International Journal of Artificial Intelligence and Knowledge Discovery Vol.

3, Issue 2, April, 2013

A Sentimental Analysis of Movie Reviews Involving Fuzzy Rule-

Based
Amit P. Pimpalkar Prof. Mahendra Kumar Rai
Department of Computer Science (Software Systems) Department of Information Technology
Shri Ram Institute of Technology, Jabalpur Shri Ram Institute of Technology, Jabalpur
[email protected] [email protected]

Abstract- With the express development of online sentiment of their person behind. Currently Google searches
resources, discussion forums, groups and blogs; people for facts and facts can be expressed with keywords; but
communicate through these means of internet on daily basis. Google does not explore for opinions, since opinions are
Today, enormous amounts of subjective text are available on hard to articulate with keywords.
the internet. Business forecasters are revolving their eyes on
the internet in order to obtain realistic and subjective
Sentiment Analysis (SA) is a "Web Opinion Mining",
information (opinions) for their companies and products.
Opinion mining can aid in a number of potential applications where the primary objective is to classify the opinions
in areas such as search engines, market research and according to a variety and range. The boundaries on the
recommender systems. range usually correspond to +ve or -ve feelings about a
Sentiment Analysis (SA) is extensive contribution of product or brand which in fact determines sentiment
Natural Language Processing (NLP) which covenant with the orientation of an individual or a group of people. Current
computational measures of opinion, sentiment, subjectivity ranking strategies are not appropriate for opinion mining.
and objectivity in the given text. SA is the process of extracting Primarily the research has mostly persistent on the
intellectual capacity from the people’s judgment, assessment classification of the text data. Sentiments are naturally
and emotions toward an entities, an event and their attributes.
subjective from individual to individual, and can be
These opinions significantly make impact on consumers to
take their preference regarding shopping, choosing products absolute illogical. It's critical to analyze relevant sample of
and entities. As a result, it is desired to develop an efficient data when attempting to measure sentiment. No particular
and effective SA system for customer reviews and comments. data point is necessarily relevant. An individual's sentiment
We consider the quandary of determining the polarity of toward a brand or a product may be inclined by one or more
sentiments in reviews when negation words occur in the mind and someone might have a terrible day and tweet a -
sentences. ve remark about something they otherwise had a pretty
This project uses SentiWordNet to assigns sentiment neutral opinion.
scores to each word found in comments. Sentiments of the With large enough samples, outliers are diluted in the
words are assigned three sentiment scores: Positivity,
aggregate. Also since sentiment are very likely to changes
Negativity and Objectivity with a word which lies in between
the range from 0 to 1. The project also uses Rule-Based Fuzzy over the time according to a person's mood, world events,
measures Approach and gives the output. and so forth, it's usually important to look at data from the
perspective of time. Like any other type of Natural
Keywords- SentiWordNet, Natural Language Processing, Language Processing (NLP) analysis, context matters, it's
Machine Learning, Movie Review, Sentiment Analysis System, an incredibly difficult issue, sarcasm and other types of
Web Opinion Mining, Fuzzy measures, Text Tokenization. ironic language are inherently problematic for machines to
detect when looked at in isolation. It's imperative to have a
I. INTRODUCTION sufficiently sophisticated and rigorous enough approach
that relevant context can be taken into account. That would
Today the Internet holds an enormous amount of require knowing a particular person is ironic, exaggerated
textual data, which is also growing every day. The text is in or sarcastic which offers evidence to conclude whether or
ubiquitous format on the web, since it is easy to generate not a phrase is ironic.
and publish. People communicate through online resources, The focus of the system is to analyze the sentiments for
discussion forums, groups and blogs. What is hard now-a- the movie reviews. The input is to be taken from the movie
days is not accessibility of valuable information but rather review sites and the social networking sites on which the
extracting it in the appropriate context from the vast ocean comments are posted for the particular movie. The SA is
of text content. It is now beyond human supremacy and done by three methods i.e. Rule-Based approach, Machine-
time to see through it manually and therefore, the research Learning and the Hybrid Approach of the two. Rule-Based
problem of automatic categorization and organizing text is is the manually created Rules approach. Machine-Learning
perceptible. Textual information on web can be separated give the more efficient approach as compared to the Rule-
into two main fields: facts and opinions. While facts focus Based approach. The Hybrid Approach of both above
on objective data transmission, the opinions express the

Print-ISSN: 2231-2021 e-ISSN: 2231-0312

© RG Education Society (INDIA)
 International Journal of Artificial Intelligence and Knowledge Discovery Vol.3, Issue 2, April, 2013

approaches will give the more superior and filtered outcome helps to attain the best accuracy so far in this field. B. Pang
of the analysis for the movie reviews. [12] gives the idea about the Support Vector Machines
technique related to sentiments analysis. It also includes the
II. LITERATURE REVIEW other technique like Naive Bayes, Maximum Entropy.

