Performance analysis of the

PERFORMANCE ANALYSIS OF THE
RECENT ROLE OF OMSA APPROACHES IN
ONLINE SOCIAL NETWORKS
J. Ashok Kumar1, S. Abirami2, S. Murugappan3
Research Scholar1, Assistant Professor2, Associate Professor3
1, 2Department of Information Science and Technology,
Anna University, Chennai, India
jashokkumar83@gmail.com, abirami@annauniv.edu
3Department of Computer Science,
Tamil Nadu Open University, Chennai, India
drmryes@gmail.com
ABSTRACT
In this emerging trend, it is necessary to understand the recent developments taking place in the
field of opinion mining and sentiment analysis (OMSA) as part of text mining in social networks,
which plays an important role for decision making process to the organization or company,
Government and general public. In this paper, we present the recent role of OMSA in Social
Networks with different frameworks such as data collection process, text pre-processing,
classification algorithms, and performance evaluation results. The achieved accuracy level is
compared and shown for different frameworks. Finally, we conclude the present challenges and
future developments of OMSA.
KEYWORDS
Sentiment Analysis, Opinion Mining, Classification Algorithms, Social Media.
1. INTRODUCTION
Opinion Mining and Sentiment Analysis (OMSA) plays a vital role in social media to get positive
or negative sentiment and opinions expressed by the user’s or public using the mode of online
feedback forms, emails and OSN websites such as Facebook, Twitter, LinkedIn, YouTube,
MySpace, Blogs and forums etc. Shusen Zhou et al. [14] states that OSN sites are one of the most
important tools of the Web 2.0 to share or disseminate views. OMSA helps a lot to predict the
product sales, service, quality, policy initiatives, Institutions, forecasting political opinions, and
news contents for the company or organization, Government and general public. The main task of
OMSA is used to classifying the polarity at the document, sentence, or feature / aspect, and which
are expressed as positive, negative or neutral. The sentiment analysis research is also done at this
polarity level. The general system architecture of OMSA is constructed as shown Fig. 1, and the
main characteristics are analyzed like [1] educational data mining approach and reported
performances [12]. This paper is organized as follows. Section 2 presents the recent developments
in the field of OMSA with different frameworks and algorithms. Section 3 discusses the obtained
results by using datasets and its volume. Section 4 states the challenges and future developments.
Finally, Section 5 concludes the paper.
David C. Wyld et al. (Eds) : SAI, CDKP, ICAITA, NeCoM, SEAS, CMCA, ASUC, Signal - 2014
pp. 21–32, 2014. © CS & IT-CSCP 2014 DOI : 10.5121/csit.2014.41103

22 Computer Science & Information Technology (CS & IT)
2. RECENT DEVELOPMENTS IN OMSA
This section presents the recent development of OMSA approach with different frameworks,
methods, techniques, and algorithms. The main characteristics of OMSA approach is shown in
Table 1, which gives the complete or overall reference to the researchers and the process is
explained below. It is necessary to understand the present work to carry out future work without
duplication. First, the various framework and algorithms in opinion mining with data collection
approach, pre-processing stage and classification of polarity. Second, describes the various
framework and algorithms in sentiment analysis with data collection approach, pre-processing
stage and classification of polarity.
2.1 Twitter Opinion Mining (TOM) Frame work and Polarity Classification
Algorithm
Farhan Hassan Khan et al. [5] proposed a new TOM framework to predict the polarity of words
into positive or negative feelings in tweets, and to improve the accuracy level of this
classification. TOM framework is constructed into three stages. First, data acquisition process,
which is used to obtain the Twitter feeds with sparse features through Twitter streaming API from
OSN. Twitter4J library has been used to extract only English language tweets. Second, pre-processing,
which process each tweets individually for the refinement operations such as
detection and analysis of slangs/abbreviations, Lemmatization and correction, and stop words
removal. Then the refined tweets pass into the classifier. Third, Polarity Classification Algorithm
(PCA), it classifies the twitter feeds on basis of Enhanced Emoticon Classifier (EEC), Improved
Polarity Classifier (IPC), and SentiWordNet Classifier (SWNC) by using set of emoticons, a list
of positive and negative words, and SentiWordNet dictionary respectively. In this stage, reducing
the number of neutral tweets is the major issue. For this problem, final classification is performed
to indicate more accurate results than its predecessors based on the scores of EEC, IPC, and
SWNC.
2.2 Standard election prediction model by using User Influence factor
Malhar Anjaria et al. [10] introduced a model to predict the election result by applying the user
influence factor (re-tweets and each party garners) and extracting opinions using direct and
indirect feature on the basis of the supervised algorithms such as NB (simple probabilistic model),
MaxEnt (Uniform classification model), SVM (achieves maximum margin hyper plane), ANN
(feed forward network), and SVM with PCA (dimension reduction). This model is built into
several steps. Step1, data collection approach is used to collect tweets with Candidate’s name.
Step2, normalization and feature reduction includes the refinement operations Internet acronyms
and emoticons, duplicate tweets, candidate accounts, word expansion, URLs, repeated words and
repeated characters for to get original sentence format into usable format of tweets. Step3, feature
extraction and extended terms used for the purpose of unigram, bigram and a unigram + bigram.
Step4, machine learning methods, in which the polarity based classification applied to
independent features, and it fails to capture query specific sentiments so that the aforementioned
text classification algorithms used for direct features. Setp5, incorporating sentiment analysis
considers re-tweets into account for influence factor. In Step6, analyzing gender based votes
based on term frequency. Finally, US Presidential Election result 2012 and Karnataka State
Assembly Election 2013 results are shown that Twitter provides a reasonable accuracy.

