Potentials and limitations of ‘Automated Sentiment Analysis

Potentials and Limitations Of Automated
Sentiment Analysis of Weblogs
GEORG-AUGUST-UNIVERSITÄT GÖTTINGEN
06 - Jun -14 1
Seminar Presented by :
Gollapinni Karthik
Student Id: 11337533
Course: Masters of Applied Computer Science

Foundations
Approaches of ASA
Potentials of ASA
Limitations of ASA
Conclusions
AGENDA
Introduction
06 - Jun -14 2

INTRODUCTION(1/2)
 Web is an excellent source for gathering customers opinions.
 What others think has always been an important piece of information.
 85% of the people gather information online before buying a product
1) Pre-Web
 Friends
 Customer reports
2) Post-Web
 Blogs
 Review Sites
 Discussion forums
06 - Jun -14 3

INTRODUCTION(2/2)
06 - Jun -14 4
 150 Million Number of blogs online
 1 billion Number of blog/forums readers
 55 million Number of tweets in a month
 180 million Number of unique visitors to twitter everyday
Answers Research Questions :-
 Which application fields is ASA useful?
 How far are limitations restricting the exploitation of potentials?

Introduction
Approaches of ASA
Potentials of ASA
Limitations of ASA
Conclusions
AGENDA
Foundations
06 - Jun -14 5

FOUNDATIONS
What does Automated Sentiment Analysis mean?
 ´Opinion Mining´
 `A system which automatically identifies the sentiments and extract the information
from the emotions or options and process a finite decision of the avaliable sentiments
from the given pool of content (Butler, Eugene. "Automated Sentiment Analysis”)´
 Natural Processing Language(NLP)
 Reshape the businesses
06 - Jun -14 6

What are Weblogs?
‘A website which consists of discrete
number of series of entries which are
arranged in reverse chronological
order and are often updated
frequently with new information
about the particular topics (Rouse,
Margaret. “Weblogs”)’
 Log of our times
 ´User-generated data´
Personal, Business or Community
are 3 types of weblogs.
06 - Jun -14 8

Introduction
Foundations
Potentials of ASA
Limitations of ASA
Conclusions
AGENDA
Approaches of ASA
06 - Jun -14 9

Computer Coding Vs Human Coding
06 - Jun -14 10

DIFFERENCES BETWEEN COMPUTER CODING AND
HUMAN CODING
06 - Jun -14 11
COMPUTER CODING HUMAN CODING
Uses Dictionary based or Machine
Learning Methods.
Follows standard code book and
coding form.
Involves software for automation
process.
Involves people as coders.
Automated tabulation of variables. Human observation is recorded on
pre-established variables.

06 - Jun -14 12
BUILDING BLOCKS OF ASA
5 main factors ASA looks at:
 Topics: Main areas of discussion.
 Aspects (subtopics and attributes):
What about the topics being talked ?
 Sentiment: What is the sentiment of
the content?
 Holder: Whose is being analyzed?
 Time: When was the content posted?
LEVELS
Of
ASA
Statement
Level
Document
Level
Aspect
Level

06 - Jun -14 13
MACHINE LEARNING METHODS

06 - Jun -14 14
DICTIONARY METHOD

Introduction
Foundations
Approaches of ASA
Potentials of ASA
Limitations of ASA
Conclusions
AGENDA
Potentials of ASA
06 - Jun -14 15

METHODOLOGY FOLLOWED
Followed ‘Literature Review’ to categorize the Areas of the Potentials and Limitations.
Used scholar.google.com to take the books and papers.
Most used books and papers:
1. (Bing Liu 2011)
2. (Brain Conlin 2012)
3. (Michael Gamon)
4. (Kimberly Neuendarf 2002)
5. (Yayan Meng 2012)
6. (Adam Westerki 2013)
7. (Bing Liu 2012)
8. (Ben Donkor 2013)
9. (Sitaram/Bernardo 2010)
06 - Jun -14 16

POTENTIAL AREAS OF ASA
06 - Jun -14 17
Areas
Of
ASA
Financial
Domain
Individual
Needs
Supervised
Spam
Detection
Marketing
Domain

Advertisement Placement
 Product benchmarking and Market Intelligence
 Opinion Summarization
 Opinion search and Retrieval
Practical examples :
Analyze the comments made on the weblogs to provide better customer support.
Monitoring the Company brand.
06 - Jun -14 18
MARKETING DOMAIN