C. Hauff [5] gives the way how to handle the negation III. PROBLEM DEFINITION
words like not, no, neither, couldn’t, etc words in the
sentence. It may happen that even if the negative words are • Named Entity Recognition - What is the person actually
present in the sentence still its meaning is in a positive way. talking about, e.g. Is 300 Spartans is a group of Greeks or
A. Neviarouskaya [7] performs fine grained categorization a movie?
of sentences using ten categories: nine emotions (‘Anger’, • Anaphora Resolution - The problem of resolving what a
‘Disgust’, ‘Fear’, ‘Guilt’, ‘Interest’,’ Joy’, ‘Sadness’ pronoun, or a noun phrase refers to. "We watched the
(‘distress’), ‘Shame’ and ‘Surprise’) and neutral. The movie and went to dinner; It was terrible." What does "It"
proposed rule-based approach processes each sentence in refer to?
stages, including symbolic cue processing, detection and • Parsing - What is the subject and object of the sentence,
transformation of abbreviations, sentence parsing and which one does the verb and/or adjective actually refer
word/phrase/sentence-level analyses. Each analyzed to?
sentence is automatically annotated with emotion or neutral • Sarcasm - If you don't know the source author you have
label and numerical intensity; the strength of emotion. It no idea whether 'bad' means bad or good.
also mentioned the links where the datasets would be • Ambiguous or more complex situations - The hotel room
available. B. Tierney [8] presents the results of applying the is on the ground floor right by the reception”. Is that
SentiWordNet lexical resource to the problem of automatic positive, negative or its neutral? The answer is probably
sentiment classification of film reviews. It comprises that it is different to different people.
counting +ve and -ve term scores to determine sentiment • Scale - What is the quantity of data input as a proportion
orientation, and an improvement is presented by building a of the total universe of users? 10% of the Twitter gives
data set of relevant features using SentiWordNet as source, you a rough idea of what's going on, but the results are
and applied to a machine learning classifier. M. Thelwall nowhere close to the resolution if you get with 50% of
[9] gives the hybrid knowledge for involving the Rule- the reviews.
Based and Support Vector Machines method. The hybrid
approach gives the maximum efficiency for analysis of IV. PROPOSED WORK
sentiments.
S. Kawathekar [1] studies the role of negation in an The development of a complete review or opinion
opinion-oriented information-seeking system. They mining application might involve attacking each of the
investigate the problem of determining the polarity of following problems. If the application is integrated into a
sentiments in movie reviews when negation words, such as general-purpose search engine, then one would need to
not and hardly occur in the sentences. A. Shukla [3] determine whether the user is in fact looking for subjective
presented a tool which tells the quality of document or its material. This may or may not be a difficult problem in and
usefulness based on annotations. Annotation may include of it: perhaps queries of this type will tend to contain
comments, notes, observation, highlights, underline, indicator terms like “review”, “reviews”, or “opinions”, or
explanation, question or help etc. Collective sentiments of perhaps the application would provide a “checkbox” to the
annotators are classified as positive, negative, objectivity. user so that he or she could indicate directly that reviews
Recent research has tried to automatically determine are what are desired. Besides the still-open problem of
the “PN-polarity” of subjective terms F. Sebastiani [10] in determining which documents are topically significant to an
order to aid the extraction of opinions from text, i.e. opinion-oriented query, a supplementary challenge we face
identify whether a term that is a marker of opinionated in our new setting is simultaneously or subsequently
content has a positive or a negative connotation. P. determining which documents or portions of documents
Bhattacharyya [11] proposed a technique for the effective contain review-like or opinionated material. Sometimes this
SA of movie reviews. It also describe a novel approach to is relatively easy, as in texts fetched from review-
process the predictions for individual documents of the test aggregation sites in which review-oriented information is
dataset to improve the accuracy over the entire set. It presented in relatively tagged format: examples include
presented a WorldNet based method for the effective Epinions.com and Amazon.com. However, blogs also
incorporation of linguistic information in the system disreputably contain quite a bit of subjective content and
without any kind of experts’ intervention. It also presents a thus are another obvious place to look and are more
generic method that can be used to improve the accuracy of relevant than shopping sites for queries that concern
classification over a test dataset in any kind of classification politics, people, or other non-products, but the desired
task. It shows how the application of this technique to SA

