Doozer - Revision history

Topher at 22:36, 11 January 2013

2013-01-11T22:36:48Z

Topher at 05:28, 1 November 2011

2011-11-01T05:28:02Z

Topher at 04:57, 1 November 2011

2011-11-01T04:57:09Z

Topher at 04:56, 1 November 2011

2011-11-01T04:56:09Z

Topher at 04:55, 1 November 2011

2011-11-01T04:55:38Z

Topher at 03:15, 1 November 2011

2011-11-01T03:15:44Z

Topher at 02:12, 1 November 2011

2011-11-01T02:12:26Z

Topher at 02:12, 1 November 2011

2011-11-01T02:12:05Z

Topher at 00:28, 1 November 2011

2011-11-01T00:28:15Z

Topher at 00:26, 1 November 2011

2011-11-01T00:26:46Z

← Older revision		Revision as of 22:36, 11 January 2013
Line 446:		Line 446:
	== References ==		== References ==
	<references/>		<references/>
		+
		+
		+	[[Category: Information Extraction]]
		+	[[Category: Ontology Learning]]

@@ Line 34: / Line 34: @@
 ====Models for Information Retrieval Tasks====
-The following Images show models that were created on-the-fly for Information Retrieval purposes.
+The following Images show models that were created on-the-fly for Information Retrieval purposes. The following example models were created using the Web Service interface with the following keyword queries:
 '''Example 1: Beatles Songs and related concepts'''

← Older revision		Revision as of 04:57, 1 November 2011
Line 44:		Line 44:

	'''Example 3: 2008 Financial Crisis - TARP'''		'''Example 3: 2008 Financial Crisis - TARP'''
−	[[File:tarp-all-large1.png\|1024px\|A model ~~containing~~ important concepts ~~from~~ the ~~Harry Potter Universe~~]]	+	[[File:tarp-all-large1.png\|1024px\|A model describing important concepts surrounding the TARP program]]

@@ Line 43: / Line 43: @@
 [[File:Doozer-Model-HarryPotter1.png|1024px|A model containing important concepts from the Harry Potter Universe]]
-'''Example 2: Harry Potter'''
+'''Example 3: 2008 Financial Crisis - TARP'''
 [[File:tarp-all-large1.png|1024px|A model containing important concepts from the Harry Potter Universe]]

@@ Line 40: / Line 40: @@
-'''Example 1: Harry Potter'''
+'''Example 2: Harry Potter'''
 [[File:Doozer-Model-HarryPotter1.png|1024px|A model containing important concepts from the Harry Potter Universe]]
 ==Approach==

@@ Line 17: / Line 17: @@
 An introductory video can be found here.
 <!--
@@ Line 22: / Line 23: @@
 -->
-==Example Models==
+===Example Models===
 In this section we show some models that were automatically extracted using Doozer. These range from models that were inspired by Information Retrieval and filtering tasks to models that described scientific domains. The IR-inspired models revolved around filtering of ongoing events, such as the Arab Spring in early 2011, specifically the protests in Egypt. The scientific models were built for the domains of Neoplasms and of Cognitive Performance.
-[[File:mesh-07152011e-bold-tree.png|thumb|A tree representation of a model describing the Neoplasms domain.]]
+====Cognitive Science====
 The following Images show models that were created on-the-fly for Information Retrieval purposes.
 ==Approach==

@@ Line 18: / Line 18: @@
 An introductory video can be found here.
-<!--
 ==Example Applications==
--->
 ==Example Models==
@@ Line 346: / Line 346: @@
 ===Comparison against MeSH===
-[[File:mesh-051508e-treemap-oncology.png|The Oncology subtree of the created MeSH-related category hierarchy. The size of the rectangles indicates the number of descendants. Lighter rectangles indicate categories, darker rectangles indicate individuals]]
+[[File:mesh-051508e-treemap-oncology.png|1024px|The Oncology subtree of the created MeSH-related category hierarchy. The size of the rectangles indicates the number of descendants. Lighter rectangles indicate categories, darker rectangles indicate individuals]]
 Comparing the generated list of domain terms to a Gold Standard such as MeSH allows us to get a quantitative evaluation of the extraction quality in a scientific field. However, this kind of comparison is biased towards the concepts in the gold standard. Terms in the extracted model that are relevant, but not in the gold standard will not count towards the model's precision and recall. Domain ontologies and glossaries usually contain terms for immediate domain concepts rather than terms that are highly indicative of a domain. The term "cancer", for example is very important for, but not highly indicative of the oncology field. The content of the created domain models are meant to be used in retrieval and classification tasks. For this reason, finding the phrase ''"MIT Center for Cancer Research"'' in an article is very indicative of the oncology domain, but not useful in a specific biomedical taxonomy such as MeSH. Nevertheless, in order to have a numerical evaluation of the ontology creation process, an automated Gold-Standard evaluation <ref>J. Brank, M. Grobelnik, and D. Mladenic, A survey of ontology evaluation techniques, in Proceedings of the Conference on Data Mining and Data Warehouses (SiKDD 2005), 2005, p. 166--170.</ref> is performed. We extracted all MeSH terms in the Neoplasms subtree to compare them against an automatically generated Neoplasms domain model. For the extraction, the algorithm was run with the following description:
@@ Line 405: / Line 405: @@
-[[File:growingFieldsMeSHComparison2004-2008v.png|Evaluation wrt. full set of terms and Wikipedia subset of terms in MeSH versions of 2004 and 2008.]]
+[[File:growingFieldsMeSHComparison2004-2008v.png|800px|Evaluation wrt. full set of terms and Wikipedia subset of terms in MeSH versions of 2004 and 2008.]]
 We performed the comparison with the MeSH C04 Neoplasms subtree using three different experimental setups for domain hierarchy generation; ranging from more precision-oriented to more recall-oriented. By changing the thresholds in various steps in the algorithm as seen in Table , we achieve more expansion or more reduction. Recall oriented means using a lower search threshold (more initial results), a lower expansion threshold (more similar nodes) and a low domain-importance threshold (fewer nodes deleted because of conditional probability). Precision-oriented means that higher thresholds were set. We evaluated the created domain hierarchies with respect to the MeSH versions of 2004 and 2008. We achieve a precision of up to 48 Even though the scope of the extracted hierarchy is different from the gold standard, the focus was the same. Out of a total of 222 extracted instances in the high-precision model (Experiment 3), 135 belong to the category ''types of cancer''. Other categories and terms that are relevant to the ''neoplasms'' domain can also be found, such as ''radiobiology'' and ''therapy'', which, amongst others, share the instance ''Radiation therapy'' as well as ''Chemotherapeutic agents'',  ''Tumor suppressor gene'' and  ''Carcinogens''. These additional, non-MeSH finds exemplify the difference in scope between Doozer's domain hierarchy and MeSH. Whereas the MeSH C04 subtree restricts itself to listing different types of neoplasms, Doozer discovers many related concepts that are important for classification, but do not represent types of neoplasms. We take this as a strong indication that Doozer performs well in the task of producing domain hierarchies even is such a specialized domain as the ''neoplasms'' field.
@@ Line 412: / Line 412: @@
 == References ==
 <references/>

← Older revision		Revision as of 00:26, 1 November 2011
Line 412:		Line 412:
	== References ==		== References ==
	<references/>		<references/>
		+
		+	--[[User:Topher\|Topher]] 20:26, 31 October 2011 (EDT)