A fuzzy logic based on sentiment

International Journal of Data Mining & Knowledge Management Process (IJDKP) Vol.4, No.4, July 2014
DOI : 10.5121/ijdkp.2014.4403 27
A FUZZY LOGIC BASED ON SENTIMENT
CLASSIFICATION
J.I.Sheeba1
and Dr.K.Vivekanandan2
1
Assistant Professor,
Department of Computer Science & Engineering,
Pondicherry Engineering College, Puducherry, India
2
Professor,
Department of Computer Science & Engineering,
Pondicherry Engineering College, Puducherry, India
ABSTRACT
Sentiment classification aims to detect information such as opinions, explicit , implicit feelings expressed
in text. The most existing approaches are able to detect either explicit expressions or implicit expressions of
sentiments in the text separately. In this proposed framework it will detect both Implicit and Explicit
expressions available in the meeting transcripts. It will classify the Positive, Negative, Neutral words and
also identify the topic of the particular meeting transcripts by using fuzzy logic. This paper aims to add
some additional features for improving the classification method. The quality of the sentiment classification
is improved using proposed fuzzy logic framework .In this fuzzy logic it includes the features like Fuzzy
rules and Fuzzy C-means algorithm.The quality of the output is evaluated using the parameters such as
precision, recall, f-measure. Here Fuzzy C-means Clustering technique measured in terms of Purity and
Entropy. The data set was validated using 10-fold cross validation method and observed 95% confidence
interval between the accuracy values .Finally, the proposed fuzzy logic method produced more than 85 %
accurate results and error rate is very less compared to existing sentiment classification techniques.
KEYWORDS
Sentiment classification, Fuzzy logic, Fuzzy rules, Fuzzy C-means algorithm, Text mining
1. INTRODUCTION
Now a days, the immense amount of text data is available in the online, text mining has been
applied to discover hidden knowledge from the text in many applications and domains. Users
express their opinions about TV shows ,Programs ,Products or Services and they consume in blog
posts, shopping sites or review sites. For example in the business field, sentiment classification is
mainly used to find out consumers, sentiments and opinions about company’s products and
services and identifying new business opportunities. Manually discovering sentiments and
opinions from large volume of data is very difficult and also, another issue of this problem is to
extract the expressions like implicit or explicit from large volume of unstructured text data (Ex :
Meeting transcripts)[1].So it is needed to detect the expressions automatically from the
unstructured text data. Automatic classification of sentiment is important for several applications
like credibility analysis of news sites on the Web, survey of marketing report, recommendation
system ,mining online discussion, opinion mining and opinion summarization.
Sentiment analysis is one of the latest, highly active fields in Natural Language Processing. It is a
task of classifying sentiments according to their transcripts. Sentiment analysis is also known as

