1. Jong Y. Choi, Joshua Rosen, Siddharth Maini, Marlon E. Pierce, and Geoffrey C. Fox Community Grids Laboratory Indiana University

2.  Delicious example 1 Bookmark Tags Social Networks Social Networks People- generated

3.  Collaborative Tagging  Online bookmarking with annotations  Create social networks  Utilize power of people’s knowledge  Pros and cons  High-quality classifier by using human intelligence  But lack of control or authority 2

4. 3

5. 4 Search Result SOAP, REST, … Repository Query with various options RDF RSS Atom HTML Populate Bookmarks/ tags Distributed Tagging Data CCT System Data Coordinator User Service Data Importer Collective Collaborative Tagging (CCT) System

6. 5  1 st - Service and algorithm development  Identify services and algorithms  2 nd - Interface development  Web2.o style interface  REST, SOAP, …  3 rd – Export/import service development  Merging distributed data sets  Export data to build mesh-up sites  So far, we are mainly in 1 st stage and do some experiments in 2 nd stage

7. 6 Different Data Sources Various IR algorithms Flexible Options Result Comparison

8. 7 Searching Given input tags, returning the most relevant X (X = URLs, tags, or users) Latent Semantic Indexing (LSI), FolkRank I I Recomme ndation Indirect input tags, returning undiscovered X II Clustering Community discovering. Finding a group or a community with similar interests K-Means, Deterministic Annealing Clustering III Trend detection Analysis the tagging activities in time- series manner and detect abnormality Time Series Analysis IV Service Description Algorithm Type

9.  Vector-space model (bag-of-words model)  Assume n URLs and q tags  A URL can be represented by q-dimension vector, d i = (t 1, t 2, …, t q )  A total data set can be represented by n-by-q matrix  Pairwise Dissimilarity Matrix  n-by-n symmetric matrix  Distance (Euclidean, Manhattan, … )  Angles, cosine, sine, …  O(n 2 ) complexity 8

10. 9 (Source : MSI-CIEC)  Graph model  Building a graph with nodes and edges  Edges are indicating relationship  Becoming complex networks (tag graph)  Dissimilarity  Related with path distance  Finding path is important (Shortest path problem)  Naive approach : O(n 3 ) complexity

11.  Latent Semantic Indexing  Using vector-space model, find the most similar URLs with user’s query tags  Dimension reduction from high q to low d (q >> d)  Removing noisy terms, extracting latent concepts 10 Precision Recall 2 terms 4 terms 8 terms 20% dim. reduction None Ideal Line

12.  Discover the group structures of URLs  Non-parametric learning algorithm  Non-trivial optimization problem  Should avoid local minima/maxima solution 11

13.  Deterministically avoid local minima  Tracing global solution by changing level of energy  Analogy to physical annealing process (High  Low) 12

14.  Classification  To response more quickly to user’s requests  Training data based on user’s input and answering questions based on the training results  Artificial Neural Network, Support Vector Machine,…  Trend Detection  Can be used for prediction/forecasting  Time-series analysis of tagging activities  Markov chain model, Fourier transform, … 13

15.  The goal of our Collective Collaborative Tagging (CCT) system  Utilize various data sets  Provide various information retrieval (IR) algorithms  Help to utilize people-powered knowledge  Currently various models and algorithms are being investigated  Service interfaces and import/export function will be added soon 14

16. 15

17. 16 -. Distances, cosine, … -. O(N 2 ) complexity -. Distances, cosine, … -. O(N 2 ) complexity Dis- similarity Vector-space Model -. Paths, hops, connectivity, … -. O(N 3 ) complexity -. Paths, hops, connectivity, … -. O(N 3 ) complexity Graph Model -. Latent Semantic Indexing -. Dimension reduction schemes -. PCA -. Latent Semantic Indexing -. Dimension reduction schemes -. PCA Algorithm -. PageRank, FolkRank, … -. Pairwise clustering -. MDS -. PageRank, FolkRank, … -. Pairwise clustering -. MDS -. q-dimensional vector -. q-by-n matrix -. q-dimensional vector -. q-by-n matrix Represen- tation -. G(V, E) -. V = {URL, tags, users} -. G(V, E) -. V = {URL, tags, users}

18.  Pairwise clustering  Input from vector-based model vs. graph model  How to avoid local minima/maxima? (e.g, K-Means) 17 Graph model Vector-space model