WebPlan - Dynamic Planning for Domain-Specific Search in the Internet

The objective of the project WebPlan is the development of a generative, interactive tool for planning-based Information-Gathering in the Internet. Because of the enormous increase of available information in the internet, searching for needed information has become a nontrivial task. The various existing search engines frequently respond to queries with a wealth of answers but the most interesting are often missing. The main reason for this is, that only syntactic criterions are considered, whereas the semantic and pragmatic aspects of the query are neglected.
In field-specific search it is possible to overcome these deficits, since field-specific background knowledge could lead to the real purpose of the query. Moreover, searching has mostly a dynamical character, because the query develops itself during searching. So, action planning is a appropiate method to do field-specific dynamical search, since the combination of Information-Gathering and deductive actions can solve the problems mentioned above.

For this approach the classical AI planning was extended with following features:

  • handling uncomplete knowledge
  • interleaving of planning and execution
  • integration of Information-Gathering actions
  • supporting interactive planning

Funding

WebPlan was funded by the "Deutsche Forschungsgemeinschaft (DFG)". The project was carried out between 1998 to 2002.

Architecture of the WebPlan System

 

\begin{figure}

Planning Kernel: This module realizes the basic planning algorithm for domain-specific Internet search. It can be seen as an extension of the planning kernel of CAPlan, an SNLP-based plan-space planner.
Action Execution: This component is responsible for controlling the execution of actions. It interacts frequently with the planning kernel which determines operators that should be executed. The execution module monitors the execution of the actions which correspond to the operators and gives the results to the planning kernel.
Query Interpretation: In this module the user's query has to be transformed into a planning problem.
Knowledge Base: The knowledge base consists of three main parts: (1) The definition of the deductive and information seeking operators. This declarative knowledge could be compared with the domain definition in a classic planning system and is needed by the planning kernel. (2) The code for the domain-specific information seeking actions. This procedural knowledge is needed by the execution module. (3) The vocabulary and the mechanisms for the queries. This knowledge is mainly used by the Query Input, Control and Interpretation component.
Knowledge Base Editor:
This editor supports the application developer to create and maintain the domains. It offers mechanisms to edit the three parts of the knowledge base mentioned above.
Control: The control module of the WebPlan system monitors the search process and presents the current state (current plan and results of executed actions) to the user, which is done via the Result Presentation module. The user should be able to interfere in the search process as described in the preceding sections.
Result Presentation:
This component works up the search results and presents them to the user. Retrieved information could be used to reformulate the initial query.
Query Input & Control: This module implements a WWW-User-Interface for the input and refinement of queries, as well as for interactive control of the search process.

Research Team

Two members of the Wi2-group have been working within WebPlan.

Publications

  • Jochem Hüllen, Ralph Bergmann, and Frank Weberskirch. WebPlan: Dynamic planning for domain-specific search in the Internet. In J. Köhler, editor, 13. Workshop "Planen und Konfigurieren" (PuK-99), 1999.