Computer Science & Information Technology (CS & IT) 23
2.3 Similarity shaped membership function
Tapia-Rosero A et al. [15] employed a method to detect similarity shaped membership functions
in group decision making process. This method is constructed by using the symbolic notation,
similarity measure, and grouping membership functions by shape similarity. The symbolic
notation has two component algorithms to get shape-string and feature-string which represent a
sequence of symbols and sequence of linguistic terms respectively for each segment of
membership functions. The similarity measure used the linguistic terms from extremely short to
extremely long to obtain an overall similarity at unit interval. In a group decision making
environment, the grouping membership functions by shape similarity aims to gather groups using
similarity matrix.
2.4 Three-level similarity method (TLSM)
Jun ma et al. [8] stated a method to reduce the chance of applying inappropriate decisions in the
multi-criteria group decision making (MCGDM). In this aspect, a Gradual Aggregation
Algorithm (GAA) developed and established a TLSM. GAA faced two practical issues. First,
how to handle missing values. Second, how to generate a decision dynamically in MCGDM?. To
solve these issues, GAA is implemented in two ways, i.e., OGA (ordinary gradual algorithm) and
WGA (weighted gradual algorithm). The OGA does not explicitly process the criteria weights and
leaves it to the aggregation operator but the WGA does. TLSM will be measuring the similarity of
two participants opinion at three sequence levels, i.e., assessment level, criterion level, and
problem level. In Level-1, the term set will be divided into several semantic-equal groups for the
criterion and then used HCFSM. In Leve-2, identifies an appropriate SUF for each criterion at
PSA and PRW. In Level-3, each individual criterion provides a single perspective to observe
similarity of two opinions.
2.5 Unsupervised dependency analysis-based approach
Xiaolin Zheng et al. [17] presented an unsupervised dependency analysis-based approach to
extract AEP (Appraisal Expression Pattern) from reviews. The problem statement is defined at
different terminologies such as domain, aspect word, sentiment word, background word, and
review. Then the AEP is applied to represent the syntactic relationship between aspect and
sentiment words by using Shortest Dependency Path (SDP), Confidence score (CS) and
parameter inference. Finally, AEP-LDA (Latent Dirichlet Allocation) model is employed to
jointly identify the aspect and sentiment words. It outperforms the base line method when
compared to other supervised methods such as LDA, Local-LDA, Standard LDA, AEP-LDA (no-
AEP).
Pre-processing
Steps
Classification
Method (M) /
Technique
(T)
Result
Data
Collection
Process
Internet
Messages
Online Social
Networks
(OSN) Sites
Fig. 1 General System Architecture of OMSA approach