28
sentiment classification or opinion mining. It is a computational technique which seeks to
understand and explain expressions and sentiments by analyzing large amounts unstructured data
in such an efficient way as to assist in human decision making. It is essentially useful for all the
fields in day to day life. For example In politics, it can predict shift in public opinion regarding
election candidates and also it will give the suggestion to select the candidate.
On a daily basis life, it suggests the people to select electronic products like mobile, washing
machine, air conditioner and it will also give a better decision to see the movies or books to
read[1].Sentiment classification has been modelled as the problem of training a classifier using
reviews annotated for Positive, Negative and Neutral sentiment and it will express differently in
various domains[2].It is very useful for both the consumers as well as for the producers to know
what general public think about a particular product or service .The proposed framework has been
introduced for extracting sentiment classification from conversations. So that the users can get
the information about the conversation and also they can get the important keywords, topic and
expressions. Automatic sentiment classification is the task of classifying given transcripts with
respect to the sentiment expressed by the viewers of the conversations.
The most existing approaches are proposed for identifying only explicit expressions of sentiments
in the document. Generally, emotions are expressed directly for example “I like to see Sony
Television” in the transcripts. Detecting implicit expression from the transcripts is little bit hard,
the meanings appeared in the transcripts are indirectly like “ This Sony Television is costly but
sound system performance is less compared to other Televisions” and the system has to identify
these type of words, it would require the training dataset[3].In this proposed approach
incorporating features for both implicit and explicit expression detection with in a single
framework itself. This paper aims to identify the both expressions and topic of the particular
conversations and also to add some additional features for improving the sentiment extraction
method . Here, this method is performed by both Human transcripts and ASR transcripts and the
sentiments are extracted through MaxEnt and SVM classifier. Finally, the quality of the extracted
sentiments are improved using Fuzzy Logic technique.
2. RELATED WORKS
In Sentiment Classification method (Chenghua Lin, Yulan He 2011) proposed unsupervised
framework called Joint Sentiment-Topic (JST) model, it works based on Latent Dirichlet
Allocation (LDA). In JST model detects sentiment and topic simultaneously from text [4]. In this
JST model detects only explicit emotions are presented in the transcripts. It can not detect for
implicit emotions and neutral words from the transcripts.In another method(Giacomo
Inches,2011) proposed the ideas of author characterization, author identification and also
additionally identify the Topic of the particular document, using these methods it is easy to detect
the people who are involved in the conversation, find out author character and topic of particular
transcripts[5]. In this method detect author's characterization only and it can not support to detect
emotions present in the input. In (Yulan He Chenghua Lin,2011)proposed a modified JST model
by incorporating word polarity priors through modifying the topic-word Dirichlet priors. Here
polarity bearing topics are extracted by modified JST model[6].Another approach (Ahmed
abbasi,2011)proposed Feature Relation Network for Sentiment Classification, it will identify N-
gram based sentiments only. In this method, two new concepts like subsumption, parallel relation
introduced for reducing redundant words[7]. Using this approach, it can not supported for
identifying the topic of the particular input In (Alexandra Balahur ,2012) proposed to detect
implicit expression from text. In this method it will identify only seven implicit words which have
taken from EmotiNet dataset [3]. In this method, it will identify only seven implicit words which
have taken from EmotiNet dataset other words, it cannot detect. In another method (J.I.Sheeba
and Dr.K.Vivekanandan ,2012) proposed for improved sentiment classification from meeting

