International Journal of Engineering & Technology, 7 (2.

32) (2018) 473-476

International Journal of Engineering & Technology

Website: www.sciencepubco.com/index.php/IJET

Research paper

Sentiment Analysis for Social Networks Using Machine

Learning Techniques
Dorababu Sudarsa1, Siva kumar.P2, L.jagajeevan Rao3
1
Asst.Prof , K L E F , vijayawada, India .
2
Assoc. Professor K L E F , vijayawada, India,
3
Asst.Professor, K L E F , vijayawada, India.
*Corresponding author E-mail: dorababu.sudarsa@gmail.com

Abstract

The tremendous of the overall enormous net has conveyed a present day way of communicating the feelings of individuals. It's
additionally a medium with a vast amount of data in which clients can see the assessment of different clients which can be ordered into
exceptional entailment summons and are progressively more boom as a key component in decision making. This paper adds to the
supposition assessment for customers assessment class that is utilized to analyze the records inside the type of the assortment of tweets
wherein investigates are very unstructured and are both high fine or terrible, or somewhere in the middle of these . For this we first pre-
prepared the dataset, after that extract the adjective from the dataset that has a couple of significance this is alluded to as capacity vector,
at that point decided on the component vector posting and from that point accomplished device examining based write calculations
particularly navie bayes, most entropy and svm along the edge of the semantic introduction based absolutely based on word net which
extracts synonyms and similarity for the content characteristic. In the end, we measured the performance of the cl assifier in terms of
considering, precision and accuracy.

Keywords:Machine Learning, Semantic Orientation, Sentiment Analysis, Twitter.

where in the tweets convey surveys, however looking to harvest the

I I I I I I I I I I I

general comprehension of these unstructured records (audits) might

1. Introduction I I I I I I

be extremely time ingesting. those unstructured measurements

I I

(audits) on a particular site are seen through by the customers and

The age of internet has modified the manner humans explicit their
I I I I I I I I I I I
subsequently growing a picture about the items or offerings and
perspectives. it's miles now executed via weblog posts, on line dia-
I I I I I I I I I I
accordingly at long last delivering a beyond any doubt judgment.
logue forums, product review web sites and so forth. people depend
I I I I I I I I I I I
these surveys are then being summed up to secure criticisms for
upon this person generated content to a terrific quantity. while a
I I I I I I I I I I I
unprecedented purposes to offer valuable audits in which we uti-
person needs to shop for a product, they will appearance up its eval-
I I I I I I I I I I I I
lize assessment examination. Slant assessment is a procedure in
uations online earlier than taking a choice. the amount of user gen-
I I I I I I I I I I I
which the dataset incorporates emotions, mentalities or evaluation
erated content is simply too huge for a
I I I I I I I I
which mulls over the way a human thinks [1]. In a sentence, hop-
ing to secure the positive and the negative segment is a totally
ordinary consumer to research. so that you could automate this,
I I I I I I I I I I
hard endeavor. the capacities used to classifications the sentences
numerous sentiment analysis strategies are used. Symbolic strate-
I I I I I I I
ought to have a totally Sturdy modifier with the expectation to
gies or understanding base approach and machine mastering tech-
I I I I I I I I
abridge the assessment. Those s are even composed in unmistaka-
niques are the 2 important strategies used in sentiment analysis.
I I I I I I I I I I
ble strategies which are not without issues concluded with the
Expertise base approach requires a large database of predefined
I I I I I I I I I
guide of the clients or the organizations making it extreme to
feelings and an green expertise illustration for figuring out senti-
I I I I I I I I I
group them. Opinion assessment influences clients to classes
ments. Machine learning approach uses a education set to expand a
I I I I I I I I I I I
whether the information around the item is decent or no longer
sentiment classifier that classifies sentiments. Given that a prede-
I I I I I I I I
before they gather it. Advertisers and organizations utilize this
fined database of entire emotions is not required for system learn-
I I I I I I I I I I
investigation to catch about their administrations or items in the
ing method, it's far alternatively easier than knowledge base tech-
I I I I I I I I I
kind of way that it might be furnished as with regards to the indi-
nique. on this paper, we use special device getting to know tech-
I I I I I I I I I I I
vidual's needs. There are two styles of machine picking up infor-
niques for classifying tweets. I I I
mation of techniques which may be typically utilized for notion
investigation, one is unsupervised and the other is managed [2].
The forefront of this exploration paper covers the assessment of the
I I I I I I I I I I I
Unsupervised becoming more acquainted with does not comprise
substance at the web covering bunches of zones that are wrapping
I I I I I I I I I I I
of a class and that they don't give the ideal destinations at all and
exponentially in numbers notwithstanding in volumes as destina- I I I I I I I
therefore conduct bunching. Managed learning is construct abso-
tions are devoted to specific kinds of stock and that they have prac-
I I I I I I I I I I I I
lutely with respect to named dataset and hence the names are out-
tical experience in storing up clients' audits from different locales,
I I I I I I I I I I
fitted to the model all through the strategy. those sorted dataset are
for example, Amazon and numerous others. Considerably twitter is
I I I I I I I I I