10
 International Journal of Artificial Intelligence and Knowledge Discovery Vol.3, Issue 2, April, 2013

material within blogs can vary quite widely in content, • It might be more appropriate to construct an idea of
style, presentation, and even level of grammaticality. sentiment data rather than a textual summing up of it,
Once one has target documents in hand, one is still whereas textual summaries are what are usually created in
faced with the problem of identifying the overall sentiment standard topic-based multi-document summarization.
expressed by these documents and/or the accurate opinions
regarding particular features or aspects of the items or PHASES OF SENTIMENT ANALYSIS SYSTEM
topics in question. Again, while some sites make this kind (SAS)
of extraction easier for e.g., user reviews posted to Yahoo!
Movies must specify grades for predefined sets of i. Data Collection: In this step, it is important to get
distinctiveness of films more free-form text can be much the item of interest, that is, what we actually want to know
harder for computers to analyze, and indeed can pose the opinion about. It is also important to remove all facts
additional challenges; for e.g., if quotations are included in that don't express opinions like news and objective phrases.
a newspaper article, care must be taken to attribute the The focus is on the user's opinions.
views expressed in each quotation to the correct entity. ii. Pre-processing: The pre-processing is also
important in order to remove unnecessary words or
irrelevant words from the user's opinions. Our processing
system deals only the description part of each review, here
processing means splitting review into sentences to create a
plain text file of reviews.
iii. Classification and part of Speech Tagging
(POS): The polarity of the content that must be identified.
Generally, the polarities used are positive, negative or
neutral. POS is a process whereby tokens are sequentially
labeled with syntactic labels, such as “Verb" or "Noun" or
"Adjective". Not all the words in review sentences are
sensibly useful for recognize product features and
orientations of the discussed product. As Hu and Liu
recommend, nouns and noun phrases in the sentences are
most likely to be the features that customers comment on,
while adjectives are frequently used to express their
opinions and feelings.
We collect frequent nouns and adjectives, adverbs from
the review file after part-of-speech (POS) tagging using a
SENTIWORDNET. POS taggers are not constantly ideal
and review sentences might be pretty complex or irregular.
Therefore, tagging errors cannot be avoided. A
SENTIWORDNET triplet of numerical scores Φ(s, p) (for p
∊ P = {P,N,O}) describing how strongly the terms
contained in s enjoy each of the three properties.
Figure 1: Sentiment classification system design
∑ {P,N,O} =1
Finally, the system needs to present the sentiment
iv. Stop Words Removal: We remove words like
information it has achieve in some reasonable summary
digits, prepositions, articles, and proper nouns like name of
fashion. This can involve some or all of the following
Movie etc from the POS tagged review file, as their
actions:
existstance are pointless in our system. It helps better
extraction of opinion phrases/words from the POS tagged
• Aggregation of “votes” that may be registered on file.
different scales (e.g. one reviewer uses a grade (good, v. Summarization of Results: In this step, the
bad, average), but another uses star system (1, 2, 3….)) categorizations of several opinions were summarized in
• Selective standing of some opinions. order to be presented to the user. The objective is to
• Representation of points of disparity and points of facilitate the perceptive and give a general grasp about what
harmony. people are talking about an item. This summarization can
• Classification of communities of opinion holders. be articulated in graphics or text.
• Accounting for diverse levels of influences among
opinion holders.

11
 International Journal of Artificial Intelligence and Knowledge Discovery Vol.3, Issue 2, April, 2013

number of annotations present on the document. In second

protocol, it computes collective sentiment of annotators.

