tanaysh7 / Information-Extraction

Information extraction in Python using Conditional Random Fields to segment/chunk the course description of UCLA dataset by adding markup tag to indicate the type of chunk.

Information-Extraction

Information extraction in Python using Conditional Random Fields to segment/chunk the course description of UCLA data set. This project will be using markup tag to indicate the type of chunk.

Sample Input

Lecture, four hours; outside study, eight hours. Requisite: course 180. Additional requisites for each offering announced in advance by department. Selections from design, analysis, optimization, and implementation of algorithms; computational complexity and general theory of algorithms; algorithms for particular application areas. Subtitles of some current sections: Principles of Design and Analysis (280A); Distributed Algorithms (280D); Graphs and Networks (280G). May be repeated for credit with consent of instructor and topic change. Letter grading.

Here is the chunking of the above sample:

<format> Lecture, four hours; outside study, eight hours. </format> <requisite> Requisite: course 180. Additional requisites for each offering announced in advance by department. </requisite> <description> Selections from design, analysis, optimization, and implementation of algorithms; computational complexity and general theory of algorithms; algorithms for particular application areas. </description> <others> Subtitles of some current sections: Principles of Design and Analysis (280A); Distributed Algorithms (280D); Graphs and Networks (280G). <others> <others> May be repeated for credit with consent of instructor and topic change. </others> <grading> Letter grading. </grading>

Execution format:

python extract.py ucla.model test_raw test-ucla.txt

• arg1 - model
• arg2 - descriptions as lines
• arg3 - tagged file to evaluate text

Output:

• Prints Evaluation
• O/p file – test_generated.txt

Features implemented:

    'bias',         
    'word.lower=' + word.lower(),         
    'word[-3:]=' + word[-3:],         
    'word[-2:]=' + word[-2:],         
    'word.isupper=%s' % word.isupper(),         
    'word.istitle=%s' % word.istitle(),         
    'word.isdigit=%s' % word.isdigit(),         
    'postag=' + postag,         
    'postag[:2]=' + postag[:2],                  
    features.append('BOS')         
    features.append('EOS') 

Model Evaluation:

Prediction on training set :

	Precision	Recall	F1 Score
format	1.000000	1.000000	1.000000
description	0.993631	1.000000	0.996805
grading	1.000000	1.000000	1.000000
others	1.000000	0.993464	0.996721
requisite	1.000000	1.000000	1.000000

Prediction on test set :

	Precision	Recall	F1 Score
format	1.000000	1.00000	1.000000
description	0.927273	1.00000	0.962264
grading	1.000000	1.00000	1.000000
others	1.000000	0.83871	0.912281
requisite	0.973684	1.00000	0.986667