23
ContextSensitive Search Literature Review By: Noran Hasan

Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Context‐Sensitive Search Literature Review

By: Noran Hasan

Page 2: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

The Problem with Search as We Know ItThe Problem with Search as We Know It

• Search queries are often too short or too vague Sea c que es a e o te too s o t o too agueresulting in too many results, many of which are irrelevant.

• The large number of results are practically impossible to be fully browsed by any user.

• The use of ubiquitous devices, such as smart phones, emphasize these problems due to their limited screen space limited ease of use and thelimited screen space, limited ease of use, and the nature of the user’s activity while using it (usually on the move).)

Page 3: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Typical Search ProcessTypical Search Process

QQueryResults

Search Engine Results

Page 4: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Possible Improvement StrategiesPossible Improvement Strategies

Refine Query Modify engineFilter

QQueryResults

Search Engine Results

Page 5: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Using Context to Improve SearchUsing Context to Improve Search

• Context can be used to improve search by oneContext can be used to improve search by one of the following approaches:– Using context to refine search by adding context– Using context to refine search by adding context words to query.

– Using context to filter results and getting the mostUsing context to filter results and getting the most relevant.

Page 6: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Searching with ContextSearching with ContextKraft et. Al in [1] present 3 algorithms to implement contextual search:search:

Page 7: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Searching with ContextSearching with Context– Query rewriting: 

• query AND context.• Less results (danger of low recall)• Less results (danger of low recall)

– Rank‐Biasing: • requires a modified search engine.

t t i ti l ki t• context is an optional ranking term.• has the same level of recall as query without context. • E.g. <query> =  <selection=  cat> <optional = persian, 2>

It ti filt i t h– Iterative filtering meta‐search:1) Candidate context words are generated2) Query template is applied 

– E.g.  Query q and context words a b  cAll ibl bi ti b b b b» All possible combinations: qa, qb, qc, qac, qab, qac, qbc,  qabc

» If window size is 2: qab, qac, qbc» Force to contain high ranked term, e.g. if a is high ranked, then: qa, 

qac, qab, qabc3) Results from multiple queries are aggregated.

Page 8: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Desktop ContextDesktop Context

• In desktop search context may be inferredIn desktop search, context may be inferred from:– Previous queries– Previous queries

– URLs clicked/recently browsed documents

Documents on the desktop– Documents on the desktop

– Activities in standard applications

Page 9: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Desktop Context (cont’d)Desktop Context (cont d)

• In [2], they rank documents not just to the relevance to current query, but to the series of queries issued in the current session.

• The table below shows some search logs:The table below shows some search logs:

Page 10: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Desktop Context (cont’d)Desktop Context (cont d)

• In [3] previous queries and URLs clicked areIn [3], previous queries and URLs clicked are used to identify the context.– If a user issues the query “Micheal Jordan”– If a user issues the query  Micheal Jordan  preceded by the query “NBA”, the  user probably intends to query about the basket ball player not q y p ythe machine learning researcher at UC Berkeley.

• In [4], they summarize desktop documents [ ], y pand select terms to be used as context works.

Page 11: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Desktop Context (cont’d)Desktop Context (cont d)

• In [5] query expansion and re‐ranking areIn [5], query expansion and re ranking are performed based on recently browsed documents with special focus on news domains.p

• In [6], context is extracted from user’s activities in standard applications such as word processors.standard applications such as word processors. The context is used as a query to PRISM’s internal system to find a search engine that covers similar y gtopics. If no match is found, the query is forwarded to general‐purpose search engines.

Page 12: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

How is Mobile Search Different?How is Mobile Search Different?

• Mobile search has introduced:Mobile search has introduced:1. Limitations, in terms of limited 

hardware/software capabilities. • Not all pages can be viewed on mobile devices.• Deficient input emphasize the need for intelligent 

applications to help the user reach his target usingapplications to help the user reach his target using minimum keystrokes/browsing.

2. Increased amount of accessible of context information.  Mobile devices can provide more interesting information that can be used for contextcontext.

Page 13: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Aiding Mobile SearchAiding Mobile Search

• The are several ways enhance the user’sThe are several ways enhance the user s experience while performing mobile search:– Query suggestions that appear as the user enters– Query suggestions that appear as the user enters the query

– Expanding the query or filtering search results toExpanding the query or filtering search results to achieve the best precision

– Only return results that can be viewed by theOnly return results that can be viewed by the querying device

– A combination of some/all of the above./

Page 14: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Dealing with Mobile H/W and S/W llimitations

• In [7] they rank contents not only on the• In [7], they rank contents not only on the textual relevancy to the user’s query but also how adaptable it is to the target mobile p genvironment.

• The main components in this system:– Delivery Context Categorization (legacy, Under DDC, DDC, Over DDC, Advanced, Computer)

– Adaptive Context Crawling (off‐line, sends multiple HTTP t t di if it i d i d t ti )requests to discover if site is doing adaptation)

– Web Page Classification (mobile optimized/mobile friendly/mobile unaware).Mobile Readiness Ranking (optimal/fair/bad/not valid)– Mobile‐Readiness Ranking (optimal/fair/bad/not valid)

Page 15: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

What Kind of Context Data Do Mobile Devices Provide?

