1. Searching with context
Searching with context
Presented by
Ana Hilda Morales
R Kraft, C.C. Chang, F. Maghoul, and R. Kumar.
Proceedings of the Fifteenth International
World Wide Web Conference NY, ACM Press
1/18/2019

2. Outline Introduction Algorithms Methodology Conclusions
Query Rewriting
Rank-Biasing
Iterative, Filtering Meta Search
Methodology
Conclusions
1/18/2019

3. Introduction
Contextual search refers to proactively capturing the information need of a user by automatically augmenting the user query with information extracted from the search context.
1/18/2019

4. Introduction Contextual search User interface Y!Q [Kraft et al, 2005]
Extracting and representing the context
Presentation layer
User interface
Front-end
Logic layer
Data layer
Back-end
1/18/2019

5. Terminology
Context: a piece of text (e.g., a few words, a sentence, a paragraph, an article) that has been authored by someone.
Query: regular keyword search queries.
Context term vector: subset of the words/phrases/entities in the content of the context.
Contextual Search Query: a search query that comprises a keyword query and context.
Query-less: when the keyword query is empty
1/18/2019

6. Terminology Families of Queries
Simple Query SQ: keywords based search queries.
Complex Query CQ: keyword based search queries and typically consist of keywords or phrases plus ranking operators.
Standard Search Engine (Std. SE): refers to web search engine.
Modified Search Engine (Mod. SE): refers to a web search engine that has been modified to support complex search using rank-biasing operators.
Contextual Search Engine (CSE): is an application front-end that supports contextual search queries. .
1/18/2019

7. Algorithms Query Rewriting (QR) Rank Biasing (RB)
Iterative, Filtering Meta-Search (IFM)
1/18/2019

8. Query Rewriting (QR)
Send one simple query per contextual search query to a standard search engine
Query
Generator
1/18/2019

9. Process QR Creating queries composed of all query and context
term vector
using AND semantics
Input:
Simple Query
Result set
Parameter: The number of terms taken from the terms vector
1/18/2019

10. Example QR QR1 q (a) QR2 q (a b) q (a, b, c, d, e, f)
QR5 q (a b c d e)
q (a, b, c, d, e, f)
Result set
1/18/2019

11. Advantages and Disadvantages QR
This approach is simplicity, especially since conjunctive semantics is supported in all major search engines.
Not requires a modified search engine back-end
The more terms are added in a conjunctive way, the more restricted the query is, and the less results it will likely return.
The query and context term vector together may comprise more terms than the search engine supports for evaluation.
1/18/2019

12. Rank-Biasing (RB)
Send one complex query per contextual search query to a modified search engine.
Query
Generator
1/18/2019

13. Generating and sending
Process RB
Generating and sending
a complex query that
contains ranking
instructions
to modified search
engine back-end
Input:
Parameters
Result set
Parameters: Number of selection terms to use
Number of rank operators
Weight multiplier for each RANK operator
1/18/2019

14. Example RB
<query> = <selection=cat> <optional=persian, 2.0>
Weight
Selection terms
Rank operators
<query> = <selection=peugeot306> <optional=sale, 5.0><optional=blue, 2.0>
1/18/2019

15. Advantages and Disadvantages RB
List of the results ordered sensibly
Requires a modified search engine back-end
The need for a specialized search engine for back-end support
1/18/2019

16. Iterative Filtering Meta-search (IFM)
Send multiple, simple queries per contextual search query to a standard search engine. The model is based on the concept of meta-search.
Query Generation
Ranking/Filtering
1/18/2019

17. Structure IFM
Query
Generator
1/18/2019

18. Process IFM Parameters: Number of terms taken from the terms vector
Creating queries composed of all
query and context term vector
using AND semantics
Result
Input:
Multiple
Simple Query
Re-Rank
Filter
Result set
Creating queries composed of all
query and context term vector
using AND semantics
Result
Creating queries composed of all
query and context term vector
using AND semantics
Result
Parameters: Number of terms taken from the terms vector
1/18/2019

19. Example IFM Combinations q (a) q (a b) q (a b c) q (a,b,c,d)
q (b) q (b c) q (a b d)
q (c) q (c d) q (a b c d)
q (d)
IFM-SW1 q (a), q (b), q (c), q (d)
IFM-SW2 q (a b), q (b c), q (c d)
IFM-SW4 q (a b c d)
q (a, b, c, d)
Result set
1/18/2019

20. Advantages and Disadvantages IFM
Not requires a modified search engine back-end
Dependency of the standard search engine
1/18/2019

21. Methodology 24,566 judgments from 28 expert judges
Judgments arose out of a benchmark of 200 contexts.
Is this result relevant to the context?
“Yes”
“Somewhat”
“No”
“Can’t tell”
1/18/2019

22. Conclusions
QR, a simple technique that closely emulates human query reformulation and can be easily implemented on top of a commodity search engine, performs surprisingly well and is likely to be superior to manual reformulation.
RB and IFM break the recall limitations of QR
IFM is very effective and outperforms the others both in terms of recall and relevance at an added engineering cost.
These three techniques offer a good design spectrum for contextual search implementors.
1/18/2019

23. References
R Kraft, C.C. Chang, F. Maghoul, and R. Kumar. Searching with Context. In Proceedings of the Fifteenth International World Wide Web Conference (WWW-06), New York, NY, ACM Press.
R. Kraft, C. C. Chang, and F. Maghoul. Y!Q: Contextual search at the point of inspiration. In Proceedings of the 14th Conference on Information and Knowledge Management (CIKM), pages 816–823, 2005.
1/18/2019

24. Thank you for your attention?
1/18/2019