2.6 SuperedgeRank algorithm
Ning Ma et al. [11] introduced a SuperedgeRank algorithm to identify opinion leaders in online
public opinion supernetwork model. This model is built up into three stages. First, data processing
which identifies the main information from the public comments collected from internet and then
applies the ICTCLAS for splitting sentence into words. Second, online public opinion
supernetwork modeling includes four types of elements i.e., Social Subnetwork, Environment
Subnetwork, Psychological Subnetwork, and Viewpoint Subnetwork to form four layers of
supernetwork. Third, Indexes of online public opinion supernetwork includes Node Superdegree,
Superedge Degree to identify opinion leaders and Superedge-superedge Distance, and Superedge
Overlap to evaluate and verify the results. The aforesaid algorithm ranks the superedges in
supernetwork by using the indexes: the influential degree of information dissemination, the
transformation likelihood between different psychological types, and the similarity between
keywords of viewpoints.
2.7 Hybrid opinion mining framework for e-commerce application
Vinodhini G et al. [16] introduced two frameworks by the combination of classifiers with
principal component analysis (PCA) to reduce the dimension of feature set. First, PCA with
Bagging which is used to construct each member of the ensemble, and to predict the combination
over class labels. Second, Bayesian boosting model is employed using rapid miner tool. For data
pre-processing, a word vector representation of review sentences is created for the aforesaid
models. Finally, the results proved that the PCA is a suitable dimension reduction method for
bagged SVM and Bayesian boosting methods.
2.8 Opinion mining model for ontologies
Isidro Penalver-Martinez et al. [7] presented an innovative method called ontology based opinion
mining to improve the feature-based opinion mining by employing the ontologies in selection of
features and to provide a new vector analysis-based method for sentiment analysis. The
framework composed by four main modules namely, NLP module, Ontology-based feature
identification module, polarity identification module, and opinion mining module. In module1,
obtains the morphologic and syntactic structure of each sentence by including the pre-processing
and POS. In module2, extract the features from the opinions expressed by users. In module3,
provides the positive, negative and neutral values of nouns, adjectives and verbs. In module4, the
vector analysis enables an effective feature sentiment classification. Each feature is represented
using three coordinates. Finally, the results obtained in the movie review domain.
2.9 Sentiment extraction and change detection
Alvaro Ortigosa et al. [2] introduced a new method is called sentiment extraction and change
detection. The method includes the operations for extracting sentiments from texts are pre-processing,
segmentation into sentences, tokenization, emotion detection, interjection detection,
token score assignation (building the lexicon, removing the repetitive letters, spell checking),
syntactical analysis, polarity calculation, and for sentiment change detection are building the user
regular patter and then comparing weeks. In this aspect, a Facebook application is implemented in
SentBuk, which obtains the data from Facebook with the following permissions: Offline_access,
Read_stream, and User_about_me and using the user interface performs user regular pattern and
combining weeks for classification.

2.10 Semi-supervised Laplacian Eigenmap (SS-LE)
Kyoungok Kim et al. [9] presented SS-LE to reduce the dimensionality of the data points. The
experimental process is taken into text cleaning, text refinement, vectorization, applying PCA,
and applying SS-LE to reduce the dimension to 2 or 3. SS-LE constructs the graph by utilizing
label information and without label information. From this two graphs, graph laplacian matrices
are calculated separately, and then dimensionality reduction process used to minimize the
distance between two data points and its neighbors by weights.
2.11 Three ensemble methods with five learners in Facebook application
Gang Wang et al. [6] conducted the comparative assessment to measure the performance of three
ensemble methods i.e., Bagging, Boosting, and Random Subspace with five learners: NB which is
a simple probabilistic classification method, MaxEnt doesn’t make any assumptions in relations
between features, Decision Tree is a sequence model for a sequence of simple tests, K-Nearest
Neighbor classifies majority of its vote of its neighbors, and SVM has ability to model non-linearity.
For the above mentioned three ensemble methods, the base learners are constructed
from the training data set using random independent, weighted versions, and random subspaces of
the feature space respectively.
2.12 VIKOR and Sentiment Analysis framework
Daekook Kang et al. [4] presented a new framework in two stages by combining the VIKOR
approach and sentiment analysis for measurement of customer satisfaction in mobile services.
VIKOR is a compromising ranking method for MCDM. First, data collection and pre-processing
stages involves into the operations like preprocessing data from relevant website. Second,
compiling dictionaries of service attributes and sentiment words, it combines the dictionary of
attributes and dictionary of sentiment words, and then expressed in verb phrases, adjective
phrases and adverbial phrases for sentiment words into positive, negative and neural polarity, at
the last assigns score with WordNet. Third, the constructed keyword vectors of customer’s
opinions with reference to the dictionaries. Fourth, the customer satisfaction is measured with
respect to each service attribute. Finally, evaluates the customer satisfaction by considering all
attributes.
2.13 Fuzzy deep belief network
Shusen Zhou et al. [14] constructed a two step semi- supervised learning method for the
sentiment classification. In this sense, the general DBN is trained by using abundant unlabeled
and labeled reviews, design a fuzzy membership function for each class of reviews, and then
DBN maps each review into the output space for constructing the FDBN. In the first step, all
unlabeled and labeled reviews trained by using the general DBN and estimate the parameters
based on mapping results of all reviews. In second step, all unlabeled and labeled reviews trained
based on membership functions. Also, AFD proposed by combining active learning with FDBN
for labeling and uses them for training.
2.14 Construction of constrained domain-specific sentiment lexicon
Sheng Huang et al. [13] proposed an automatic construction strategy of domain specific sentiment
lexicon based on constrained label propagation. In this method, each steps presented sequentially
as follows. In step1, sentiment term extraction, which is used to detect and extract candidate term
from the corpus. In this step, nouns and noun phrases are expressing objective states, adjectives,