Algorithm 1 Find the total count of annotation found in the

document.
Input: Annotated document
Output: Total Count of Annotation present in the
Document
1. for each annotated document do

Count   Annotation

2. End for.
3. Output Count.
Figure 2: The graphical representation adopted by
SENTIWORDNET for representing the opinion-related Algorithm 2 Find the average score of each annotation
properties of a synset found in document
Input: List of sentiment words extracted from comments of
annotation
Output: Sentiment score and Sentiment Review

1. for each sentiment word from List do

Get polarity as well as sentiment scores using
SentiWordNet.
2. Compute the intensity of each word using Fuzzy
Intensity Finder algorithm.
3. Final Sentiment Score (SS)

MaxPolarity
SS  
Count

4. Output Sentiment Score.
5. Output Overall Sentiment Review.

ALGORITHM FUZZY INTENSITY FINDER

Figure 3: Sentiment Analysis System (SAS)
We calculate the weight of the extracted opinionated
vi. Fuzzy Rule-Based Approach: In the Rule-Based
phrases as the weights of individual words in the phrases.
approach, rules are to be defined which contain an if-else
Case 1: There are few adverbs like very, really,
relation which have originator and its associated resultant.
extremely, simply, always, never, not, absolutely, highly,
In this methodology, certain rules are to be organized and
overall, truly, too, etc. (we imply 15 such adverbs) which
then the sentiments should be analyzed depending on rules.
may be used positively or negatively like {very good, very
A consequent represents a sentiment that is either positive
bad}, {extremely acceptable, extremely unacceptable}, {too
or negative, and is the result of meeting the condition
good, too bad}, {simply outstanding, simply disgusting}
defined by the originator. {token-1 ^ token-2 ^ . . . ^ token-
etc.
n} =) {+|−}. The two simple rules listed below depend
solely on two sentiment bearing words, each of which
Case 2: Again never, not etc. will change the
represents an antecedent. {excellent} = {+}; {absurd} =
orientation of the opinion like {not good will be bad}
{−}.
{never accepted will be always rejected} etc.
OPINION ANLAYZER
Case 3: Case 1 and Case 2 may also come together in
opinion phrases like {not very good}, {not absolutely
Opinion analyzes annotations to obtain the collective
recommended}, {not truly reliable one} etc. We consider
sentiments of annotators in two protocols to judge the
them accordingly in the calculation of fuzzy weights of
quality or usefulness of documents. In first protocol, it
opinion phrases.
extracts all the annotations created by annotators which are
of type highlights, underline, and comments and count total

12
 International Journal of Artificial Intelligence and Knowledge Discovery Vol.3, Issue 2, April, 2013

Case 4: It is also possible that in some cases like {It is Bad" 0.0625 Pathetic " 0.1406
not only good but also it’s an awesome} etc. appears in This is not only bad but it’s pathetic 0.0206
sentiments. Table 4: Fuzzy Measure of (But, Also, Nor) Phrases

In Case 1, we consider the weight (W) of the opinion V. RESULT

like (very /extremely/highly etc) (Adj)
 Value Of Adj if ValueOf(Adj) ≥ 0.5
 Value Of Adj" if ValueOf(Adj) < 0.5

Good 0.625 Very good 0.7906

Bad 0.25 Very bad 0.0625
Awesome 0.875 Simply Awesome 0.9354
Pathetic 0.375 Highly Pathetic 0.1406

Table 1: Fuzzy Measure of (Adverb, Adjective) Phrases

Figure 4: Summary for Topmost and Bottom Most
In Case 2, we consider the weight (W) of the opinion
Sentiments
like (not/never)(Adj /Verb)
= 1 - ValueOf(Adj/Verb)

Good 0.625 Not good 0.375

Bad 0.25 Not bad 0.75
Conclusive 1.0 Never Conclusive 0.0
Stunning 0.75 Never Stunning 0.25

Table 2: Fuzzy Measure of (Not/Never, Adjective) Phrases

In case 3, we consider the weight (W) of the opinion

like (not/never) (very /extremely/highly etc)(Adj)
 A # B
Where A = (very/extremely/highly etc) (Adj)
 Value Of Adj if ValueOf(Adj) ≥ 0.5
Figure 5: Preprocessing, Classification and POS tagging

 Value Of Adj" if ValueOf(Adj) < 0.5

And B = (not/never) (Adj)

Good 0.625 Not Good 0.375 Not Very Good 0.5445

Bad 0.25 Not Bad 0.75 Not Very Bad 0.2165
Table 3: Fuzzy Measure of (Not, Adverb, Adjective)
Phrases
In case 4, we consider the weight (W) of the opinion
like (but/also/nor) words. Firstly we divide the sentiment
into two opinions group. Then these two opinions
separately calculated as follows. Finally the average of both
cases will be taken as a Fuzzy measure for the given
Figure 6: Fuzzy measure graph for SA
sentence.

