Application of Text Classification and Clustering of Twitter Data for Business Analytics

Abstract

In the recent years, social networks in business are gaining unprecedented popularity because of their potential for business growth. Companies can know more about consumers’ sentiments towards their products and services, and use it to
better understand the market and improve their brand. Thus, companies regularly reinvent their marketing strategies and campaigns to fit consumers’ preferences. Social analysis
harnesses and utilizes the vast volume of data in social networks to mine critical data for strategic decision making. It uses machine learning techniques and tools in determining patterns and trends to gain actionable insights.
This paper selected a popular food brand to evaluate a given stream of customer comments on Twitter. Several metrics in classification and clustering of data were used for analysis. A Twitter API is used to collect twitter corpus and feed it to a Binary Tree classifier that will discover the polarity lexicon of English tweets, whether positive or negative. A k-means clustering technique is used to group together similar words in
tweets in order to discover certain business value. This paper attempts to discuss the technical and business perspectives of text mining analysis of Twitter data and recommends appropriate future opportunities in developing this emerging field.

Full text

Application of Text Classification and Clustering
of Twitter Data for Business Analytics
Alrence Santiago Halibas
Faculty of Computing Sciences
Gulf College
Muscat, Oman
alrence@gulfcollege.edu.om
Abubucker Samsudeen Shaffi
Faculty of Computing Sciences
Gulf College
Muscat, Oman
abobacker.shaffi@gulfcollege.edu.om
Mohamed Abdul Kader Varusai Mohamed
Faculty of Computing Sciences
Gulf College
Muscat, Oman
varusai@gulfcollege.edu.om
Abstract — In the recent years, social networks in business are
gaining unprecedented popularity because of their potential for
business grow th. Companies can know more about consumers’
sentiments towards their products and services, and use it to
better understand the market and improve their brand. Thus,
companies regularly reinvent their marketing strategies and
campaigns to fit consumers’ preferences. Social analysis
harnesses and utilizes the vast volume of data in social networks
to mine critical data for strategic decision making. It uses
machine learning techniques and tools in determining patterns
and trends to gain actionable insights.
This paper selected a popular food brand to evaluate a given
stream of customer comments on Twitter. Several metrics in
classification and clustering of data were used for analysis. A
Twitter API is used to collect twitter corpus and feed it to a
Binary Tree classifier that will discover the polarity lexicon of
English tweets, whether positive or negative. A k-means
clustering technique is used to group together similar words in
tweets in order to discover certain business value. This paper
attempts to discuss the technical and business perspectives of text
mining analysis of Twitter data and recommends appropriate
future opportunities in developing this emerging field.
Keywords— Twitter, Sentiment Analysis, Decision Tree, k-means,
Social Media
I.
I
NTRODUCTION
The social media has redefined the nature of how
companies strategize their business processes. The social media
contains a massive volume of unstructured data (e.g. tweets,
comments, blogs, forum discussions, user post, and reviews)
that can be used for business intelligence such as customer
profiling and content analytics. Twitter, which is a social
networking online service, is mainly used as a marketing and
promotion tool by most companies. Specifically, twitter data
contains not only user information, but also texts that contain
subjective information (such as user sentiments) towards a
particular issue. From a business perspective, the wealth of
tweets is enough for companies to gather sufficient feedback
about their products and services from their customers without
having to spend for costly customer surveys and interviews. On
the other hand, analyzing and extracting information from
unstructured data poses a formidable challenge to data miners.
Humans can easily find patterns and trends in documents but
this ability is limited when a large amount of data is involved.
However, with the help of analytical tools and techniques, such
challenge is achievable. Hence, the industry that surrounds
sentiment analysis is gaining momentum and currently the
focus of social media research [1]. In fact, [2] forecasted that
by 2022, the market for text analytics will rise to $8.79 billion
with a growth rate of 17.2%. The report described that the
increasing growth trend is attributed to the pressing need of
companies for social media analytics, predictive analytics and
the customization of their business applications. In a way,
companies have seen the benefits of text mining using
sentiment analysis in improving their services, monitoring of
brand or company reputation, gaining competitive advantage,
and other analytical uses with financial gains [3]. Its business
value cannot be ignored, especially by marketing and customer
support units of an organization.
This paper mainly examines the use of sentiment analysis
in business applications. Furthermore, this paper demonstrates
the text analysis process in reviewing the public opinion of
customers towards a certain brand and presents hidden
knowledge (e.g. customer and business insights) that can be
used for decision making after the text analysis is performed.
