Contextual Ranking of Keywords Using Click Data ICDE`09 Utku Irmak Vadim von Brzeski Vadim von Brzeski Reiner Kraft
Outline Introduction Contextual Shortcuts Methodology Feature Space Evaluation and Results Framework Conclusion
Introduction Application ◦ Contextual advertising ◦ Text summarization ◦ User-centric entity detection system Approach
Contextual Shortcuts Entity Detection ◦ Pattern based ◦ Named ◦ Concept Generating a Concept Vector ◦ Term vector ◦ Unit vector
Contextual Shortcuts 1)in term,not in unit 2)in unit, not in term 3)in both
Methodology (a) Determining whether the entity is relevant to the given context (b) Determining whether the entity is interesting outside of the context
Feature Space Interestingness of a Concept ◦ Search Engine Query Logs ◦ Search Engine Result Pages ◦ Text Based Features ◦ Taxonomy Based Features ◦ Other Relevance of a Concept in a Context ◦ Search engine snippets ◦ Related query suggestions
Evaluation and Results Cross Validation Approach [A,B,C,D] ◦ R1=[A,B,D,C] R2=[B,A,C,D] [(A, 0.15), (B, 0.05), (C, 0.02), (D,0.01)]
Evaluation and Results
Editorial Evaluation
Framework
Conclusion Utilize User-click feedback, interestingness and relevance to ranking the key concept that to improve overall performance.