Difference between revisions of "Comparing Unsupervised Learning of Narrative Event Chains and Mining the web for fine-grained semantic verb relations"

From Cohen Courses
Jump to navigationJump to search
 
(25 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
== Papers ==
 
== Papers ==
1.Unsupervised Learning of Narrative Event Chains, by N. Chambers, D. Jurafsky. [http://malt.ml.cmu.edu/mw/index.php/Chambers_and_Jurafsky,_Unsupervised_Learning_of_Narrative_Event_Chains,_ACL_2008]
+
1. Unsupervised Learning of Narrative Event Chains, by N. Chambers, D. Jurafsky. [http://malt.ml.cmu.edu/mw/index.php/Chambers_and_Jurafsky,_Unsupervised_Learning_of_Narrative_Event_Chains,_ACL_2008]
  
2. VerbOcean:Mining the Web for Fine-Grained Semantic Verb Relations, by Timothy Chklovski, Patrick Pantel. http://malt.ml.cmu.edu/mw/index.php/Chklovski_and_Pantel_(2004)_Verbocean:Mining_the_web_for_fine-grained_semantic_verb_relations
+
2. VerbOcean:Mining the Web for Fine-Grained Semantic Verb Relations, by T. Chklovski, P. Pantel. [http://malt.ml.cmu.edu/mw/index.php/Chklovski_and_Pantel_(2004)_Verbocean:Mining_the_web_for_fine-grained_semantic_verb_relations]
].
 
  
== Problem ==
+
== Problem, method and dataset ==
Although both papers study the popularity dynamics of user-generated videos, the focus is quite different. Paper 1 tries to characterize the evolution of YouTube video popularity in a general sense, while Paper 2 focuses on the content-agnostic factors only, and deliberately removes the impact that might be caused by video content. In paper 1, a three-phase evolution model is proposed to explain the popularity dynamics, while paper 2 does not have such characterization.
+
Goal of first paper was to [[AddressesProblem::extract narrative event chains using unsupervised methods]] . Whereas, goal of second paper is to [[AddressesProblem::find the relation between verbs]]. Relations considered in second paper are similarity, strength, antonymy, enablement, and temporal relations.
  
==Method==
+
Paper 2 uses approach motivated by mutual information to compute strength of association/relation between verb pairs and surface pattern. Paper 1 uses mutual information motivated approach to induce narrative events. It uses [[UsesMethod::Support Vector Machines]] for temporal ordering of connected events.
For characterizing the popularity evolution of videos, Paper 1 proposes a time-to-peak distribution based on empirical analysis, and uses a three-phase (before, at and after peak) model to account for temporal varying factors that impact popularity, while Paper 2 uses PCA, correlation analysis techniques, and a multi-linear regression model for assessing the importance of different content-agnostic factors.
 
  
== Dataset==
+
Paper 1 uses [[UsesDataset::Gigaword corpus]] and  [[UsesDataset::TimeBank Corpus]]. Whereas, in Paper 2 Google is used to find number of hits for relation verb pair <math> V1,V2 </math> and surface pattern <math> P </math> from WWW. Similarly other hits are also found using Google.
Paper 1 uses a dataset tracking the views of recently uploaded YouTube videos over the duration of eight months. Paper 2 uses clone sets of videos to explicitly remove the content factors that might impact the popularity.
 
  
==Other==
+
Both papers address different problems. Though approaches inspired my mutual information are used in some way in both papers.
Both papers were written by the same authors, and Paper 2 is a follow-up work of Paper 1 for specifically focusing on the content-agnostic factors.  
 
  
 
==Additional Questions==
 
==Additional Questions==
1. How much time did you spend reading the (new, non-wikified) paper you summarized? 1 hour
+
1. How much time did you spend reading the (new, non-wikified) paper you summarized? 2 hour
  
2. How much time did you spend reading the old wikified paper? 1.5 hours (It's a much longer paper)
+
2. How much time did you spend reading the old wikified paper? 1 hours  
  
3. How much time did you spend reading the summary of the old paper? 15 minutes
+
3. How much time did you spend reading the summary of the old paper? 10-15 minutes
  
4. How much time did you spend reading background materiel? 30 minutes
+
4. How much time did you spend reading background materiel? 10 mins.
  
5. Was there a study plan for the old paper? No
+
5. Was there a study plan for the old paper? No.

Latest revision as of 12:09, 6 November 2012

Papers

1. Unsupervised Learning of Narrative Event Chains, by N. Chambers, D. Jurafsky. [1]

2. VerbOcean:Mining the Web for Fine-Grained Semantic Verb Relations, by T. Chklovski, P. Pantel. [2]

Problem, method and dataset

Goal of first paper was to extract narrative event chains using unsupervised methods . Whereas, goal of second paper is to find the relation between verbs. Relations considered in second paper are similarity, strength, antonymy, enablement, and temporal relations.

Paper 2 uses approach motivated by mutual information to compute strength of association/relation between verb pairs and surface pattern. Paper 1 uses mutual information motivated approach to induce narrative events. It uses Support Vector Machines for temporal ordering of connected events.

Paper 1 uses Gigaword corpus and TimeBank Corpus. Whereas, in Paper 2 Google is used to find number of hits for relation verb pair and surface pattern from WWW. Similarly other hits are also found using Google.

Both papers address different problems. Though approaches inspired my mutual information are used in some way in both papers.

Additional Questions

1. How much time did you spend reading the (new, non-wikified) paper you summarized? 2 hour

2. How much time did you spend reading the old wikified paper? 1 hours

3. How much time did you spend reading the summary of the old paper? 10-15 minutes

4. How much time did you spend reading background materiel? 10 mins.

5. Was there a study plan for the old paper? No.