Copyright © 2018 Authors. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.
474 International Journal of Engineering & Technology

prepared to supply reasonable yields when experienced all through mark and specific capacity. 'advisor highlight' is the records that
I I I I I I I I I I

basic leadership. to help us to secure the opinion examination in a speaks to a class and 'specific element' is the information that en-
I I I I I I I I I I I

superior way, this exploration paper is based at the supervised courages in recognizing directions. the utilization of the ones
I I I I I I I I I

system learning. weights, they figured the likelihood of every class and henceforth
I I I I I I I I I I

ventured forward the bayesian arrangement of rules. barbosa et al.

I I I I I I I I I I

2. Related Work [12] outlined a 2-step programmed opinion assessment method for
I I I I I I I I I

arranging tweets. they utilized a boisterous tutoring set to lessen

I I I

the marking endeavor in developing classifiers. above all else,

There are basic techniques to unearth notions from content. They
I I I I I I I I I I
they sorted tweets into subjective and objective tweets. from that
represents methodologies and contraption becoming acquainted I I I I I I
point forward, subjective tweets are named as gigantic and nega-
with systems [2]. The accompanying two areas adapt to these meth-
I I I I I I I I I I
tive tweets. celikyilmaz et al. [13] propelled an elocution princi-
odologies. I
pally based expression grouping approach for normalizing bois-
terous tweets. in elocution based word grouping, phrases having
A. Representative Techniques Much of the examination in unsu-
I I I I I I I I
similar articulation are bunched and doled out typical tokens. they
pervised conclusion characterization utilizing emblematic proce- I I I I I
likewise utilized content handling systems like doling out compa-
dures makes utilization of accessible lexical assets. Turney [3] uti-
I I I I I I I I I
rable tokens for numbers, html joins, client identifiers, and objec-
lized pack of-words approach for assessment examination. In that
I I I I I I I I I
tive business undertaking names for standardization. subsequent to
approach, connections between the individual words are not con- I I I I I I I I
doing standardization, they utilized probabilistic models to distin-
sidered and a report is spoken to as a minor gathering of words. To
I I I I I I I I I I I I I I
guish extremity vocabularies. they performed class the utilization
decide the those qualities are joined with some total capacities.
I I I I I I I I I I
of the boostexter classifier with these extremity dictionaries as
capacities and obtained a decreased bungles cost. wu et al. [14]
He found the extremity of an audit in view of the normal semantic
I I I I I I I I I I I I I
proposed an affect opportunity show for twitter assessment. in the
introduction of tuples separated from the survey where tuples are I I I I I I I I I I
event that @username is situated inside the casing of a tweet, it's
phrases having descriptors or verb modifiers. He found the seman-
I I I I I I I I I
far impacting development and it adds to affecting Shot. any tweet
tic introduction of tuples utilizing the internet searcher Altavista.
I I I I I I I I I
that begins with @username is a rewet that speaks to a motivated
Kamps et al. [4] utilized the lexical database WordNet [5] to decide
I I I I I I I I I I I I
movement and it adds to empowered probability. they found that
the passionate substance of a word along various measurements.
I I I I I I I I I
there is a solid relationship among these probabilities. pak et al.
They built up a separation metric on WordNet and decided the se-
I I I I I I I I I I I
[15] made a twitter corpus through mechanically storing up tweets
mantic introduction of descriptors. WordNet database comprises of
I I I I I I I I
the utilization of twitter programming interface and routinely
words associated by equivalent word relations. Baroni et al. [6]
I I I I I I I I I I
commenting on the ones the use of emotions. utilizing that corpus,
built up a framework utilizing word space show formalism that
I I I I I I I I I I
they built an estimation classifier fundamentally based at the mul-
beats the trouble in lexical substitution assignment. It speaks to the
I I I I I I I I I I I
tinomial guileless bayes classifier that utilizations n-gram and pos-
nearby setting of a word alongside its general dissemination. Bala-
I I I I I I I I I
labels as abilities. in that
hur et al. [7] presented EmotiNet, an applied portrayal of content
I I I I I I I I I I I