More so, [4] stressed that there is limited academic literature
surrounding text analytics of Twitter data, as a result, this paper
attempts to contribute in this developing field by providing a
practical guide on how to mine and analyze customers’ tweets.
This paper is divided into four (4) sections. Following this
section is Section II which provides Background Information
and Related Work of the following topics: Sentiment Analysis,
Decision Tree, and Text Clustering. Section III demonstrates
the Experimental Setup combined with the Results and
Discussion and Section IV presents the Conclusion and Future
Work.

II. BACKGROUND
AND
RELATED
WORK
A. Sentiment Analysis
Sentiment Analysis is used by companies to analyze data to
understand the users’ sentiments or opinions regarding their
products or services. According to [5], Sentiment Analysis is
“a process that automates the mining of attitude, opinions,
views, and emotions from text, speech, tweets, and database
sources through Natural Language Processing (NLP)”. It is
also referred to as opinion mining and emotion analysis. It uses
textual data to analytically collect, analyze, model and validate
for various business intelligence applications. Fig. 1 illustrates
different approaches and techniques to sentiment analysis.
.
Fig. 1 Sentiment Analysis, Source: [4, Fig. 1]
Several studies have pointed out the importance and
benefits of customer sentiment analysis for business operations.
Majority of the purposes are for business improvement and
decision support [7] [8]. In the study of [9] for instance, a
decision support system using sentiment analysis with data
mining was proven to be more efficient as opposed to the use
of qualitative and quantitative methods in customer satisfaction
research. The study pointed out that customer satisfaction is the
key factor that influences decision making, thus, it is important
to know customers’ sentiments to manage the quality.
Similarly, a case study conducted by [10] revealed a good use
of unstructured text analytics for product decisions. The study
stressed the significance of information triangulation from
different unstructured text sources, including newspaper
articles, to create confidence in the information for critical
decisions. Moreover, text analytics is used by the Information
Technology Services (ITS) of Florida State University to
capture the stream of incoming service requests across business
units [11]. Its success in extracting insights from the texts has
allowed the company to provide an effective IT service
management. It is said that when sentiments are positive, it is
likely that the customers are satisfied [12]. Therefore, knowing
how to positively exploit customers ‘data presents infinite
advantages for companies.
B. Text Classification using a Decision Tree
Text classification is an automated process of classifying
text in natural language to predefined categories [13]. It is best
used for predicting binary or nominal class labels [14]. A
plethora of learning algorithms are used for text classification,
including Naïve Bayes Classifier, Decision Trees, Support
Vector Machines, Neural networks, and many others. The
statistical approach in solving a classification problem involves
two learning methods: supervised learning method (as shown
in Fig. 2) which uses a training dataset to build the
classification model prior to application in a test set, and
unsupervised learning method which does not use any known
labels in mining the data. Both of these methods produce
quantitative evaluation results that can be easily reported.
A decision tree, which is of interest in this paper, is an
example of a supervised learning method. Alpaydin [15]
defines a decision tree as an efficient nonparametric method for
classification and prediction of data. Furthermore, he states that
it is a machine learning technique that uses a hierarchical data
structure. According to [16], decision trees learn and respond
quickly and interpretable, hence, they are normally used for
classification. Moreover, decision tree algorithms are greatly
utilized in data mining since they are proven to produce
rational classification models and good accuracy levels [17].
The experiment of [18] revealed that the Decision Tree
performed with 100% accuracy as opposed to Naïve Bayes’
86% accuracy performance. Because of a decision tree’s
simplicity to be understood and interpreted, and accuracy
performance, it has been chosen as the machine learning
technique for this experiment.
Fig. 2 Text Classification Building Model Approach, Source:
[11, Fig. 4.3]
C. Text Clustering
Clustering is one of the commonly used unsupervised
learning methods in analyzing the context of text data in
natural language form [19]. It is a mathematical approach in
collecting and segmenting similar documents into clusters. It
helps trim down the volume of unstructured text and provide a
simpler understanding and thematic structure of the data. It also
provides the keywords in each cluster that is useful in
extracting valuable insights, hence, customer sentiments can be
summarized using these keywords. The clustering process is
illustrated in Fig. 3.
K-means is a clustering technique in cluster analysis that
finds a user-specified number of clusters that are represented
by their centroids.

