content providers to note words in our vocabulary for which they think their pages should be retrieved as matching documents. Our initial work is concerned with synonym, subclass, and instance relationships. We are not automatically integrating this into the topic hierarchy but instead we are using it to support the owner of the vocabulary in the evolution task. Issues that we are evaluating in our environment are:
  1. Understanding - We need an interface which can teach users about the notions of class relationship, instance relationship, and synonyms.
  2. Compliance - User understanding may not always lead to compliance. We also need to convince the users of the benefits of compliance with our approach.
  3. ``Cheating'' - Clever users can discover ways to cheat the system once they understand how our approach works. We need to understand how much of a problem cheating is likely to be and then if it is noticeable, devise ways to limit its impact.
  4. Interest (Apathy) - A collaborative approach only works if we can convince our users to participate.

To further support maintenance, we are encoding the background knowledge in a more controlled language for object organization- the CLASSIC knowledge representation system[1]. This will then automatically identify conflicts when subclasses may be stated incorrectly as synonyms or superclasses and may also be used to identify many more complex relationships.

We also would like to describe all of our possible content sources in the knowledge base which could then support automatic identification of which sources to search from a user query. This method has already been used quite successfully in the Information Manifold project[9].

Finally, we will be gathering usage data to evaluate our experiments with background knowledge and collaborative environments for support of domain knowledge evolution.


•Abstract
•Background
•Search Goals
•FindUR: A Hybrid Approach to Search
•Future Directions
•References