that stores the structure and the semantics of genuine occasions for
I I I I I I I I I I I
Approach, there is a danger of mistake for the reason that feelings
a particular space. Emotinet utilized the idea of Finite State Autom-
I I I I I I I I I I
of tweets in preparing set are arranged exclusively basically based
ata to distinguish the passionate reactions activated by activities.
I I I I I I I I I
at the extremity of emotions. the instruction set is in like manner
One of the members of SemEval 2007 Task No. 14 [8] utilized
I I I I I I I I I I I I
less proficient since it contains most straightforward tweets having
coarse grained and fine grained ways to deal with distinguish as-
I I I I I I I I I I
emotions.
sessments in news features. In coarse grained approach, they per- I I I I I I I I I

formed double characterization of feelings and in fine grained ap-

I I I I I I I I I

proach they grouped feelings into various levels. Knowledge base

I I I I I I I I I
3. Our Approach
approach is found to be difficult due to the requirement of a huge I I I I I I I I I I I I I

lexical database. Since social network generates huge amount of

I I I I I I I I I In our technique we used the twitter dataset and analyzed it.
data every second, sometimes larger than the size of available lexi-
I I I I I I I I I I

cal database, sentiment analysis became tedious and erroneous.

I I I I I I I This investigations sorted datasets the utilization of the unigram
include extraction system. we utilized the structure wherein the
B. Machine learning methods: gadget acing strategies utilize an
I I I I I I I I I preprocessor is done to the uncooked sentences which make it
instruction set and a test set for order. Tutoring set contains input I I I I I I I I I I I I additional proper to perceive. so also, the unique gadget acing
work vectors and their comparing class names. utilizing this prepa-
I I I I I I I I I systems prepares the dataset with trademark vectors and after that
ration set,a type display is developed which endeavors to order the
I I I I I I I I I I I the semantic assessment offers an extensive arrangement of
information highlight vectors into comparing class names. at that I I I I I I I I I equivalent words and similitude which bears the extremity of the
point a test set is utilized to approve the adaptation by foreseeing
I I I I I I I I I I I I substance. The entire depiction of the system has been portrayed
the class names of inconspicuous element vectors. various gadget
I I I I I I I I I in next sub segments and the piece outline of the same is graph-
picking up learning of methodologies like credulous bayes (nb),
I I I I I I I I I ically spoken to in fig.1. Chart of the strategy to issue a. pre-
most extreme entropy (me), and bolster vector machines (svm) are
I I I I I I I I I I preparing of the datasets the tweets contain various studies around
utilized to arrange feelings [9]. a portion of the abilities that can be
I I I I I I I I I I I I I the information which are communicated in particular strategies
utilized for slant write are day and age nearness, day and age recur-
I I I I I I I I I I I I through people
rence, nullification, n-grams and grammatical feature [1].those
I I I I I I I

capacities can be utilized to find the semantic introduction of I I I I I I I I I I The twitters dataset utilized on this work is now arranged. Ar-
words, expressions, sentences and that of records. Semantic intro-
I I I I I I I I ranged dataset has a negative and top notch extremity and there-
duction is the extremity which can be either positive or negative.
I I I I I I I I I I I fore the examination of the data ends up smooth. the crude infor-
dominos et al. [10] found that guileless bayes works legitimatelyI I I I I I I I I I mation having extremity is very inclined to irregularity and repeti-
for specific issues with very settled capacities. that is shocking as
I I I I I I I I I I I tion. the charming of the certainties impacts the outcomes and
the central supposition of innocent bayes is that the capacities are
I I I I I I I I I I I subsequently which will enhance the pleasant, the crude insights is
fair. zhenniu et al. [11] presented a fresh out of the box new model
I I I I I I I I I I I I I I pre-handled. it offers with the planning that kills the rehashed
wherein effective techniques are utilized for work determination,
I I I I I I I I words and accentuations and enhances the productivity the reali-
weight calculation and class. the new form is construct absolutely
I I I I I I I I I I ties. for instance, "that artistic creation is beauuuutifull #" in the
with respect to bayesian arrangement of tenets. ideal here weights
I I I I I I I I I I wake of preprocessing proselytes to "painting staggering." fur-
of the classifier are balanced by method for utilizing advisor trade-
I I I I I I I I I I ther,"@geet is currently persevering" believers to "geet now dedi-
International Journal of Engineering & Technology 475