29
transcripts .They incorporate the features for both JST and FRN method and also included for the
author classification method [8]. In (Sheeba and Vivekanandan,2012) proposed a another method
to combine both keyword extraction and sentiment classification into a single model which will
perform both the works at a single time[15].but in this model it cannot detect the implicit words.
In (Long Thanh Ngo and Dinh Sinh Mai ,2012) introduced a method to improve the
computational efficiency of the interval type-2 Fuzzy C-Means Clustering(IT2-FCM) based on
GPU platform and applied it to land-cover classification from multi-spectral satellite image. In
this process the performance of GPU is more faster than CPU[16]. In (Nayana Mariya
Varghese,2012) proposed a cluster optimization methodology based on fuzzy logic which is used
for eliminating the occurrence of redundancies in data after the clustering was completed by the
web page mining methods. For Clustering Fuzzy C-Means algorithm is used in this method
[17].In (G. Tadayon Tabrizi,2012 ) proposed a new algorithm that avoids being trapped in local
peaks using fuzzy logic and PSO algorithm. It also finds a global optimal response or optimal
place of cluster centers[18]. In (Tushar,2009) introduced Fuzzy logic based approach for
determining the input-output relationship of some manufacturing processes. Here, fuzzy logic
controller is used to determine regression equations and also developed two types of clustering
algorithm namely entropy-based approach and fuzzy C-means algorithm[19].In (S.Selva
Kumar,2013) introduced a new technique mixed C-means clustering .This method is mainly
used to test against a brain tumour gene expression[20]. In (Maciej Piasecki,2013) proposed a
new tool WordNetLoom an application for WordNet development it mainly consists of two
methods namely a form-based and graph-based and it also discussed the role of the application
in WordNet development[21].Based on the above papers inferred fuzzy logic and wordnet
techniques and it was adopted for this proposed work.
Generally most existing approaches are detecting either explicit expressions or implicit
expressions of sentiments in the text. In the existing method it can identify only seven implicit
expressions which was available in the emotinet dataset . But in this proposed framework along
with emotinet dataset we can add word net dataset and general words also so that it can able to
detect more implicit words compare to existing method. In this proposed method is going to
detect both implicit and explicit expressions ,identify topic , author characterization and author
identification using fuzzy logic technique likes fuzzy rules and fuzzy C-means algorithm.
3. PROPOSED FRAMEWORK
In this proposed framework the process is done step by step for the best result in extraction of
sentiment classification . In this proposed framework along with emotinet data set , the word net
dataset has been used. So that it can detect any type of implicit emotions that are available in the
transcripts. The data whichever matches with emotions available in the data set will be listed.
Based on this list of words, it is concluded whether the Conversation is based on the Positive
conversation, Negative conversation or Neutral conversation. Here Convote debate dataset has
been taken as input. The data preprocessing has been implemented for this text file and the output
will be sent to input to the sentiment classification step. The sentiments are classified both
implicit and explicit expression of emotions using training dataset. The output of the explicit and
implicit emotions will be sent to the input of the fuzzy logic based sentiment classification for
reducing the emotions. Here ,the input will be taken from the output of the sentiment
classification step. The weights are assigned for each word in which the weights are calculated
from SentiWordNet. Based on the weight of the word, the “Threshold” value is calculated. . The
threshold value is calculated based on the average of each listed word. The list of positive,
negative and neutral words are listed which is greater than or equal to the Threshold value. This
method will returns the reduced emotions list. The output of the fuzzy emotions list will be given
to the input of clustering step.

