23. juli 20101 by benjamin riedel collaborative web
TRANSCRIPT
Benjamin Riedel 2
Outline
• A Survey of Collaborative Web Search Practices (Meredith Ringel Morris, 2006)
• Social summarization in collaborative web search (Oisín Boydell and Barry Smyth, 2009)
• Discussion
Benjamin Riedel 3
A Survey of Collaborative Web Search Practices (2006)
• Survey of the using of collaborative web search
• Demographics: 204 knowledge workers at a technology company 80,4% male 21 to 61 years old (median 36) 38% researchers, 22% software developers, 17% program managers 73,5% see themselves as web searching experts
Benjamin Riedel 4
What counts as Collaborative Web Search?
• Term is not clear:
While only 53,4% of participients say they ever cooperated to search the web, only 2,9% have not used any of the collaborative search activities listed in the study
People do collaborate when searching, but they are not aware of it!
Benjamin Riedel 5
Most frequent engaged collaborative activities
Watched over someone‟s shoulder as he/she searched the Web, and suggested alternate query terms. (87,7%)
E-mailed someone links to share the results of a Web search. (86,3%)
Showed a personal display to other people to share the results of a Web search. (85,3%)
E-mailed someone a textual summary to share the results of a Web search. (60,3%)
Called someone on the phone to tell them about the results of a Web search (49%)
Printed Web pages on paper to share the results of a Web search. (41,2%)
Benjamin Riedel 6
Frequency (only those 53,4% who knew they are searching collaboratly)
Always Remember: The Study was conducted in the year 2006 – the numbers probably changed by now – a lot.
Frequency of collaborting on Web search Respondants
Daily 0,9%
Weekly 25,7%
Monthly 48,6%
Yearly 24,8%
Benjamin Riedel 7
Searching tasks people collaborate on
Task Respondants
Travel planning 27,5%
General shopping tasks 25,7%
Literature search 20,2%
Technical information 16,5%
Fact finding 16,5%
Social planning 12,8%
Medical information 6,4%
Real estate 6,4%
Benjamin Riedel 8
Types of collaboration
Brute Force: All participants just search and share their results afterwards
Divide-and-conquer:Sharing out sub-topics or places to search to everyone beforehand, so they will find different relevant results
Race:Trying to find something before everyone else does
Benjamin Riedel 9
Obstacles
No way to be sure that you are following a trail no one else of your group followed before
Difficult to share results:When communicating only per voice: URLs too long to dictate, hard to navigate people to your findings
Realizing the need to share the findings not until you can't find them anymore:Browser history not very helpful when looking for a specific site in a search
No UI that helps you to teach people how to search
Benjamin Riedel 10
Social summarization in collaborative web search (Oisín Boydell and Barry Smyth, 2009)
• Improving snippets by using context provided by community of like-minded searchers
Benjamin Riedel 11
Basic Idea
• Like-minded people will probably find similar pieces of a text interesting and therefore need similar snippets.
• So searches have to be compared to searches that were helpful for similar people in the past
Benjamin Riedel 12
Basic Idea
• Like-minded people will probably find similar pieces of a text interesting and therefore need similar snippets.
• So searches have to be compared to searches that were helpful for similar people in the past
Benjamin Riedel 14
How to achieve this?
• 1. Create surrogates (Sc) for every document (r) in a search (q) that gets a hit and save the used snippet s(r, q)
Benjamin Riedel 15
How to achieve this?
• 2. Sort and replace snippet fragments in each Surrogate, so that there is only one homogenous representation of each fragment
Therefore: Find overlapping fragments Search those for fragment pairs where the overlap is greater than a threshold
Replace the shorter fragment of the pair with the longer one
Benjamin Riedel 16
How to achieve this?
• 3. Rank the fragments by counting in how many snippets they occur
• 4. Create a social summary for a document by putting together all fragments of its surrogate, sorted by their rank
Example: If there are 3 Fragments: „eating a lot pizza“ (Rank 3), „using computers“ (Rank 5), „commenly referenced to as nerds“ (Rank 2) there social summary would by:
„...using computers … eating a lot pizza … commenly referenced to as nerds ..“
Benjamin Riedel 17
Problem: We created only a generel social summary, but we want to have one specific to each query
Answer: Weighting the rank of each fragment by the compliance between the used query and the current query
Example: fragment „eating Pizza“ can be found in the snippets for three different querys:
„what do nerds do“„what should I eat today“„how to eat pizza“By just counting the snippets the rank would have been 3Now we look at our query: „what pizza is best“:
25% of the first fragment equal our query, so we take 0.25 points for that, 0.2 for the second and 0.25 for the third. Thus the rank of „eating Pizza“ for this query is 0.25 + 0.2 + 0.25 = 0.7