The basic k-means algorithm is described as:
1 Select k points as initial centroids
2 Repeat
3 Form k clusters by assigning each point to the closest centroid
4 Recompute the centroid of each cluster
5 Until centroids do not change
The algorithm iteratively assigns each point to one of the k
groups based on the specified features and the points are
clustered based on feature similarity [20]. Firstly, the points are
assigned to initial centroids. Each point is assigned to the
nearest centroid and the centroid of that cluster is updated by
computing the mean of all points that are assigned to the
centroid’ cluster. This process is repeated until a stop criterion
is satisfied.
Fig. 3 Text Clustering, Source: [3, Fig.4]
III. EXPERIMENTAL
SETUP
A. Software Specification
Rapidminer is a data-driven data mining tool that discovers
patterns in large collections of text [21]. It is an analytic
platform that integrates machine learning and predictive model
deployment used by data scientists. It contains rich libraries of
data science and machine learning algorithms [22]. The
software used in this experiment is the latest version -
RapidMiner Studio Educational version 8.0.001.
B. Hardware Specification
The experiment was developed using the following
hardware specifications:
Processor: Intel Celeron CPU N2830 @2.16GHz 2.16GHz
RAM: 4 GB (3.98 GB usable)
System Type: 64-bit Operating System
C. Text Classification Process Model
The text classification process involves many tasks such as
data extraction, data cleaning, pre-processing and feature
extraction, and classification [23]. The detailed process
diagram in Rapidminer is illustrated in Fig. 4. The processes
applied to this experiment are each described in the succeeding
sections.
Fig. 4 Text Classification Process in Rapidminer
1) Data Extraction
An access token, which is used to get authentication
access to extract data from the Twitter database, is given
upon logging into a Twitter account. Authorization details
that include the API key are required to establish the
connection and allow a search query. An example set of
Twitter is generated when setting the parameters for the
Search Twitter operator. In this experiment, a query
parameter is configured using the brand name of a popular
food brand. Other search parameters include returned tweet
and language limit of 500 English tweets, and the result
type of recent or popular tweets. This contains attribute and
label types including tweet ID, username, number of
retweets, original text, date and time it was created,
language, and many others. This example set is saved in an
MS Excel file.
2) Manual Data Cleaning and Sentiment Labelling
The returned example set is reviewed by human
labeler who was tasked to filter irrelevant data and retaining
those which are significant to the experiment. The labeler
was instructed to label the tweets as positive, negative, or
neutral with an accompanying guideline. Out of the 500
tweets, 352 neutral tweets were discarded since they serve
no purpose in the analysis. The remaining example set of
148 tweets (approximately 42%) which comprise the
training corpora. The sentiment labeling is manually
conducted by two (2) experts who classified each tweet as
either positive or negative. Additionally, the experiment
ensured that there is a balanced dataset, hence, it included
71 positively-labelled tweets and 76 negatively-labelled
tweets.
3) Feature Extraction
A preprocessing stage is necessary to clean (or
remove the “noise” in) the data in preparation to build the
model. Five operations were applied at the preprocessing

stage, namely Transform Cases, Tokenize, Filter
Operator, Filter Token, Stem Porter to extract useful
features of the data. The following describes each
operation in order of execution:
a) The Transform cases operator converts the characters
in a document to lowercase so that the words, “Hello” and
“hello” are the same.
b) The Tokenize operator splits the texts of a document
into a sequence of words (or tokens) by setting the non-
letters mode as splitting points.
c) The Filter Stopwords (English) operator removes the
unwanted words from a document such as is, are, the, of,
etc. These words are commonly used in the text but are
deemed useless when used because it provides no content
information.
d) The Generate n-Grams (Terms) operator creates n-
length of tokens in a document. The operator will check
words that frequently follow one another. In text analysis,
it is essential that tokens are grouped together in order to
extract more meaning. Single tokens such as “health”,
“living”, “summer”, and “breeze” may provide little
information as opposed to paired (or 2 gram) tokens, such
as “healthy_living” and “summer_breeze”. Thus, it will
be easier to understand the context of the associated
terms.
e) The Filter Tokens (by Length) operator removes
tokens based on their length. For instance, if the minimum
and maximum characters are set to 3 and 20, respectively,
then it will remove texts having less than 3 characters and
more than 20 characters in length.
f) The Stem Porter operator reduces the length of the
words until a minimum length is reached or when the
words are in its base form. For example, the words
“electricity” is shortened to electric”, “revival” is
shortened to “reviv”, “communism” is shortened to
“commun”, and so on. This operation makes the
comparison between words easier.
