jcdl4: Social Networks

[deborah]

Social Networks

Can social bookmarking enhance search in the web? (Yanbe, Jatowt, Nakamura, and Tanaka)

This was a fascinating paper.

How do you measure the popularity of a web page? Linkage is no longer a good metric for popularity, because of auto links from wikis and blogs, as well as spam, and links with different purposes. Can we trust Google PageRank, which is the common metric?

They developed SBRank: social bookmark based on popularity, and a search engine to use SBRank, called SBSearch. SBSearch uses tagging and other social bookmarking systems

• Content vs Sentiment tags: search mechanism allows for linking because "useful", "amazing", or "awful" -- so doesn't high rank pages linked for their sheer horribleness


• temporal factor allowed in search
  
  • age of page in social bookmarking system
  
  
  • popularity variance over time


• estimates controversy ranking of pages

Then they did a data analsysis of social bookmarks based on their ranking system, to clarify characteristics

• what kinds of bookmarks exist


• what kind of relations between ranking meaures

They created a dataset using del.icio.us popular tags, to find popular pages, and then added google PageRank to the metric. This analyses how pagerank compares to the others, as well.

• more than half the pages found have a pagerank of 0 on a 0-10 scale!


• pages bookmarked for the first time long ago have a higher pagerank value


• there is a positive correlation between SBRank and PageRank


• exposes user determined popularity of pages

Example: search "digital library" [rank in SBRank (rank in PageRank)]
1 (77): internet archive
2 (19): citeseer
3 (2): online books page

ToDo:

• combat spam links


• meta-search

Task-based interaction with an integrated multilingual, multimedia information system: a formative evaluation (Zhang, Plettenberg, Klavans, Oard, and Soergel)

This paper, on speech recognition and machine translation to transform broadcast news to English: how do information analysts cope with MT errors?, was not that interesting to me. It might have been a good paper but MT is not my field.

modeling personal and social network context for even annotation in images (Shevade, Sundaram, and Xia)

These folks wanted to build a framework to annotate images and design recommendations based on user annotations. They used what they called a "who, when, where, what context".

So pick an event: eg a birthday party. The context is that it has attributes (eg participants, date, location, activity). Can you use correlation to build networks among users?

• compare similarity among who,when,where,what facets


• compare images themselves with image comparison algortihms


• use relational graph to fileter recommendations


• showed that people within a social netowrk agree about images more often than users with no relation

longitudinal study of changes in blogs (Bogen II, Francisco-Revilla, Furuta, Hubbard, Karadkar, and Shipman)

They wanted to measure changes in blogs. With 62 blogs (from 100 technorati top, minus non-english, minus dead links, minus not blogs), they cached blogs every day and measured

• absolute changes from base date


• change deltas

measurement revealed that

• the basic blog template doesn't change much; mostly just the entries.


• people have more time to blog on weekends

This is very preliminary research. still to study: should they separate by days of week; by blog type; by content vs template; by post vs comments; etc.