Transcript Associative Query Answering via Query Feature Similarity

Associative Query Answering via
Query Feature Similarity
Outline
•
•
•
•
•
Associative Query Answering
Approach and system overview
Database schema and semantic model
Query feature and similarity
Searching associative attributes from case
bases
• Conclusions
Associative Query Answering
To provide additional relevant information to
the queries that:
• not explicitly asked
• user does not know how to ask
Examples
• “List airports in Tunisia that can land a C-5 cargo
plane.”
Associative information depends on user type
– Planner: railway facility near the airports.
– Pilot: runway condition and weather condition.
• “Tourists ask visitor’s information about a city.”
Additional information depends on the selection
condition.
– Florida: hurricanes
– California: earthquakes
Associative Information
• For a relational query:
1. Simple associative attributes: attributes of
relations in the query
2. Extended associative attributes: attributes of
relations introduced to the query by joins
3. Statistical associative information: aggregate
functions related to the entity in the query
• We focus on the first two types
Approach
Similar case searching based on:
• User type: case bases are separate for user
types
• Query context: query features
• Similarity measure: based on domain
knowledge represented by semantic model.
Property of Semantic Model
•
•
•
•
Semantic model derived from database schema
As a directed graph
Nodes: entities and complex associations
Edges: relationships among nodes, including userdefined
• Weights on edges express the relative information
of content
• Nodes have equal weights
Conclusions
• Query feature vector as a query representation for
similarity matching.
– Query topic
– Output attribute list
– Selection constraints
• Developed a new type of semantic model constructed from
database schema, user types, and user-defined
relationships.
• Methods to evaluate similarity measure of case base
queries based on query feature vector via the semantic
model.