cated". b. trademark extraction the ventured forward dataset after The description of the manner in pseudo code shape is shown.
I I I I I I I I I I

pre-handling has heaps of one of a kind homes. the capacity ex-

traction approach, extricates the component (descriptive word) input: categorized dataset I I

from the dataset. later this descriptor is utilized to demonstrate the output: fantastic and bad polarity with synonym of I I I I I I I

high caliber and horrendous extremity in a sentence which is valu- words and similarity between phrases I I I I

able for deciding the feeling of the general population the use of step-1 pre-processing the tweets: I I I

unigram show [15]. unigram show separates the modifier and pre-processing () I

isolates it. it disposes of the past and progressive word happening eliminate url: I

with the descriptive word in the sentences. for above case, i.e. eliminate unique symbols I I

"depict shocking" by means of unigram demonstrate, easiest de- convert to lower: I I

lightful is removed from the sentence. step-2 get the feature vector listing: I I I I I

for w in phrases:I I I

After the tutoring and classification we utilized semantic investi- update or more words I I I

gation. semantic investigation is gotten from the word net data- strip:
base in which each day and age is related with each extraordinary. if (w in stop words)
I I I I

This database is of English words that are connected together. on hold

the off chance that expressions are close to each other, they're else:
semantically comparable. additional particularly, we're ready to append the document I I

decide equivalent word like likeness. we delineate and take a gan- go back function vector
I I I

der at their relationship in the cosmology. the key mission is to step-3 extract features from feature vector listing:
I I I I I I

apply the spared documents that contain terms and afterward in- for phrase in feature listing
I I I I

vestigate the comparability with the expressions that the individu- capabilities=phrase in tweets words I I I

al employments of their sentences. in this manner it's miles gainful go back features
I I

to uncover the extremity of the assessment for the clients. as an step-4 integrate pre-processing dataset and feature
I I I I I

example inside the sentence's am fulfilled" the word ''fulfilled'' vector listing I

being a descriptor gets chose and is in correlation with the spared pre-processed record=course call of the record I I I I I

work vector for equivalent words. Give us a chance to expect 2 stopwords=file direction name I I

phrases; 'fulfilled' and 'happy' have a tendency to be particularly function vector listing=document path of characteristic vector I I I I I I

similar to the expression 'happy'. presently after the semantic as- listing
sessment, 'happy' replaces 'fulfilled' which gives a fine extremity. step-5 training the step 4 I I I I

practice classifiers classes I I

4. Usage and End Result IIIII step-6 discover synonym and similarity of the characteristic vec-
I I I I I I I I

tor
for each sentences in function listing
I I I I I

we utilized python and normal dialect gadget unit to teach and extract feature vector within the tweets () I I I I I I

group the credulous bayes, most entropy and guide vector frame- for every function vector: x
I I I I

work. by and large we utilized records set of length 19340 out of for each function vector: y
I I I I

which 18340 have been utilized for preparing and 1000 for look- locate the similarity(x, y) I I I

ing at. for training fig 2. show the overall waft of techniques. if (similarity>threshold)
I

fit observed I

feature vector: x= feature vector: y I I I I I

classify (x, y) I I

print: sentiment polarity with similar feature phrases

I I I I I I

Fig. 3. Pseudo code of the procedure:

I I I I I I

5. Conclusion
I

In this paper, we proposed a set of techniques of system gaining

I I I I I I I I I I I I

knowledge of with semantic evaluation for classifying the sentence I I I I I I I I I

and product opinions based totally on twitter facts. the important

I I I I I I I I I I

thing aim is to analyze a massive amount of reviews through using

I I I I I I I I I I I I

twitter dataset which are already classified. the naïve byes approach
I I I I I I I I I I

which offers us a better end result than the maximum entropy and
I I I I I I I I I I I I

svm is being subjected to unigram version which gives a better end

I I I I I I I I I I I I

result.