The result of the Process document operator is a list
of words with total and document occurrences. In this
case, the operation produces 1819 word list.
In between the Process Document and Split Validator
operators are two operators, namely Set Role and Select
Attributes operators. The Set Role is used to identify the
target attribute role (label). The Select Attributes selects
attributes with no missing values. The example set
contains 1819 regular attributes. Each attribute is listed
separately in columns. A Term Frequency-Inverse
Document Frequency (TF-IDF) score, which is the
weighting factor, is assigned to each attribute in order to
identify its significance to a document in a corpus. A TF-
IDF score of 0 indicates that the term did not appear in a
document, whereas a TF-IDF score of 0.617 means that
the term is significant as opposed to a TF-IDF score of
0.283.
4) Split Validation using Decision Tree
The Split Validation operator contains three (3) other
operators, namely Decision Tree, Apply Model and
Performance.
This experiment used supervised learning, which has
a training data set and a test data set, in training the
classifier to identify the polarity. This method is essential
in training the classifier on how to predict the polarity of
unknown Twitter data. The classification model is first
tested and built using the training dataset and the
classification accuracy is run using the test model.
Table 1 shows the performance vector for a decision
tree with varying split ratios. A split ratio that is set to
50%-50% means that 1/2 of the data is used in the training
while the remaining 1/2 of the data is used in the testing.
The table further shows that the accuracy is better for a
70%-30% split ratio as compared to the other two split
ratios, hence, this split ratio (with Accuracy of 70.45% and
Classification Error of 29.55%) is used in this experiment.
This experiment ensured that appropriate parameters that
can provide the best possible model to maximize the
prediction are selected.
TABLE 1 Performance Vector for Decision Tree
Split Ratio
50%-50% 70% - 30% 80% -20%
Accuracy/
Classification
Error
53.42% /
46.58%
70.45% /
29.55%
68.97% /
31.03%
5) Apply Model
The Apply Model operator applies a model on an
example set. In this experiment, the Decision Tree mode,
which was first trained, is applied as illustrated in Fig. 1.
A new example set of 1000 tweets was extracted, pre-
processed, and feed into the model to predict the polarity
of each tweet. This example set is processed to be
compatible with the model with the same attributes that
were used to generate the model.
6) Results and Discussion for Text Classification
There are several metrics that measure the
performance of the prediction. Some of the most widely
used metrics are accuracy, precision, and recall. These
metrics use four (4) prediction outcomes that are defined
in Table 2.
TABLE 2 Prediction Outcomes
Abbr Outcome Description
tp true positive No. of correct positive prediction
tn true negative No. of correct negative prediction
fp false positive No. of incorrect positive prediction
fn false negative No. of Incorrect negative prediction

Precision is computed as the number of true positives
divided by the total number of predicted positive while
Recall (also known as Sensitivity) is computed as the
number of true positive divided by the total number of
positives. Likewise, Accuracy is computed as the total
number of correct predictions divided by the total number
of the dataset. The formulas for accuracy, precision, and
recall are as follows:
Table 3 presents the confusion table that summarizes
the prediction performance of the binary classification
model.
TABLE 3 Confusion Matrix
True
Negative
True
Positive
Class
Precision
Predicted Negative 23 (tn) 13 (fn) 63.89%
Predicted Positive 0 (fp) 8 (tp) 100%
Class Recall 100% 38.10%
From the test data, the model correctly predicted 8
positive terms with no predicted error on false positives,
thus, garnering a precision of 100%. On the other hand,
the model correctly predicted 8 positive terms but
wrongly predicted 13 positive terms, thus garnering a
recall of 38%. The accuracy of the model is 70.45%.
Fig. 5 Tree Description
Fig. 5 provides the Decision Tree description. The
minimal leaf size is set at 2 (positive and negative). The
minimal gain ratio is set at 10% since the tree produced a
good representative of words given this value, otherwise,
this value will be decreased to let the tree grow. The tree
depth is set to a maximum depth of 15 in order to make it
readable and comprehensible
The attributes that contribute to the improvement of
the tree are shown in Fig. 5. For instance, the attributes:
love, meal, thank, old, drive, expect, card, date, eat, good,
look are highly significant to the label. Moreover, if the
TF-IDF score for the term “love” is greater than 0.075,
then the attribute is categorized as positive, otherwise, it
is negative. The TF-IDF score of each term are listed in
Fig.5.