Stock Markets
Analyze market
sentiments
Public Relations
Investor Relations
06 - Jun -14 19
FINANCIAL DOMAIN

INDIVIDUAL NEEDS
ASA system could be potentially used by :
Internet Users
Summarized view of posts for community review sites
Other decision making tasks
Examples:
1. Movie reviews or product reviews on Amazon book store (Hu and Liu, 2004)
2. Product reviews on Other websites(Hariharan et al., 2010)
06 - Jun -14 20

 Posts are being abused.
Fake comments and reviews
Generic reviews
Random texts
Example : -
“Get one for FREE!!! Have a look at this video first:
http://www.youtube.com/watch?v=DFKYVE__Mug Just take 2 minutes and read this.
This is very EASY to do…”
06 - Jun -14 21
SUPERVISED SPAM DETECTION

Introduction
Foundations
Approaches of ASA
Potentials of ASA
Conclusions
AGENDA
Limitations of ASA
06 - Jun -14 22

06 - Jun -14 23
LIMITATIONS OF ASA(1/2)
Language Specific
Limitations
Technological Limitations Data Specific Limitations
Words have different meanings Extraction of Entities Noisy data
Language is a barrier Cannot identify the root cause of
the review
Videos and Images
Comparative and Complex
statements
Entity level vs Article level
Sentiment.
Large data sets and lots of spam
Scarsam statements. Ex:Irony Human Accuracy is missing. Not available for all domains
Short forms and many
representations of the words.
Only three categories to categorize
the reviews.
Fails to classify with respect to
others prespective
Co–relationship between the
sentences.
Focused on explicit opinion. Knowledge of the data

LIMITATIONS OF ASA(2/2)
06 - Jun -14 24
Limitations of
Document –Level – ASA
Limitations of
Statement – Level –
ASA
Limitations of Aspect –
Level – ASA
Need more addational details of
the aspects liked and disliked.
Cannot deal with the conditional,
sarcasm and question statments.
Fails in aspect extraction and
aspect sentiment classification
and the sentiment words can
handle up to 60% of problems.
Not applicapable for weblogs and
non-review forums like
discussions, blogs
One common language. Not available for all the domains.
Does not provide fine grained
analysis
Complex statements have
different sentiments on different
targets.
Noisy data.

FACTORS ASA SHOULD FOCUS
Demystifying accuracy
Content Type Filter
Entity level vs. Article level Sentiment
Human Accuracy
Sentiment Override
06 - Jun -14 25

Introduction
Foundations
Approaches of ASA
Potentials of ASA
Limitations of ASA
AGENDA
Conclusions
06 - Jun -14 26

CONCLUSIONS
Sentiment analysis tackleschallenging tasks that involve NLP and text mining.
There are many challenging problems yet to be solved.
Has a strong commercial interest.
Future of Sentiment Analysis
The gap between the social media, blogs and the sentiment analysis can be covered by bridging the gap
between the insight and action.
The future can hold by using concepts of influencer analytics with analysis can help for better results.
In near future on the whole ASA can be an advantage to various social analyses, predictions and finally a master
piece insight to the company social performance.
06 - Jun -14 27

Thanks for listening
The END
06 - Jun -14 28

06 - Jun -14 29
RERERENCES
 Westerski, Adam. Sentiment Analysis: Introduction and the State of the Art Overview. Tech. N.p.: n.p., n.d.
Print.
 Ganesan, Kavita, and Hyun Duk Kim. "Opinion Mining Tutotrial." Slideshare.net. Web.
 Sentiment Analysis. Digital image. Itresearches.ru. N.p., n.d. Web.
 Liu, Bing. "Chapter 1 Sentiment Analysis: A Fascinating Problem." Sentiment Analysis and Opinion Mining. 2nd
ed. Vol. 1. San Rafael, CA: Morgan & Claypool, 2012. Pp. 7-48. Print.
 Butler, Eugene. "Automated Sentiment Analysis." , . 4 Dec. 2010. Lecture.
 McGlohon, Mary, Natalie Glance, and Zach Reiter. Star Quality: Aggregating Reviews to Rank Products and
Merchants. Tech. N.p.: n.p., n.d. Proceedings of the Fourth International AAAI Conference on Weblogs and
Social Media
 Local, Reach . "150 Smart Stats about Online Marketing." , . 29 Aug. 2012. Lecture.
 Outline of Sentiment Analysis. Digital image. Softwareunwound. N.p., n.d. Web.
 "How to Learn About Micro Blogging." WikiHow. Ed. Teresa. N.p., n.d. Web.
 "Types of Weblogs." Southbourne. N.p., n.d. Web.