Than the use of it alone. Further the accuracy is once more pro-
I I I I I I I I I I I I

gressed while the semantic evaluation word net is followed up by

I I I I I I I I I I I

way of the above method taking it to 89.9% from 88.2%. the train-
I I I I I I I I I I I I

ing facts set can be improved to improve the feature vector related
I I I I I I I I I I I I

sentence identity process and can also expand word net for the
I I I I I I I I I I I

summarization of the critiques. it may supply better visualization of I I I I I I I I I I

the content in better manner with a view to be beneficial for the

I I I I I I I I I I I I I

customers.

There are distinct symbolic and device gaining knowledge of strat-

I I I I I I I I I

egies to become aware of sentiments from textual content. device

I I I I I I I I I I

Fig. 2. Flow Diagram of the proposed methodology gaining knowledge of techniques are less complicated and green
I I I I I I I I I

than symbolic techniques. these techniques may be applied for twit-

I I I I I I I I I
476 International Journal of Engineering & Technology

ter sentiment evaluation. there are sure issues at the same time as
I I I I I I I I I I I I [11] M. A. Russell, Mining the Social Web: Data Mining Face-
managing figuring out emotional key-word from tweets having
I I I I I I I I
book,Twitter, LinkedIn, Google+, GitHub, and More. O'Reilly Me-
multiple key phrases. it's miles additionally difficult to address
I I I I I I I I I
dia, Inc,2013.
[12] L. Bing, K. C. Chan and C. Ou, Public sentiment analysis in twitter
misspellings and slang words. to deal with those issues, an efficient
I I I I I I I I I I I
data for prediction of a company's stock price movements, in
function vector is created through doing characteristic extraction in
I I I I I I I I I
EBusiness Engineering (ICEBE), 2014 IEEE 11th International
two steps after right preprocessing. In step one, twitter precise func-
I I I I I I I I I I
Conference on, 2014, pp. 232-239.
tions are extracted and delivered to the feature vector. after that,
I I I I I I I I I I I [13] M. Mittermayer, Forecasting intraday stock price trends with text
these features are removed from tweets and once more characteris-
I I I I I I I I I mining techniques, in System Sciences, 2004. Proceedings of the
tic extraction is executed as if it's far accomplished on regular tex-
I I I I I I I I I I I 37th Annual Hawaii International Conference on, 2004, pp. 10-pp.
tual content. these features are also introduced to the characteristic
I I I I I I I I I I
[14] R. Socher, A. Perelygin, J. Y. Wu, J. Chuang, C. D. Manning, A.
vector. category accuracy of the function vector is examined the
I I I I I I I I I I
Y.Ng and C. Potts, Recursive deep models for semantic com-
positionality over a sentiment treebank, in Proceedings of the Con-
use of different classifiers like nave bayes, svm, maximum entropy
I I I I I I I I I I
ference on Empirical Methods in Natural Language Processing
and ensemble classifiers. most of these classifiers have almost simi-
I I I I I I I I I
(EMNLP), 2013, pp. 1642.
lar accuracy for the new function vector. this feature vector per-
I I I I I I I I I I
[15] M. Sokolova and G. Lapalme, A systematic analysis of per-
forms nicely for digital products domain
I I I I I I formance measures for classification tasks, Information Processing
& Management, vol. 45, pp. 427-437, 2009.
[16] Go, R. Bhayani and L. Huang, Twitter sentiment classifi-cation us-
6. Conclusion ing distant supervision, CS224N Project Report, Stan-ford, pp. 1-
12,2009.
In this paper we contributed a methodical survey of supposition [17] Bermingham and A. F. Smeaton, Classifying sentiment in mi-
examination and sentiment mining. The multifaceted nature of croblogs: Is brevity an advantage? in Proceedings of the 19th ACM
data Presentation and dimensionality, distinctive use necessities, International Conference on Information and Knowledge Manage-
ment, 2010, pp. 1833-1836.
the conclusion examination or sentiment mining developed as
[18] Pak and P. Paroubek, Twitter as a corpus for sentiment analysis and
basic research objective thinking about that 10 years. This assess opinion mining. in Lrec, 2010, pp. 1320-1326.
investigated the notion assessment method, contemporary assess- [19] L. Barbosa and J. Feng, Robust sentiment detection on twitter from
ment of The machine acing based absolutely assumption assess- biased and noisy data, in Proceedings of the 23rd Interna-tional
ment designs found in late writing, effect of contraption learn- Conference on Computational Linguistics: Posters, 2010, pp. 36-44.
ing .Conclusion investigation and plausible and limit thinks about [20] Oh and O. Sheng, Investigating predictive power of stock micro
focuses for predetermination look into. At some point or another, blog sentiment in forecasting future stock price directional move-
we finish up the Manuscript by utilizing saying that all the opinion ment, 2011.
[21] Tayal and S. Komaragiri, Comparative Analysis of the Impact of
assessment obligations are hard, because of the reality ability and
Blogging and Micro-blogging on Market Performance, Internation-
know-how of the inconvenience and its answers are as yet obliged. al Journal, vol. 1, pp. 176-182, 2009.
The principle object is that it's far a home grown dialect handling [22] M. Sokolova and G. Lapalme, A systematic analysis of per-
undertaking, and Herbal dialect preparing has no simple issues. Be formance measures for classification tasks, Information Processing
that as it may, numerous huge advances were made. Obvious to & Management, vol. 45, pp. 427-437, 2009.
finish that the sentiment evaluation is having potential scope for [23] S. Dreiseitl and L. Ohno-Machado, Logistic regression and artifi-
destiny research and certainly one of that is exposing The scope of cial neural network classification models: a methodology re-view,
evolutionary computational or soft computing strategies and the J.Biomed. Inform., vol. 35, pp. 352-359, 2002.
[24] Gartner, http://www.gartner.com/newsroom/id/766215 ac-cessed on
hybridizing these techniques in the direction of Function extrac-
Mar 30, 2015.
tion, selection to categories the sentiment. [25] IDC, http://blogs.idc.com/ie/?p=190, accessed on Mar 30,2 015.
[26] The451group,www.451group.com/reports/execu=ve_summary.php
References ?id=619,accessed on Mar 30, 2015.