verbs and their phrases are used for reviewed objects, adverbs and their phrases are used to
enhance or weaken the adjectives and verbs opinions. In step2, sentiments seeds extraction, which
maintains consistent sentiment polarities across multiple domains and it extracted from semi-structured
format i.e., Title, Pros, Cons, and Text. It describes rated aspect, positive and negative
aspect (nouns and adjective-noun phrases) respectively. In step3, association similarity graph
construction, which constructs similarity graph to propagate sentiment information. In step4,
constraints definition and extraction focused two types of constraints called contextual constraint
and morphological constraint. In contextual constraint, coherence or incoherence relations are
used to maintain sentiment polarity directly or reversely. In morphological constraint, coherence
or incoherence relations between sentiment terms and also maintains sentiment polarity directly
or reversely. In step5, constraint propagation defines a matrix and encodes the direct and reverse
constraints into pair-wise constraints. In step6, constrained label propagation in which each
sentiment term receives sentiment information from its neighbors and retains its initial polarity
label.
Table 1. Main characteristics of OMSA approach published in 2014.

2.15 Ranked WordNet graph for sentiment polarity
Arturo Montejo-Raez et al. [3] employed a method for sentiment classification by using weights
of WordNet graph. In this method, the polarity of measurement consider in the interval [-1, 1] and
defines a function where values over zero refers positive polarity, values below zero refers
negative polarity and values to closer to zero refers neutral. This function is computed by
expanding senses and final estimation. Expanding senses intends to expand few concepts in order
to calculate the global polarity of the tweet by using the graph of WordNet according to random
walk algorithm. The final estimation i.e., final polarity score is evaluated by the combination of
SentiWordNet score and random walk weights.
3. EVALUATION RESULTS OF OMSA APPROACH
The OMSA approaches are demonstrated in particular resources with different datasets and its
volume as stated in Table 2. Also, the evaluated classification performance was analyzed by using
the feature selection process with the different types of metrics such as confusion matrices,
precision, recall and F-measure, etc., and their key findings in all the OMSA approach as shown
in Table 3. In this paper, we have used the Polarity Classification Algorithm (PCA) and
evaluation procedure to verify the accuracy for the above mentioned approach by using the
Sanders-Twitter Sentiment Corpus [18]. The corpus contained 5513 hand-classified tweets which
are focused on the topic of the companies (Apple, Google, Microsoft and Twitter) and products.
The tweet sentiments are labeled as positive, neutral, negative and irrelevant. The datasets count
is given in Table 4. These datasets has been measured with 43 trained tweets by using the
confusion matrices (Table 5), precision, recall, F-measure and accuracy. The results are shown in
Table 6 and Figure 2. In this analysis, the overall system accuracy is only considered and shown.
Table 4. Dataset count
Datasets No. of. Tweets
Apple 1313
Google 1381
Microsoft 1415
Twitter 1404
Table 5. Confusion matrix
P Q R S
P tpP ePQ ePR ePS
Q eQP tpQ eQR eQS
R eRP eRQ tpR eRS
S eSP eSQ eSR tpS
Precision P = tpP / (tpP+eQP+eRP+eSP), Recall P = tpP / (tpP+ePQ+ePR+ePS), F-measure = 2 x
(Precision x Recall) / (Precision + Recall), Accuracy = (True Positive + True Negative + True
Neutrals + True irrelevant) / (True Positive + False Positive + True Negative + False Negative +
True Neutrals + False Neutrals + True Irrelevant + False Irrelevant). Based on this results and
observations, the accuracy level of the OMSA approach remains the same for the resources.

