Task4

From Kolflow Project

Title : Traces and explanations: documenting inferences, query solving and interac- tions

Code : 4

Responsible : Edelweiss

Activities : Design and experiment tracing and explanation approaches

Start Date :

End Date :

Objectives : As the knowledge bases grow larger and as more and more actors (human users, artificial agents) contribute to these knowledge bases, the ability to trace and understand the activity is becoming vital. This task will focus on local means an individual knowledge base should provide to trace and analyse the users’ interactions both for the edition and the querying of the knowledge, each action being the subject of a subtask.

Subtask T4.1: "Alter Ego Assistant" will lead to the design of an assistant based on traces to help users to better understand the Kolflow environment. This subtask will be conducted between month 01 and month 18. The results of this subtask will then be reused and improved within task 5. Subtask T4.2: will be conducted in parallel from month 01 to month 36.

Work plan This task is directly linked to one of the PhD subjects defined in this proposal and includes several milestones

1. a state of the art on query and inference documentation and explanation (T0+6);
2. an approach to explain basic query mechanisms (e.g. projection) (T0+10);
3. a way to provide performance and errors indicators (e.g. most significant points of failure) (T0+15);
4. an extension to explain ontology-based reasoning during query-solving (T0+20); 5. a way to suggest queries and changes to queries (T0+26).

Success criteria : The ultimate goal of this task would be to demonstrate a query interface implementing explanation and suggestion means.

Risks : Query documentation on the semantic web is not really addressed so far and the query mechanisms are more often optimized and more and more complex than opened and explained. One of the major risks is the fracture between the complexity of the query solving algorithm and data structures on one hand and the simplification of interfaces on the other. However we believe that the use of formal knowledge together with its natural language documentation can help automate the explanation.

Deliverables :

Sub-tasks :

Participants :

• GDD (2)
• Edelweiss (67)
• Orpailleur (6)
• Silex (25)