Stefan Decker Keynote at CSHALS

Editor's Notes

• #12: The Mouse; Word Processing; Data Sharing; voice & video 4 minutes
• #13: The Mouse; Word Processing; Data Sharing; voice & video 4 minutes
• #15: The Mouse; Word Processing; Data Sharing; voice & video 4 minutes
• #25: In order to create a global data network the data format needs to be able to identify entities. Therefore it needs a world-wide accepted way to identify entities. Object identity does not impose name uniqueness: information is created independently. Different object identifiers may be used for the same object as information is been created independently.
• #26: In order to create a global data network the data format needs to be able to identify entities. Therefore it needs a world-wide accepted way to identify entities. Object identity does not impose name uniqueness: information is created independently. Different object identifiers may be used for the same object as information is been created independently.
• #43: Work by Conor Hayes, Vaclav Belak et at, DERI.
• #45: Agglomerative hierachical clustering
• #46: Using principle component analysis to analyse the features, we found that the the amplitude of the largest principal component constituted more than 95% of the variance in the features, and the size of the ego-centric networks was the dominant feature in the largest component. Hence, we use the size of the ego-centric networks as our feature to partition the users into the three bands. We discard the lowest band, which consists one-post users, and the middle band, which does not have enough neighbours to have an accurate power law exponent fit. Using agglomerative hierarchical clustering, we cluster the feature profile data of the remaining top band users from all forums. To determine the optimal number of clusters, we used five different validation techniques: Rand, Silhouette, RS, Root mean square and DB Index (Handl, Knowles, and Kell 2005). We found that the optimal number of clusters was either 8, 13, 15 or 21. After manual inspection, we selected 8 and 15 as the best numbers of clusters. Each cluster approximately corresponds to one user role type. The average value of the nine features and the number of users in each cluster are used to build a quantitative description of the clusters/user role types.
• #48: For example, the taciturn role makes up 95% of all users in the Personal Issues forum (grouping 1). This suggests that, despite its name, there is little dialogue happening.
• #49: strong component of popular initiators, suggesting that a few users regularly initiate threads that subsequently generate discussion (large percentage of popular participants and supporters).
• #54: Instead of paradigm shift, we were looking for community shift Instead of paradigm articulation, we were looking for community specialization We call it ‘community shift’, because we reveal less dramatic changes in the scientific discourse. Very significant and important shifts may eventually turn out to be ‘paradigm shifts’. Similar argument applies also to the use of the notion ‘community specialization’. Paradigm articulation – scientists successfully apply the methods within the paradigm to new problems, until they eventually reach the limits of the paradigm by finding problems/questions not answerable/solvable by the methods, which leads to the crisis of the paradigm and call for a radical change in the scientific field.
• #55: Publications from major IR and SW conferences obtained from DBLP for 2000–2009 (ISWC, ESWC, SIGIR, …) Co-citation network of 5772 authors and 817642 edges over all years was extracted 3-year time-steps with 2-years overlap: 2000–2002, 2001–2003, 2002–2004, . . . Total number of articles was 39314 for which we were able to scrape 22975 abstracts and 3740 full-texts Nearly 70% coverage by content 10% coverage by author-provided keywords
• #56: We used non-overlapping community detection algorithms, because at that time there was no suitable implementation of overlapping communities detection algorithm. We currently work with overlapping communities.
• #58: In 2007 there was a strong inflow from community 15 “semantic web and IR” into community 26, which caused a change of topics towards “semantic web”.