REFERENCES
 Neuendorf, Kimberly A. “Chapter 3 Beyond Description : An Integrative Model of Content Analysis” The
Content Analysis Guidebook. 1st Edition. 1. United States Of America: Sage Publications, 2002. Pp. 35-36
 Federico P. "LAPS: Sentiment Analysis Project Presentation." Nextbit. N.p., n.d. Web.
 ‘Sentiment Analysis -The Next Big Thing In Social Media Marketing’
 "Introduction to Sentiment Analysis Applied to the Stock Market." QuantShare Trading Software. N.p., n.d.
Web.
 Wang, Jun. Sentiment Analysis in Practice. Publication no. 1. N.p.: n.p., n.d. Print.
 Ogneva, Maria. "How Companies Can Use Sentiment Analysis to Improve Their Business“
 Kaefer, Mark. Let's Get Social: Web User Preferences, Habits and Actions in Spring 2012. Rep. N.p.:
Burstmedia, n.d. Print.
 Hu,M., B., August 2004. Mining and summarizing customer reviews. In: Proceedings of ACM SIGKDD
International Conference on Knowledge Discovery and Data Mining (KDD’04). Seattle, Washington, USA, pp.
168-177.
06 - Jun -14 30

REFERENCES
 Hariharan, S., Srimathi, R., Sivasubramanian, M., Pavithra, S., 2010. Opining Mining and summarization of
reviews in webforums. In: Proceedings of the Third Annual ACM Bangalore Conference.
 (Bing Liu 2011) : Bing Liu. Sentiment Analysis and Opinion Mining, Morgan &Bollen, Johan, Huina Mao, and
Xiao- Jun Zeng. Twitter Mood Predicts the Stock Market. Tech. N.p.: n.p., 2011. Print.
 (Brain Conlin 2012) : Bing Liu. Sentiment Analysis and Opinion Mining, Morgan &Bollen, Johan, Huina Mao,
and Xiao- Jun Zeng. Twitter Mood Predicts the Stock Market. Tech. N.p.: n.p., 2011. Print.
 (Michael Gamon) : Gamon, Michael. "Linguistic Correlates of Style: Authorship Classification with Deep
Linguistic Analysis Features - Microsoft Research."Http://research.microsoft.com/. International Conference
on Computational Linguistics/ Coling.
 (Kimberly Neuendarf 2002) : Neuendorf, Kimberly A. The Content Analysis Guidebook. 1st Edition. 1. United
States Of America: Sage Publications.
 (Bing Liu 2012) : Liu, Bing. Sentiment Analysis and Opinion Mining. 2nd ed. Vol. 1. San Rafael, CA: Morgan &
Claypool, 2012. Print.
 (Ben Donkor 2013) : Donkor, Ben. "On Social Sentiment and Sentiment Analysis." Web log post .Brnrd.me.
N.p.,
06 - Jun -14 31

REFERENCES
 (Yayan Meng 2012) : The Truth About Sentiment & Natural Language Processing." Current
Technological Meng, Yanyan. SENTIMENT ANALYSIS: A STUDY ON PRODUCT FEATURES. Thesis.
University of Nebraska, 2012. Lincoln: DigitalCommons, 2012. Print. Miller, George A., Richard
Beckwith, Christiane Fellbaum, Derek Gross, and Katherine.
 (Adam Westerki 2013) : Westerski, Adam. Semantic Technologies In Idea Management Systems: A
Model for Interoperability, Linking & Filtering. Thesis. Universidad Politecnica De Madrid/ Madrid,
2013. N.p.: Google, 2013. Print.
 (Sitaram/Bernardo 2010) : Asur, Sitaram, and Bernardo A.Huberman. Predicting the Future With
Social Media. Tech. N.p.: n.p., 2010. Print
06 - Jun -14 32

Potentials and limitations of ‘Automated Sentiment Analysis

More Related Content

What's hot (20)

Similar to Potentials and limitations of ‘Automated Sentiment Analysis (20)

Recently uploaded (20)

Potentials and limitations of ‘Automated Sentiment Analysis