comparable entity mining from comparative questions (1)

7/21/2019 Comparable Entity Mining From Comparative Questions (1)

1/5

Comparable Entity Mining from Comparative Questions

Abstract:

Comparing one thing with another is a typical part of human decision making

process. However, it is not always easy to know what to compare and what are the

alternatives. To address this difficulty, we present a novel way to automatically

mine comparable entities from comparative questions that users posted online.To

ensure high precision and high recall, we develop a weakly-supervised

bootstrapping method for comparative question identification and comparableentity extraction by leveraging a large online question archive. The experimental

results show our method achieves !-measure of "#.$% in comparative question

identification and "&.&% in comparable entity extraction. 'oth significantly

outperform an existing state-of-the-art method.

Architecture Diagram:


2/5

Objectives

(e develop a weakly-supervised bootstrapping method for

comparative question identification and comparable entity extraction by leveraging

a large online question archive. The experimental results show our methodachieves !-measure of "#.$% in comparative question identification and "&.&% in

comparable entity extraction. 'oth significantly outperform an existing state-of-

the-art method.

Existing system:

comparator mining is related to the research on entity and relation extraction in

information extraction )pecifically, the most relevant work is mining comparative

sentences and relations. Their methods applied class sequential rules *C)+ and

label sequential rules *)+ learned from annotated corpora to identify

comparative sentences and extract comparative relations respectively in the news

and review domains. The same techniques can be applied to comparative question

identification and comparator mining from questions.

Disadvantages:

This methods typically can achieve high precision but suffer from low recall.

Proposed system:

we present a novel weakly supervised method to identify comparative questions

and extract comparator pairs simultaneously. (e rely on the key insight that a good

comparative question identification pattern should extract good comparators, and a

good comparator pair should occur in good comparative questions to bootstrap the


3/5

extraction and identification process. 'y leveraging large amount of unlabeled data

and the bootstrapping process with slight supervision to determine four parameters.

Advantages:

To ensure high precision and high recall, we develop a weakly-supervised

bootstrapping method for comparative question identification and comparable

entity extraction by leveraging a large online question archive

cope of t!e Project

(e rely on the key insight that a good comparative questionidentification pattern should extract good comparators, and a good comparator pair

should occur in good comparative questions to bootstrap the extraction and

identification process.

Main Modules:

Pattern "eneration#comparable Entity$:

%& 'exical patterns

#. "enerali(ed patterns&. peciali(ed patterns

Pattern Evaluation#comparable )uestions$:

'exical patterns:

exical patterns indicate sequential patterns consisting of only words and symbols

*C, /start, and /end. They are generated by suffix tree algorithm with two

constraints0 1 pattern should contain more than one C, and its frequency in

collection should be more than an empirically determined number.

"enerali(ed patterns:


4/5

1 lexical pattern can be too specific. Thus, we generali2e lexical patterns by

replacing one or more words with their 34) tags. 12 generali2ed patterns can

be produced from a lexical pattern containingN words excluding $Cs.

peciali(ed patterns:

5n some cases, a pattern can be too general. or example, although a question

ipod or zune? is comparative, the pattern


5/5

HARDWARE REQUIREMENTS:

PROCESSOR : PENTIUM IV 2.6 GHz, Intel Core 2 Duo.

RM : 2 G! DD RM

MONITOR : 1"# CO$OR

HRD DIS% : &' G!

CDDRIVE : $G "2(

%E)!ORD : STNDRD 1'2 %E)S

MOUSE : * !UTTONS

SOFTWARE REQUIREMENTS:

O+ert-n /0/te : 3-n4o5/ '7 (P Pro8e//-onl

9ront En4 : V-/ul Stu4-o 2'12, C.Net.

Dt;/e : S

comparable entity mining from comparative questions (1)

Documents