TABLE 4 Prediction Classification Statistics
Table 4 shows that the model predicted
approximately 85% negative comments with a confidence
value of 0.605 as opposed to approximately 15% positive
comments with a confidence value of 0.395. This result
poses a highly significant issue to a company or brand
because it suggests a negative branding to the business.
With this kind of information, a company can exercise a
more informed critical decision. More so, there are
anecdotal evidences that support the correlation of
sentiment and brand reputation. In fact, [24] suggests that
it is essential to know the influence of Twitter data to a
brand reputation. In this case, a company can act upon
this serious issue immediately.
D. Text Clustering Process Model
The clustering process, which is implemented in
Rapidminer, is shown in Fig. 6. It includes data extraction, pre-
processing and feature extraction, and clustering. The
subsequent sections described the processes that are applied to
this experiment.
Fig. 6 Text Clustering
Nominal
Value
Absolute
Count
Fraction Confidence
Min Max Average Dev.
Negative 852 0.852 0 0.710 0.605 0.252
Positive 148 0.148 0.290 1 0.395 0.252

1) Data Extraction
The Read Excel operator was used to read the
example set from an Excel file.
2) Pre-processing and Feature Extraction
The aforementioned pre-processing and feature
extraction process is likewise applied to this experiment.
Similarly, the output of this process was feed to a Data-to-
Similarity operator which computes the similarity of each
example with every example in the given example set.
3) Clustering using k-means Algorithm
The clustering process uses the k-means operator.
Since clustering is categorized as an unsupervised
learning, it does not need a label attribute. K-means is
efficient even with multiple runs. This experiment
performed 10 repeated run times. It simply determines
the set of k clusters and assigns each example to a specific
cluster as shown in Fig. 7.
4) Results and Discussion for Text Clustering
In Fig. 7, the end process generated four (4) clusters
with corresponding items: Cluster 0 with 28 items,
Cluster 1 with 920 items, Cluster 2 with 33 items, and
Cluster 3 with 19 items. The words in each cluster
provide insights on what the cluster is all about. For
simplicity, Cluster 3 is chosen for evaluation due to its
reasonable number of items.
Fig. 7 Cluster Results
Table 5 lists the significant attributes (or words) in
Cluster 3. Note that some words were omitted (*) as it
reveal significant clues on the brand name that is used for
this experiment. This may flag ethical issues when
ignored.
TABLE 5 Cluster 3 attributes
boycott_* boycott_* boycott_sponsor cat
cat_meat dog dog_cat evil
evil_boycott * *_boycott meat
meat_trad sponsor trade trade_evil
winter boycott olympi olympi_dog
As seen in Table 5, only two negative words, such as
boycott and evil, are included in the list. Although less
significant as single words, when paired up with the other
words the insights ultimately change. Moreover, when
these insights are combined with the result in the text
classification having 85% negative classification
prediction, the theme of the insights will be strengthened.
Combining the words in Table 5 will create meaningful
information that can be used for critical decision making.
IV. CONCLUSION
AND
FUTURE
WORK
The outcome of evidence-based decision making
contributes to the improvement of a brand. Having a text
analysis of customer feedback and reviews allows effective
quality management. With sentiment analysis, companies can
now strategically reposition their businesses according to
customers’ sentiments.
This paper provided an introduction and rationale
behind the value of text analytics of Twitter data to businesses
in gaining customer views on products and services, and
brand. This paper also discussed several related work in
sentiment analysis for business applications. Importantly, it
demonstrated a practical application of text classification and
clustering of Twitter data, and revealed ways on how to
analyze these to gain business insights.
Although the classification accuracy rate for this
experiment is already acceptable in this application domain. It
is suggested that future work needs to increase the accuracy of
the classification model by improving data preparation and
experimenting with other classification algorithms.
Future work in this field can also be focused on real-
time analytics of Twitter data stream. Since there is a massive
amount of tweets collected daily, handling real-time analytics
is difficult. Therefore an automated sentiment analysis, which
runs in high processing and large memory computing
resources, is required.
R
EFERENCES
[1] B. Liu, “Sentiment Analysis and Opinion Mining,” Synth. Lect. Hum.
Lang. Technol., vol. 5, no. 1, pp. 1–167, 2012.
[2] marketsandmarkets.com, “Text Analytics Market by Component
(Software, Services), Application (Customer Experience Management,
Marketing Management, Governance, Risk and Compliance
Management), Deployment Model, Organization Size, Industry Vertical,
Region - Global Forecast to 20,” 2017.
