slide1 n.
Download
Skip this Video
Loading SlideShow in 5 Seconds..
OVERVIEW PowerPoint Presentation
Download Presentation
OVERVIEW

Loading in 2 Seconds...

play fullscreen
1 / 1

OVERVIEW - PowerPoint PPT Presentation


  • 88 Views
  • Uploaded on

Existing algorithms for FINE-GRAINED OPINION EXTRACTION can to some extent identify and characterize private states in text when they are expressed explicitly .

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'OVERVIEW' - cree


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
slide1
Existing algorithms for FINE-GRAINED OPINION EXTRACTION can to some extent identify and characterize private states in text when they are expressed explicitly.

Similarly, existing algorithms for SEMANTIC EQUIVALENCE and NOVELTY DETECTION have focused to date on facts and event data.

OVERVIEW

HIGH-LEVEL SYSTEM ARCHITECTURE

CURRENT STATE OF THE ART

Our existing systems can:

Recognize and extract explicit private states in newswire

  • Joint opinion+source recognition: 70F
  • Opinion expressions:
    • 72F (direct expressions)

The congressman criticizedObamacare.

    • 66F (indirect expressions)

They ignored the unreasonable customer.

  • Contextual polarity given polar word:
    • 78% accuracy

Identify semantic relatedness between texts using lexical matching approaches

  • Senseval2012: 0.61 Pearson

EXPECTED IMPACT

Claire Cardie (Cornell) RadaMihalcea (UNT) JanyceWiebe (Pittsburgh)

Uncovering Motivations, Stances and Anomalies

Through Private-State Recognition and Interpretation

BloggerX: The international community seems to be tolerating the Israeli campaign against the Palestinians.

Cross-Document

Tracking of Private

States

Private-State Extraction

Private-State Database

  • WE PROPOSE TO
  • Identify the rich spectrum of private states
  • expressed not explicitly but through implicature(i.e. inference) and connotation;
  • Track private states through discourse and across documents; and
  • Produce systems for private-state-awaresemantic equivalence and novelty detection.

[explicit] Int’l community:

 Israeli campaign

[inferred] BloggerX:

 Israeli campaign

 Israel

Palestinians

  • BloggerX:
  • Palestinians
  •  Israel
  • Turkey

Private-state-aware Semantic

Equivalence and Novelty

Detection

EXAMPLES

The people are happy because Chavez has fallen.

[explicit] The people:  Chavez falling

[inferred] The people:Chavez himself

BloggerY: It is no surprise then that MoveOn would attack Senator McCain

[explicit] MoveOnSenator McCain

[inferred] BloggerY

  • Our private-state-aware semantic equivalence and novelty detection algorithms will assist analysts in:
  • Recognizing shared beliefs among key participants;
  • Determining changes in the attitudes and beliefs of key participants;
  • Detecting contradictions among expressed and inferred opinions, emotions, and attitudes; and
  • Identifying emerging or disintegrating alliances.

*attitude change*: BloggerX Palestinians

  • Cross-Document Private-State Tracking
  • Within-document and cross-document coreference resolution
  • Private-state recognition in conversational data
  • Discourse-level integration of explicit private states, connotations, and inferred private states
  • Private-State-Aware Semantic Equivalence and Novelty Detection
  • Private-state aware semantic relatedness
  • Sentence-level novelty detection
  • Novelty detection on protagonist-centered event graphs
  • Private-State Extraction
  • Representation and acquisition of connotation lexical knowledge
  • Improved recognition of explicit private states
  • Compositional calculation of polarity
  • Novel framework for representing and processing private-state implicature

 ?

MO

MO

OR

Evidence

from throughout

the discourse must be

marshaled to choose which

set of inferences is more probable

….

….

 …

 …

MO attacking SM

MO attacking SM

Contact information:

cardie@cs.cornell.edu

rada@cs.unt.edu

wiebe@cs.pitt.edu

MoveOn

MoveOn

Senator McCain

Senator McCain