Semantic Grounding Joshua Shinavier 20089011

From Semantic Portal Wiki

Joshua Shinavier presented cattuto2008semantic at CSCI 6966 Advanced Semantic Web (Fall 2008) - Lesson 3

Contents 1 Presentation slides 2 Audience questions 3 Questions 4 Audience notes 5 Attendees

Presentation slides

• File:ASW-Joshua Shinavier-20080911.ppt

Audience questions

• Speaker: Joshua Shinavier
• Title: Semantic Grounding of Tag Relatedness in Social Bookmarking Systems
• Authors: Ciro Cattuto, Dominik Benz, Andreas Hotho, Gerd Stumme
• Conference: ISWC 2008
• URL: http://arxiv.org/pdf/0805.2045
• Date of Presentation: 2008/09/11

Questions

ID	Question	Name	Answer
Q1 for Josh on Semantic Grounding	Basic question. What is the form of random surfer vector? Is it in reference (13)? And also, how to define important tags and important users within the context of FolkRank? Do we have to take social metrics into considerations when saying that a specific user is more important(to some people) than other users? Web of trust has any role here	Zhenning Shangguan
Q2 for Josh on Semantic Grounding	IMHO, this paper describes roughly two categories of approaches to evaluate tag-relatedness -one qualitative and one semantic. Is it possible to combile these two kinds of approaches so that they can complement each other? E.g., when we are evaluating the relatedness between tags written in Chinese and Indian, is simply using semantic grounding enough to reach a sound relatedness between them? If not, do other qualitative metrics mentioned in this paper have any merit?	Zhenning Shangguan
Question 2	Cosine Similarity: Tags are represented by Vectors, but what is the vector form (the elements that form the vector, (user, resource)?)?	Ankesh Khandelwal
Semantic Grounding Joshua Shinavier 20089011 Jesse Weaver	What is the significance of Figure 1? Why is it significant that "the tags obtained from cosine-similarity relatedness belong to a broader class of tags, not strongly correlated with rank (frequency)"? Does this somehow indicate that that the tags are themselves more strongly related? Or is this just to further the point that cosine-similarity has significantly different related tags from the frequency and FolkRank methods?	Jesse Weaver
SemanticGroundingSep11GregoryToddWilliams1	Do the presented similarity measures approximate the actual (grounded) similarities? The paper starts by asserting that "folksonomies include many community-specific terms", and proceeds to ground tags in WordNet (apparently ignoring everything that isn't in WordNet for the relatedness assessment). Is there any indication that such an approach will actually provide accurate results when applied to all tags, even "community-specific terms"?	Gregory Todd Williams

	Author	Text
Cattuto2008semantic question 1 by lebo	Tim Lebo
Cattuto2008semantic question 2 by lebo	Tim Lebo	The introduction motivates the investigation of relatedness measures: "We believe that a deeper insight into the semantic properties of relatedness measures is an important prerequisite for the design of ontology learning procedures..." Although relatedness measures may be necessary for a "Ontology Learning" capability, they are arguably insufficient on their own. What other components are required to achieve Ontology Learning? Is this work already being done?
Cattuto2008semantic question 3 by lebo	Tim Lebo	The paper compares five relatedness metrics. We could sit down and make up ten more. What does a "good" metric look like? How do we avoid a subjective evaluation of a metric's results? (e.g.,"An interesting observation is also that java and python could be considered as siblings in some suitable concept hierarchy") If we had the perfect metric (m*), what would we do with it?
Cattuto2008semantic question 4 by lebo	Tim Lebo	Could you explain the difference between one-mode, two-mode, and three-mode analysis? What challenges are introduced when attempting three-mode analysis?
Cattuto2008semantic question 5 by lebo	Tim Lebo	Notation nit: <math>R^T</math> and <math>R^n</math> seem to be used synonymously -- what is the distinction? Shouldn't <math>R^T</math> be <math>R^
Cattuto2008semantic question 6 by lebo	Tim Lebo	I don't understand the statement "The reason for giving weight zero between a node and itself is that we want two tags to be considered related when they occur in a similar context, and not when they occur together. What is the difference between "similar context" and "occur together"? I would think that they are the same. Isn't "occurrence" the only "context" that the Folksonomy formalism provides?
Cattuto2008semantic question 7 by lebo	Tim Lebo	The description of FolkRank mentions "random surfer vector" but does not introduce the term or it's purpose. Could you describe the random surfer vector?

Audience notes

Attendees

Tim Lebo