[3] F. N. Ribeiro, M. Araújo, P. Gonçalves, M. André Gonçalves, and F.
Benevenuto, “SentiBench - a benchmark comparison of state-of-the-
practice sentiment analysis methods,” EPJ Data Sci., vol. 5, no. 1, 2016.

[4] A. Moreno and T. Redondo, “Text Analytics: the convergence of Big
Data and Artificial Intelligence,” Int. J. Interact. Multimed. Artif. Intell.,
vol. 3, no. 6, p. 57, 2016.
[5] V. A. Kharde and S. S. Sonawane, “Sentiment Analysis of Twitter Data:
A Survey of Techniques,” Int. J. Comput. Appl., vol. 139, no. 11, pp.
975–8887, 2016.
[6] W. Medhat, A. Hassan, and H. Korashy, “Sentiment analysis algorithms
and applications: A survey,” Ain Shams Eng. J., vol. 5, no. 4, pp. 1093–
1113, 2014.
[7] L. Ziora, “The sentiment analysis as a tool of business analytics in
contemporary organizations,” Stud. Ekon., pp. 234–241, 2016.
[8] S. Yaram, “Machine learning algorithms for document clustering and
fraud detection,” in Proceedings of the 2016 International Conference
on Data Science and Engineering, ICDSE 2016, 2017.
[9] N. Yussupova, M. Boyko, and D. Bogdanova, “A Decision Support
Approach based on Sentiment Analysis Combined with Data Mining for
Customer Satisfaction Research,” Int. J. Ad v. Intell. Syst., vol. 1&2,
2015.
[10] S. K. Markham, M. Kowolenko, and T. L. Michaelis, “Unstructured
Text Analytics to Support New Product Development Decisions,” Res.
Technol. Manag., vol. 58, no. 2, pp. 30–39, 2015.
[11] O. Muller, I. Junglas, S. Debortoli, and J. Von Brocke, “Using Text
Analytics to Derive Customer Service Management Benefits from
Unstructured Data,” MIS Q. Exec., vol. 15, no. 4, pp. 64–73, 2016.
[12] P. Khobragade and V. Jethani, “Sentiment Analysis of Movie Review,”
Int. J. Adv. Res. Comput. Sci., vol. 8, no. 5, p. 19411948, 2017.
[13] S. M. Kamruzzaman, F. Haider, and A. R. Hasan, “Text Classification
using Data Mining,” Science (80-. )., p. 19, 2010.
[14] T. Pang-Ning, M. Steinbach, and V. Kumar, Introduction to data
mining. 2006.
[15] E. Alpaydin, Introduction to Machine Learning. 2004.
[16] K. M. Sreerama, “Automatic Construction of Decision Trees from Data:
A Multi-Disciplinary Survey,” Data Min. Knowl. Discov., vol. 2, no. 4,
pp. 345–389, 1998.
[17] R. C. Barros, A. C. P. L. F. de Carvalho, and A. A. Freitas, Automatic
Design of Decision-Tree Induction Algorithms. Springer International
Publishing, 2015.
[18] A. Jain and P. Dandannavar, “text analytics framework using apache
spark and combination of lexical and maching learning techniques,” Int.
J. Bus. Anal. Intell., vol. 5, no. 1, pp. 36–42, 2017.
[19] M. Allahyari et al., “A Brief Survey of Text Mining: Classification,
Clustering and Extraction Techniques,” arXiv Prepr. arXiv, vol. 1707,
no. 2919, pp. 1–13, 2017.
[20] A. Trevino, “Introduction to K-means Clustering,” Datascience.com.
2016.
[21] M. Hofmann; and R. Klinkenberg;, “RapidMiner: Data Mining Use
Cases and Business Analytics Applications,” Zhurnal Eksp. i Teor. Fiz.,
2013.
[22] K. S. Rawat, “Comparative Analysis of Data Mining Techniques, Tools
and Maching Learning Algorithms for Efficient Data Analytics,” JOSR
J. Comput. Eng., vol. 19, no. 4, pp. 56–60, 2017.
[23] H. Kaur and V. Mangat, “Dictionary based Sentiment Analysis of
Hinglish text,” Int. J. Adv. Res. Comput. Sci., vol. 8, no. 5, pp. 816–822,
2017.
[24] M. H. Peetz, M. De Rijke, and R. Kaptein, “Estimating Reputation
Polarity on Microblog Posts,” Inf. Process. Manag., vol. 52, no. 2, pp.
193–216, 2016.