Table 2. Results of OMSA approach with datasets and key findings

Table 6. Dataset calculation for confusion matrices, precision, recall and accuracy
Figure 2. Comparing precision, recall, F-measure and accuracy with four different datasets

4. DISCUSSED CHALLENGES AND FUTURE DEVELOPMENTS IN OMSA
The challenges and future developments are discussed below for the above presented frameworks
of OMSA approach published in 2014. It is one of the important tools to the aspirant researchers
to focus on new innovative idea. Farhan Hassan Khan et al. [5] indicated that TOM framework
faced challenges due to their short length and irregular structure of the content such as named
entity recognition, anaphora resolution, parsing, sarcasm, sparsity, abbreviations, poor spellings,
punctuation and grammar, incomplete sentences. Also, suggested to compare the proposed
algorithm with TweetFeel and Sentiment 140 for further improvement of the accuracy. Malhar
Anjaria et al. [10] mentioned that the presence of all entities in unbiased and equal manner was
the biggest challenge to provide the accuracy in the standard election prediction model. But, there
is a chance of increasing the accuracy level in future by including the more influential factors as
age, educational background, employment, economic criterion, rural and urban and social
development index.
Tapia-Rosero A et al. [15] discussed that the group decision making process might be difficult
under a supervision of mediator. This work could be extended to final objectives i.e., strategic
planning, suitability analysis, and applications like fuzzy control, fuzzy time series to find the
similarities. Jun ma et al. [8] indicated the major issues that to reduce the risk of putting an
inappropriate decision making and measuring opinion similarity between the participants. In
GAA, the integrating information according to group of inputs and missing value and unclear
answers need to be studied in future. Xiaolin Zheng et al. [17] stated that to jointly identify aspect
and sentiment word with comparison of other models. In future work, AEP-LDA model to
assume at single sentence and extend at clause level and into aspect-based review summarization,
sentiment classification, and personalized recommendation systems.
Ning Ma et al. [11] discussed that online public opinion controlled by deleting long existing posts
with rumor from negative opinion leaders. After identifying the opinion leaders, the further work
should be focused on how to implement corresponding guidance and interference. Vinodhini G et
al. [16] misclassification is reduced, and the classification accuracy of negative opinion to be
improved. Isidro Penalver-Martinez et al. [7] addressed the present challenges that ontology-based
feature identification, and feature polarity identification. Alvaro Ortigosa et al. [2]
demonstrated to extract information from user messages and detect emotional changes. Further,
tests to be conducted to determine the values in each case for sentiment changes, and the
threshold to distinguish between small changes on user sentiment and significant changes.
Kyoungok Kim et al. [9] addressed the dimensionality reduction transformations into 2D or 3D
by using term frequency matrices, and further work suggested that the weight of the edges in the
label graph will be adjusted by using the sophisticated approach and to transform document into
term frequency. Gang Wang et al. [6] evaluated the ensemble methods, and specified that large
datasets need to be collected for validating the result in future i.e., improving the interpretability
of ensembles is an important research direction. Daekook Kang et al. [4] indicated that the
measurement of the customer satisfaction was conducted by surveys. It’s taken more time and
effort to collect useful information. Further studies suggested that to incorporate more advanced
techniques of sentiment analysis and validate the empirical results presented.
Shusen Zhou et al. [14] discussed the issues that embed the fuzzy knowledge to improve the
performance of semi-supervise based sentiment classification. Sheng Huang et al. [13]
incorporated the contextual and morphological constraints between sentiment terms, and
suggested the future works that incorporate more types of constraints knowledge between
sentiment terms and to distinguish the aspect-specific polarities. Arturo Montejo-Raez et al. [3]

stated that how to deal with negation, and to study the context of a specific tweet among the time
line of tweets from the particular user in order to identify publisher’s mood and adjust final score.
Table 3. Performance Analysis of OMSA approach using feature extraction
Based on this observation, OMSA approach is not following any single algorithm or technique or
method for data collection process, pre-processing and classification, and to solve all issues in
social media for extracting the user’s opinion. Therefore, the problem is defined with some
specific application domain to estimate the polarity of the system. The OMSA approach is also
dealt with various models as shown in Table 1 such as descriptive, predictive and statistical for
the purpose of extracting opinions by using the applications of Fuzzy sets, grouping similarity
measure, group decision making process, policy creation process, OSN sites, and symbolic
notations. The analysis of result is very useful to overcome issues and to develop various new
methods or techniques without duplication.

