Download Which are sentiment words? - LBS2016 – 13th International

November 16, 2016
Method for Construction of
Spatial Sentiment Lexicon
Using Place Reviews
Case Study on Theme Parks
Youngmin Lee
GIS/LBS LAB
Seoul National University
CONTENTS
01 Background and purpose of the study
02 How to construct spatial sentiment lexicon
03 Experiment and result
04 Conclusion
01
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Background and purpose of the study
The trend of these days
 The use of location-based services(LBS) is constantly increasing. and, reviewing and grading
places through the use of LBS has also become a common practice among users.
[ Airbnb review example ]
[ Google review example ]
 Previous reviews of a place can significantly affect their potential visitors.
3
01
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Background and purpose of the study
Visitor’s sentiment expression
 A user-created review is the result of a visitor's actual positive or negative sentiment
expression, and the sentiment could be expressed as a positive, negative, or neutral opinion.
 In order to perform sentiment analysis,
each word should be separated by its
POS (part-of-speech) through natural
language processing.
 To do this, a database of spatial
sentiment words should be constructed.
auxiliary
adverb verb
verb adjective
pronoun
Noun Noun
pronoun
verb
[ Google review example ]
4
01
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Background and purpose of the study
There is no spatial sentiment lexicon
 In the past, sentiment analysis has been used
mainly for product reviews (Chang 2009, Hu &
Liu 2004, Myung et al 2008, Scaffidi et al). No
spatial sentiment lexicon for sentiment analysis
of places has been constructed yet.
 Specifically, Korean lacks in terms of research on
sentiment word analysis, compared to English.
Korean, unlike English, has complicated
characteristics and is composed of complex
adjectives and suffixes (Jang et al. 2015).
[ Korean(Hangeul) alphabet ]
• Chang J (2009) A sentiment Analysis Algorithm for Automatic Product Reviews Classification in On-Line Shopping Mall. 14(4):19-33
• Hu M, & Liu B (2004) Mining and Summarizing Customer Reviews. In proceedings of the tenth ACM SIGKDD international conference
on Knowledge discovery and data mining. 168-177
• Myung J, Lee D, & Lee S (2008) A Korean Product Review Analysis System Using a Semi-Automatically Constructed Semantic
Dictionary. Journal of KIISE:Software and Applications. 35(6): 392-403
• Scaffidi C, Bierhoff K, Chang E, Felker M, Ng H, & Jin C (2007) Red Opal: Product-Feature Scoring from Reviews. In proceedings of
the 8th ACM conference on Electronic commerce. 182-191
• Jang K, Park S, & Kim W (2015) Automatic Construction of a Negative/positive Corpus and Emotional Classification using the Internet
Emotional Sign. Journal of KIISE. 42(4):512-521
5
01
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Background and purpose of the study
The purpose of the study
 Therefore, in this study, we propose a method to construct a spatial sentiment
lexicon using place reviews written in Korean, and we focused primarily on a ‘theme
parks’ out of the many possible place categories.
 Other types of spatial sentiment lexicon could be constructed using the same
methodology.
6
02
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
How to construct spatial sentiment lexicon
Which are sentiment words?
7
02
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
How to construct spatial sentiment lexicon
3 aspects to consider
1) The polarity of the sentiment word is analyzed.
 To do this, we calculated sentiment polarity and probability using the results of the survey.
2) It should be taken into account that some sentiment words are associated with properties of a place.
many
green
Positive
little
people
Negative
There is
3) The spatial features and predicate characteristics of a place should be classified.
 We defined the combination of the spatial feature and the predicate as the ‘spatial feature
sentence’.