• According to [12] ubiquitous computing• According to [12], ubiquitous computing environments are characterized by many sensors that can sense a variety of different contexts:

physical contexts (like location time)– physical contexts (like location, time)– environmental contexts (weather, light and sound levels)i f i l ( k )– informational contexts (stock quotes, sports scores)

– personal contexts (health, mood, schedule, activity) – social contexts (group activity, social relationships, other people in a room)

– application contexts (email  received, websites visited)

– system contexts (network traffic, status of printers)

Page 16: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Mobile Context: LocationMobile Context: Location

• Location is one of the most important contextLocation is one of the most important context data provided by mobile devices that desktops cannot providecannot provide.

• [8] proposes query refinement based on real‐world contexts of a mobile user such asworld contexts of a mobile user, such as his/her current geographic location and the typical activities at the location which aretypical activities at the location which are extracted by Blog mining.

Page 17: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Mobile Context: Location (cont’d)Mobile Context: Location (cont d)• Method:

b l– Mobile user inputs query – System Senses Real‐World Contexts (obtain geographical location)– System Translates Contexts into Contextual Words (the typical activities at the 

location) by mining blog data .– System Assigns Weight to Contextual Word– Query Refinement is enforced (generate alternative queries consisting of 

original queries and contextual word.  Let the user pick the query or return results of query with the contextual word with highest weight)

• Example: a mobile user in a bookstore issues [“da vinci”] as an original query, our system would infer from his/her original query and current place‐name as a real‐world context that he/she is requesting information about not a movie but a book of “da vinci code”, and offer or retrieve about ot a o e but a boo o da c code , a d o e o et e eautomatically by [“da vinci code” AND “book” AND “buy”] as one of its alternatives.

• Proposed future work: utilize not only current real‐world contexts but also history of continuous past ones and/or prospective oneshistory of continuous past ones and/or prospective ones 

Page 18: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Mobile Context: Location (cont’d)Mobile Context: Location (cont d)• [9] proposes a system for mobile/web searches that provides the following 

functionality:functionality:– context‐aware keyphrase inference– subtopic tree generation and context‐aware re‐formation– discovery of comparable keyphrases from the Web– meta vertical search focused on one subtopicmeta vertical search focused on one subtopic

• Future work:– utilize not only current but also past continuous and/or prospective real‐world contexts– personalize context‐based weightings based on user profiles and/or personal query logspersonalize context based weightings based on user profiles and/or personal query logs

Page 19: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Mobile Context: Location (cont’d)Mobile Context: Location (cont d)

• [10] presents a new search browser interface in which context, in the form of time and location information isand location information, is integrated with preference information derived from the queries of like‐minded qcommunities of mobile searchers. 

• The sample communities the t t dprototype used are 

entertainment and tourist sites. 

Page 20: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

Mobile Context: Location (cont’d)Mobile Context: Location (cont d)

• In [11], mobile phone users are encouraged to record lti di bl th fl i h d ith i t fmultimedia blogs on‐the‐fly, enriched with inputs from 

other physical sensors. • The blogs are geotagged and uploaded to a remote server 

that positions these blogs on a spatial platform (e g map)that positions these blogs on a spatial platform (e.g. map)• Internet users can zoom into any part of the map and 

browse multimedia blogs at those locations. Moreover, users may query selected regions for desired information.y q y g

• The Internet user selects a region and sends a query to phones in that region.

• The phone user replies to it, and the reply is transmitted b k hback to the server.

• The query and reply are associated to the blog.

Page 21: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries
Page 22: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

ReferencesReferences

1) Reiner Kraft, Chi Chao Chang, Farzin Maghoul, Ravi Kumar, “S hi ith C t t” P di f th 15th i t ti l“Searching with Context”, Proceedings of the 15th international conference on World Wide Web, 2006

2) Towards Context‐Aware Search by Learning A Very Large Variable Length Hidden Markov Model from Search Logs © 2009Length Hidden Markov Model from Search Logs © 2009

3) Context‐Aware Query Classification © 20094) Summarizing Local Context to Personalize Global Web Search 

©2006©20065) Using the Current Browsing Context to Improve Search Relevance 

© 20086) Towards Context‐Based Search Engine Selection ©2001) g7) A crawling and ranking method for improving context‐awareness 

in mobile web search © 2008

Page 23: Context Sensitive Literature Revierafea/CSCE590/Spring10/Noran/Context-Se… · Literature Review By: Noran Hasan. The Problem with Search as We Know It • SeaSea crch queque esries

ReferencesReferences

8) Activity‐Based Query Refinement for Context‐Aware Information Retrieval © 2006 

9) ReCQ: Real‐World Context‐Aware Querying © 200710) Who, What, Where & When: A New Approach to Mobile10) Who, What, Where & When: A New Approach to Mobile

Search © 200911) Micro‐Blog: Sharing and Querying Content Through 

Mobile Phones and Social Participation ©2008Mobile Phones and Social Participation ©200812) Anand Ranganathan, Roy H. Campbell, “A Middleware for 

Context‐Aware Agents in Ubiquitous Computing Environments”, IFIP International Federation forEnvironments , IFIP International Federation for Information Processing 2003

13) Query Suggestions for Mobile Search: Understanding Usage Patterns ©2008Usage Patterns ©2008