F= %√' ( √)+2 if ValueOf(Any Adj) ≥ 0.5 So, if the above value is positive (i.e. >= 50%) then,
F  A ( B  /2 if ValueOf(Both Adj) < 0.5
" " " sentiment of document is positive, otherwise negative.
Here, sentiment of collective annotator over document is
positive.
Good 0.625 Awesome 0.875
√Good 0.7906 √Awesome 0.9354 VI. BENEFITS
This is a not only good but also awesome 0.9289 • Direct, unfiltered, unlimited, unbiased and real-time
Bad 0.25 Pathetic 0.375 opinions of users.

13
 International Journal of Artificial Intelligence and Knowledge Discovery Vol.3, Issue 2, April, 2013

• Cost effective approach of capturing user and reviewer [10] Fabrizio Sebastiani and Andrea Esuli, SENTIWORDNET: A Publicly
Available Lexical Resource for Opinion Mining, Proceedings of the
feedback.
5th Conference on Language Resources and Evaluation, Genoa –
• Real time and uninterrupted with wide geographic Italy, pp. 417-422, May 2006.
reach. [11] Pushpak Bhattacharyya, Akhel Agrawal, Sentiment Analysis: A New
• Actionable market intelligence based on direct user Approach for Effective Use of Linguistic Knowledge and Exploiting
Similarities in a Set of Documents to be Classified, International
feedback and comments.
Conference on Natural Language Processing, IIT Kanpur, India,
• Better reaction time for service and quality December, 2005.
improvement for market. [12] Bo Pang, Lillian Lee, and S. K. Vaithyanathan, Thumbs up?
Sentiment Classification using Machine Learning Techniques,
Proceedings of the Conference on Empirical Methods in Natural
VII. CONCLUSION
Language Processing, Volume 10, pp. 79-86, 2002.
As we can notice that SA is a trend in the Web, with [13] Gang Li and Fei Liu, A Clustering-based Approach on Sentiment
several application with a lot of data sources provided by Analysis, International Conference on Intelligent Systems and
users. Social Networking web-sites for instance Orkut, Knowledge Engineering, Hangzhou, pp. 331-337, November 2010.
[14] Jianxiong Wang and Andy Dong, A Comparison of Two Text
Facebook and Twitter and other web services are dominant
Representations for Sentiment Analysis, International Conference on
sources for obtain opinions from the users in the Web about Computer Application and System Modeling, Taiyuan, China, pp.
any subject and especially to help to answer the question 35- 39, October 2010.
about what people are interested on. In spite of the various [15] Aditya Joshi, Balamurali AR, Pushpak Bhattacharyya and Rajat
Mohanty, C-Feel-It: A Sentiment Analyzer for Micro-blogs, The 49th
challenge, more companies and researchers are working in
Annual Meeting of the Association for Computational Linguistics:
this area until one day it would be easy for users and Human Language Technologies, Portland, Oregon, USA, pp. 127-
companies to minimally obtain complete and wealthy 132. June, 2011.
summarized fact about the opinions from the Web in order [16] Siva RamaKrishna Reddy V., DVLN. Somayajulu and Ajay R.
Dani, Classification of Movie Reviews Using Complemented Naive
to uphold them in the decision making process in their daily
Bayesian Classifier, International Journal of Intelligent Computing
life. Research, UK, Volume 1, Issue 4, pp. 162-167, December 2010.
My work is an initial discussion about how we can [17] Adnan Duric and Fei Song, Feature Selection for Sentiment Analysis
learn more about sentiment. More detail analysis and a Based on Content and Syntax Models, Proceedings of the 2nd
Workshop on Computational Approaches to Subjectivity and
better web opinion mining could be applied to compare
Sentiment Analysis, Portland, Oregon, USA, pp. 96-103, June, 2011.
another popular machine learning techniques like Genetic [18] V. Rentoumi, S. Petrakis, M. Klenner, G. A. Vouros and V.