5. CONCLUSION
Opinion mining and sentiment analysis is emerging as a challenging field as part of text mining in
social networks to an organization, Government and public. It has a lot of applications and
developments that to predict people’s polarity towards their decision making process. In this
paper, we conclude that the frameworks and algorithms of OMSA are presented with the data
collection process, preprocessing methods, classification and performance evaluation results as a
review.
REFERENCES
[1] Alejandro Pena-Ayala.: Educational data mining: A survey and a data mining-based analysis of recent
works. Expert Systems with Applications. 41, 1432-1462 (2014)
[2] Alvaro Ortigosa, Jose M. Martin, Rosa M. Carro.: Sentiment analysis in Facebook and its application
to e-learning. Computers in Human Behavior. 31, 527-541 (2014)
[3] Arturo Montejo-Raez, Eugenio Martinez-Camara, M. Teresa Martin-Valdivia, L. Alfonso Urena-
Lopez.: Ranked WordNet graph for Sentiment Polarity Classification in Twitter. Computer Speech
and Language. 28, 93-107 (2014)
[4] Daekook Kang, Yongtae Park.: Review-based measurement of customer satisfaction in mobile
service: Sentiment analysis and VIKOR approach. Expert Systems with Applications. 41, 1041-1050
(2014)
[5] Farhan Hassan Khan, Saba Bashir, Usman Qamar.: TOM: Twitter opinion mining framework using
hybrid classification scheme. Decision Support Systems. 57, 245-257 (2014)
[6] Gang Wang, Jianshan Sun, Jian Ma, Kaiquan Xu, Jibao Gu.: Sentiment Classification: The
contribution of ensemble learning. Decision support systems. 57, 77-93 (2014)
[7] Isidro Penalver-Martinez, Francisco Garcia-Sanchez, Rafael Valencia-Garcia, Miguel Angel
Rodriguez-Garcia, Valentin Moreno, Anabel Fraga, Jose Luis Sanchez-Cervantes.: Feature-based
opinion mining through ontologies. Expert Systems with Applications. 41, 5995-6008 (2014)
[8] Jun Ma, Jie Lu, Guangquan Zhang.: A three-level-similarity measuring method of participant
opinions in multiple-criteria group decision supports. Decision Support Systems. 59, 74-83 (2014)
[9] Kyoungok Kim, Jaewook Lee.: Sentiment visualization and classification via semi-supervised
nonlinear dimensionality reduction. Pattern Recognition. 47, 758-768 (2014)
[10] Malhar Anjaria & Ram Mohana Reddy Guddeti.: Influence factor based opinion mining of twitter
data using supervised learning. Sixth IEEE International conference on communication systems and
networks (COMSNETS). ISSN: 1409-5982, (2014)
[11] Ning Ma, Yijun Liu.: SuperedgeRank algorithm and its application in identifying opinion leader of
online public opinion supernetwork. Expert Systems with Applications. 41, 1357-1368 (2014)
[12] R.V. Vidhu Bhala, S. Abirami.: Trends in word sense disambiguation. Artificial Intelligence Review:
An International Science and Engineering Journal. DOI 10.1007/s10462-012-9331-5, Springer,
(2012)
[13] Sheng Huang, Zhendong Niu, Chongyang Shi.: Automatic construction of domain-specific sentiment
lexicon based on constrained label propagation. Knowledge-Based Systems. 56, 191-200 (2014)
[14] Shusen Zhou, Qingcai Chen, Xiaolong Wang.: Fuzzy deep belief networks for semi-supervised
sentiment classification. Neurocomputing. 131, 312-322 (2014)
[15] Tapia-Rosero A, A. Bronselaer, G. De Tre.: A method based on shape-similarity for detecting similar
opinions in group decision-making. Information Sciences. 258, 291-311 (2014)
[16] Vinodhini G, Chandrasekaran RM.: Measuring quality of hybrid opinion mining model for e-commerce
application. Measurement. 55, 101-109 (2014)
[17] Xiaolin Zheng, Zhen Lin, Xiaowei Wang, Kwei-Jay Lin, Meina Song.: Incorporating appraisal
expression patterns into topic modeling for aspect and sentiment word identification. Knowledge-
Based Systems. 61, 29-47 (2014)
[18] Niek J. Sanders.: Sanders-Twitter Sentiment Corpus, Version 2.0.1

Performance analysis of the

More Related Content

What's hot (20)

Similar to Performance analysis of the (20)

Recently uploaded (20)

Performance analysis of the