[1] R. Feldman, ” Techniques and Applications for Sentiment Analy-

sis ,” Communications of the ACM, Vol. 56 No. 4, pp. 82-89, 2013.
[2] Y. Singh, P. K. Bhatia, and O.P. Sangwan, ”A Review of Stud-ies
on Machine Learning Techniques,” International Journal of Com-
puter Science and Security, Volume (1) : Issue (1), pp. 70-84, 2007.
[3] P.D. Turney,” Thumbs Up or Thumbs Down? Semantic Orien-
tation Applied to Unsupervised Classification of Reviews,” Pro-
ceedings of the 40th Annual Meeting of the Association for Com-
putational Lin-guistics (ACL), Philadelphia, pp. 417-424, July 2002.
[4] Ch.L.Liu, W.H. Hsaio, C.H. Lee,and G.C.Lu, and E. Jou,” Movie
Rating and Review Summarization in Mobile Environment,” IEEE
Transactionson Systems, Man, and Cybernetics, Part C
42(3):pp.397-407, 2012. [5
[5] R.Liu,R.Xiong,and L.Song, ”A Sentiment Classification Meth-od
for Chinese Document,” Processed of the 5th International Con-
ference on Computer Science and Education (ICCSE), pp. 918 –
922, 2010.
[6] A.khan,B.Baharudin, ”Sentiment Classification Using Sentence-
level Semantic Orientation of Opinion Terms from Blogs,” Pro-
cessed on National Postgraduate Conference (NPC), pp. 1 – 7, 2011.
[7] L.Ramachandran,E.F.Gehringer, ”Automated Assessment of Re-
view Quality Using Latent Semantic Analysis,” ICALT, IEEE
Computer Society, pp. 136-138, 2011.
[8] B. Pang and L. Lee, Opinion mining and sentiment analysis, Foun-
dations and Trends in Information Retrieval, vol. 2, pp. 8-10, 2008.
[9] B. G. Malkiel, A Random Walk Down Wall Street: Including a
LifeCycle Guide to Personal Investing. WW Norton & Company,
1999.[3
[10] J. Bollen, H. Mao and X. Zeng, Twitter mood predicts the stock
market, Journal of Computational Science, vol. 2, pp. 1-8, 2011.