8
02
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
How to construct spatial sentiment lexicon
Research workflow
Naver map search
API of Google maps
Collecting theme park POIs
Collecting place reviews
Natural language process (morphological analysis)
Extracting candidate of spatial sentiment words
Survey
Vote
Sentiment classification
Positive
Negative
Neutral
Combination
of feature
sentence
Collective
intelligence
Calculation of sentiment probability
Spatial sentiment lexicon
[ Research workflow ]
9
03
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Experiment and result
Naver map search
API of Google maps
Collecting theme park POIs
Collecting place reviews
Natural language process (morphological analysis)
Naver map searching
Extracting candidate of spatial sentiment words
Survey
Vote
Sentiment classification
Positive Negative
Neutral
Combination
of feature
sentence
Collective
intelligence
Calculation of sentiment probability
Number of
search results : 118
Spatial sentiment lexicon
Theme park POIs
Place reviews of Google maps
Collecting reviews using Google API using java script
Number of
collected reviews : 204
10
03
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Experiment and result
Naver map search
API of Google maps
Collecting theme park POIs
Collecting place reviews
Natural language process (morphological analysis)
Extracting candidates of spatial sentiment
words through NLP using R
Survey for sentiment classification
Extracting candidate of spatial sentiment words
Survey
Vote
Sentiment classification
Positive Negative
Neutral
Combination
of feature
sentence
Collective
intelligence
Calculation of sentiment probability
Candidates
of spatial
sentiment
words
Spatial sentiment lexicon
 positive
 negative
 neutral
 inappropriate
Survey
result
Sentiment polarity and probability
calculation using R
# Step 1: inappropriate
if ((count(data[inappropriate]) > (count(data[positive]) &&
count(data[inappropriate]) > (count(data[negative]) &&
count(data[inappropriate]) > (count(data[neutral])){
Score = 100 * count(data[inappropriate]) / count(data[positive]) + count(data[negative]
) + count(data[neutral]) + count(data[inappropriate])
Sentiment = ‘inappropriate’}
# Step 2: neutral_1
else if ((count(data[neutral]) > (count(data[positive]) &&
count(data[neutral]) > (count(data[negative])){
Score = 100 * count(data[neutral]) / count(data[positive]) + count(data[negative]) + co
unt(data[neutral])
Sentiment = ‘neutral’}
# Step 3: positive & negative
else if ((count(data[positive]) > count(data[negative])){
Score = 100 * count(data[positive]) / count(data[positive]) + count(data[negative])
Sentiment = ‘positive’}
else if ((count(data[negative]) > count(data[positive])){
Score = 100 * count(data[negative]) / count(data[positive]) + count(data[negative])
Sentiment = ‘negative’}
# Step 4: when positive = negative = neutral
else
Score = 100
Sentiment = ‘neutral’
11
03
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Experiment and result
The result of spatial sentiment lexicon_01
(single adjective, verb, and noun)
• PA: Adjective, PV: Verb, NC: Noun
• Positive: 1, Negative: -1, Neutral: 0
Naver map search
API of Google maps
Collecting theme park POIs
Collecting place reviews
Natural language process (morphological analysis)
Extracting candidate of spatial sentiment words
Survey
Vote
Sentiment classification
Positive Negative
Neutral
Combination
of feature
sentence
Collective
intelligence
Calculation of sentiment probability
100% positive words
Spatial sentiment lexicon
c
12
03
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Experiment and result
The result of spatial sentiment lexicon_01
(single adjective, verb, and noun)
• PA: Adjective, PV: Verb, NC: Noun
• Positive: 1, Negative: -1, Neutral: 0
Naver map search
API of Google maps
Collecting theme park POIs
Collecting place reviews
Natural language process (morphological analysis)
Extracting candidate of spatial sentiment words
Survey
Vote
Sentiment classification
Positive Negative
Neutral
Combination
of feature
sentence
Collective
intelligence
Calculation of sentiment probability
Spatial sentiment lexicon
Words determined by ‘inappropriate’
13
03
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Experiment and result
The result of spatial sentiment lexicon_02
(combination with ‘many’ and ‘little’)
• PA: Adjective, PV: Verb, NC: Noun
• Positive: 1, Negative: -1, Neutral: 0
Naver map search
API of Google maps
Collecting theme park POIs
Collecting place reviews
Natural language process (morphological analysis)
Extracting candidate of spatial sentiment words
Survey
Vote
Sentiment classification
Positive Negative
Neutral
Combination
of feature
sentence
Collective
intelligence
Calculation of sentiment probability
Spatial sentiment lexicon
14
03
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Experiment and result
The result of spatial sentiment lexicon_03
(combination of feature sentence)
• PA: Adjective, PV: Verb, NC: Noun
• Positive: 1, Negative: -1, Neutral: 0
Naver map search
API of Google maps
Collecting theme park POIs
Collecting place reviews
Natural language process (morphological analysis)
Extracting candidate of spatial sentiment words
Survey
Vote
Sentiment classification
Positive Negative
Neutral
Combination
of feature
sentence
Collective
intelligence
Calculation of sentiment probability
Spatial sentiment lexicon
15
04
Method for Construction of Spatial Sentiment Lexicon Using Place Reviews: Case Study on Theme Parks
Conclusion
Contribution and future works
 Sentiment dictionaries have been constructed for analyzing product reviews, but no sentiment
dictionary has been established for the places.
 Therefore, it is meaningful that this study constructed the sentiment lexicon for place and
calculated the sentiment polarity and probability score of by word.
 The spatial sentiment lexicon could be utilized as a reference when performing sentiment
analysis on the contents of various social media platforms, and could offer useful information
to those who want to visit a place.
 In future work, we will study a method used to
1) Extending the lexicon by adding synonyms for pre-constructed sentiment words.
2) Developing the methodology to analyze syntax more precisely.
16
THANK YOU
FOR
LISTENING
Youngmin Lee
[email protected]