Programming, SVM, Decision Trees. Karkaletsis, United we stand: improving sentiment analysis by
joining machine learning and rule based methods, 7th International
Conference on Language Resources and Evaluation, Malta, pp. 1089-
REFERENCES
1094, May 2010.
[1] Swati A. Kawathekar and Dr. Manali M. Kshirsagar, Movie Review
[19] Julia Maria Schulz, Christa Womser Hacker and Thomas Mandl,
analysis using Rule-Based & Support Vector Machines methods,
Multilingual Corpus Development for Opinion Mining, Proceedings
IOSR Journal of Engineering, Vol. 2(3) pp: 389-391, March 2012.
of the 7th International Conference on Language Resources and
[2] Alaa Hamouda, Mahmoud Marei and Mohamed Rohaim, Building
Evaluation, Valletta, Malta, pp. 3409-3412, May 2010.
Machine Learning Based Senti-word Lexicon for Sentiment Analysis,
[20] Xiaoxu Fei, Huizhen Wang and Jingbo Zhu, Sentiment word
Journal of Advances In Information Technology, Vol. 2, No. 4, pp
identification using the maximum entropy model, International
199-203, November 2011.
Conference on Natural Language Processing and Knowledge
[3] Archana Shukla, Sentiment Analysis Of Document Based On
Engineering, Beijing, pp. 1-4, September 2010.
Annotation, International Journal of Web & Semantic Technology
[21] Huosong Xia, Min Tao and Yi Wang, Sentiment Text Classification
(IJWesT) Vol.2, No.4, pp 91-103, October 2011
of Customers Reviews on the Web Based on SVM, Sixth International
[4] Aurangzeb Khan, Baharum Baharudin and Khairullah Khan, 2011.
Conference on Natural Computation, 2010.
Sentiment Classification Using Sentence-level Lexical Based
[22] Alena Neviarouskaya, Helmut Prendinger and Mitsuru Ishizuka,
Semantic Orientation of Online Reviews. Trends in Applied Sciences
SentiFul: Generating a Reliable Lexicon for Sentiment Analysis, 3rd
Research, Vol. 6, pp. 1141-1157, July, 2011.
International Conference on Affective Computing and Intelligent
[5] C. Hauff, Dadvar, Maral and Jong de, Franciska, Scope of negation
Interaction and Workshops, Amsterdam, pp. 1-6, September 2009.
detection in sentiment analysis, Dutch-Belgian Information Retrieval
[23] Miho Itoh, Contextual Analysis Processing Methods able to interpret
Workshop, Netherlands, February 2011.
Sentiments Evaluation Representation, IEEE International
[6] Animesh Kar, Deba Prasad Mandal, Finding Opinion Strength Using
Conference on Semantic Computing, Keio University, pp. 71-76,
Fuzzy Logic on Web Reviews, International Journal of Engineering
2009
and Industries, volume 2, Number 1, pp 37-43, March, 2011
[24] Alena Neviarouskaya, Helmut Prendinger and Mitsuru Ishizuka,
[7] Alena Neviarouskaya, Helmut Prendinger, Mitsuru Ishizuka, Affect
Emantically Distinct Verb Classes Involved In Sentiment Analysis,
Analysis Model: novel rule-based approach to affect sensing from
International Conference Applied Computing, Japan, pp: 27-34,
text, Natural Language Engineering, Cambridge University, Vol. 17,
2009.
pp. 95- 135, September 2010.
[8] Brendan Tierney and Bruno Ohana, Sentiment Classification of [25] Cheng Mingzhi1，Xin Yang, Bao Jingbing,Wang Cong and Yang
Reviews using SentiWordNet, 9th IT&T Conference, Dublin Institute Yixian, A Random Walk Method for Sentiment Classification, 2nd
of Technology, Ireland, October, 2009. International Conference on Future Information Technology and
[9] Mike Thelwall and Rudy Prabowo, Sentiment Analysis: A Combined Management Engineering, pp. 327-330, 2009
Approach, Journal of Informatics, School of Computing and [26] Guohong Fu and Xin Wang, Chinese Sentence-Level Sentiment
Information Technology, University of Wolverhampton, UK, Classification Based on Fuzzy Sets, Coling: Poster Volume, Beijing,
Volume 3, Issue 2, pp. 143-157, April 2009. pp 312-319, August 2010

14