SNA 3C: Applications of network centrality Lada Adamic
Hospital patient transfer network 2
simulation results 3
Identifying expertise ! The Response Time Gap 10000 41 9000 8000 7000 WAITTIME(min) 6000 5000 4000 3000 2000 96 69 1000 0 N = 39 49 high low ExpertiseRating The Expertise Gap • Difficult to infer reliability of answers • Automatically ranking expertise may be helpful.
Java Forum ! 87 sub-forums ! 1,438,053 messages ! community expertise network constructed: ! 196,191 users ! 796,270 edges
Constructing an expertise network 1 B unweighted A 1 A B C C 1 B weighted by # A 2 threads Thread 1 Thread 2 C 1/2 B weighted by A shared credit 1+1//2 Thread 1 : Large Data, binary search or hashtable? user A C Re: Large ... user B Re: Large... user C Thread 2 : Binary file with ASCII data user A Re: File with ... user C 0.9 B weighted with 0.1 backflow A C
Uneven participation 0 ! � answer 10 people � may 2 α = 1.87 fit, R = 0.9730 reply to thousands of -1 cumulative probability 10 others number of -2 ! � question people one 10 replied to people � are also uneven number of people in the -3 10 one received number of replies from repliers to their posts, -4 10 but to a 0 1 2 3 10 10 10 10 lesser extent degree (k)
Not Everyone Asks/Replies The Web is a bow tie The Java Forum network is an uneven bow tie • Core: A strongly connected component, in which everyone asks and answers • IN: Mostly askers. • OUT: Mostly Helpers
fragment of the Java Forum
Relating network structure to expertise ! Human-rated expertise levels ! 2 raters ! 135 JavaForum users with >= 10 posts ! inter-rater agreement ( τ = 0.74, ρ = 0.83) ! for evaluation of algorithms, omit users where raters disagreed by more than 1 level ( τ = 0.80, ρ = 0.83) L Category Description 5 Top Java expert advanced topics deeply. Knows the core Java theory and related 4 Java professional Can answer all or most of Java concept very well, questions. Also knows one or some sub topics 3 Java user relatively well. Knows advanced Java concepts. Can program 2 Java learner not good at advanced topics of Java. Knows basic concepts and can program, but is 1 Newbie Just starting to learn java.
Algorithm Rankings vs. Human Ratings 0.9 0.8 0.7 0.6 # answers 0.5 z-score # answers indegree 0.4 z-score indegree PageRank 0.3 HITS authority 0.2 0.1 0 Top K Kendall � s τ$ Spearman � s ρ$ simple local measures do as well (and better) than measures incorporating the wider network topology
automated vs. human ratings 140 160 140 120 120 100 101 100 101 80 80 60 60 40 40 RANK of INDGR RANK of REPLY 20 20 automated ranking 40 40 # answers 0 indegree 0 -20 -20 N = 2 19 20 19 18 17 16 12 10 N = 2 19 20 19 18 17 16 12 10 2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10 160 140 LEVCOM LEVCOM 140 120 120 100 104 106 100 80 80 1 101 60 60 RANK of ZTHREADS 40 40 RANK of ZDGR 20 20 40 z # answers z indegree 0 0 -20 -20 N = 2 19 20 19 18 17 16 12 10 N = 2 19 17 19 17 17 11 12 10 2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10 160 140 LEVCOM LEVCOM 140 120 120 100 68 100 1 33 80 80 60 60 81 92 40 RANK of HITS_AUT 40 RANK of PRANK 20 20 PageRank 5 0 HITS authority 0 -20 -20 N = 2 19 20 19 18 17 16 12 10 2 3 4 5 6 7 8 9 10 N = 2 19 17 19 17 17 11 12 10 2 3 4 5 6 7 8 9 10 LEVCOM human rating LEVCOM
Modeling expertise network formation Control Parameters: ! Distribution of expertise ! Who asks questions most often? ! Who answers questions most often? ! best expert most likely ! someone a bit more expert ExpertiseNet Simulator
Simulating probability of expertise pairing suppose: expertise is uniformly distributed probability of posing a question is inversely proportional to expertise p ij = probability a user with expertise j replies to a user with expertise i 2 models: � best � preferred � just better � preferred 5 5 4 4 3 3 2 2 1 1 0 0 0 1 2 3 4 5 0 1 2 3 4 5 ( − j i ) ( − i j ) p ~ e / i p ~ e / i β γ j>i ij ij
Visualization Best � preferred � just better
Degree correlation profiles asker indegree Java Forum Network asker indegree asker indegree best preferred (simulation) just better (simulation)
Algorithm selection Preferred Helper: ‘ best available ’ Preferred Helper: ‘ just better ’
Algorithm evaluation In the � just better � model, a node is correctly ranked by PageRank but not by HITS
Recommend
More recommend