30
Figure 1: Fuzzy logic based Sentiment classification using proposed framework
Here Fuzzy C-means algorithm used to cluster the similar words for reducing the emotions list
and it is able to group the emotions based on the cluster centroid. This framework also identify
the topic of the particular inputs. Finally this proposed method will return the reduced and
accurate emotions list.
The Figure 1 shows a proposed frame work for Sentiment Classification for given dataset. In this
task, 5 Steps have been included
1. Sentiment classification for both implicit and explicit emotions
2. Sentiment classification using Fuzzy logic method
3.To implement for fuzzy C-means algorithm
4. To detect Author classification(Identification, Characterization)
5. To identify the Topic .
3.1 Sentiment classification for both implicit and explicit emotions
Data Preprocessing:
Data preprocessing is mainly used for removing irrelevant and redundant information present or
noisy and unreliable data in the inputs. It includes cleaning, selection and feature extraction,
normalization, transformation etc. In this method, stop word removal process has been done. Stop
words are natural language words which have been filtered out after processing. Some search
engines do not record extremely common words in order to save space or to speed up searches.
As the list of stop words cannot be removed automatically, it must be trained by human input.
Examples of Stop words: A, About, Being, Can, Ever, of, the, You etc.

31
Sentiment Classification:
In Sentiment classification, the preprocessed data is given as an input to find the list of Positive
words, Negative words and Neutral words for both implicit and explicit emotions. The Neutral
words are common words. The emotions are detected by comparing each preprocessed data with
the list of positive, negative and neutral emotions that are available in the data set for both
implicit and explicit emotions .
In this existing work implicit emotions are compared with emotinet dataset, it consists only seven
words like “joy, sadness ,fear, anger, shame, disgust and guilt” and it is able to detect and
consider these words, except “joy” all the other words are negative related words in which it is
only concentrate on the negative part[3]. But in this proposed framework along with emotinet
data set , the word net dataset has been used. So that it can detect any type of implicit emotions
that are available in the datasets. The data whichever matches with emotions available in the data
set will be listed. Based on this list of words, it is concluded whether the debate is based on the
Positive conversation, Negative conversation or Neutral conversation.
3.2. Sentiment classification using Fuzzy logic method
Fuzzy Logic:
In this paper it is proposed for a new framework that is capable of answering to the needs of the
users in their process of retrieving the desired information when it is enormous ,heterogeneous,
vague, imprecise or not in order. Here fuzzy logic is used for retrieving and extracting emotions
from transcripts. Fuzzy logic is an ideal tool for the management of this kind of vague and
heterogeneous information[14].Fuzzy logic is a form of many-valued logic or probabilistic logic
and it deals with reasoning that is approximate rather than fixed and exact. Generally traditional
logic they can have varying values between true or false, but in the fuzzy logic variables may
have a truth value that ranges in degree between 0 and 1 and it has been extended to handle the
concept of partial truth and the truth value may range between completely true and completely
false. Here ,the input will be taken from the output of the sentiment classification step. The
weights are assigned for each word in which the weights are calculated from SentiWordNet.
Based on the weight of the word, the “Threshold” value is calculated. The threshold value is
calculated based on the average of each listed word. Finally, the list of positive, negative and
neutral words are listed which is greater than or equal to the Threshold value.
3.3. To implement for Fuzzy C-means algorithm
Cluster:
The Fuzzy C-Means algorithm is mainly used for clustering. A cluster brings together instances in
the data that share a common set of attributes .For each of these clusters a central value
representative of the cluster’s principal value is calculated .This is called centriod of the cluster.
An array of these centroids for a collection of data elements from a database provides the cluster
centers,thus telling us which set of rows in the database are closely related. This algorithm
depends on the selection of the initial cluster center and the initial membership value. It involves
two processes : the calculation of cluster centers and the assignment of points to these centers
using a form of Euclidean distance .This process is repeated until the cluster centers have
stabilized .The main goal of this algorithm is the assignment of data points into clusters with
varying degrees of membership. This membership reflects the degree to which the point is more
representative of one cluster than another[22]. In cluster, the output of the fuzzy logic step acts
as the input to this step .Initially, the number of clusters and cluster centroid values are randomly
selected. Next, each data element is computed with each cluster centroid to find which data

32
element belongs to which cluster. After computation, each cluster will have some data elements.
Next, the cluster centroid value is updated based on the data elements . This updated cluster
centroid value is the final output of the cluster. Here the fuzzy clustering and the data elements
can also belong to one or more cluster.Fuzzy C means algorithm works by assigning membership
to each data point corresponding to each cluster center ,on the basis of distance between the
cluster center and the data point. Most of the data is nearer to the cluster center and its
membership is towards the particular cluster center. Summation of membership of each data
point should be equal to one. This algorithm is used to cluster the similar words and it is able to
group the emotions based on the cluster centroid. Finally , this algorithm gives a reduced and
accurate emotions list.
3.4. To Detect Author classification(Identification, Characterization)
Author Classification
Author identification and Author Characterization are called author classification. To identify the
author, POS tagger is used here.
POS tagger(Qtag tool)
The tool reads text from datasets and for each token in the text returns the part-of-speech which is
used for both stop word removal and author identification.Author identification is used to identify
the people who are involved in the conversation. It helps to identify the authors and their
participants in the inputs. There are many authors involved in the conversations, here, the author
has been identified with the help of POS(Part of Speech ) concept.POS tag has been done for the
datasets using QTAG tool. It tags the each word’s POS. There are some conditions to identify the
authors
• The POS of the word must be NP (Proper Noun)
• Author name should be the first word of a sentence.
• Next of the word must be a semicolon (:)
If a particular word satisfies the above mentioned rules, the word will be selected as the name of
the author of the conversation. Author characterization is used to identify the characteristics of
each author and categorize the people involved in conversation. The sentiment of each author has
been identified for detecting the attitude of the author. Depending on the author identification, the
conversation is spitted as each author’s participation. Then the sentiment will be detected for
every author using SVM classifier. Classification concepts are used to group the authors
depending on their sentiment[5].In the Figure 2 shows the output of Author identification and
characterization from the conversation.

33
Figure 2:Author identification and characterization from the conversation.
3.5. To identify the topic
Topic Detection:
In Topic Detection, the preprocessed data is given as an input which is used to find the topic,
which the debated is based on. In topic detection, each word will assign a weight which is a
number of occurrences in the whole conversation. The word which has a maximum weight is
selected as a topic for that debate. If two or more words have the same maximum weight, then all
the words are listed as a topic separated by numbers which will be useful to select any one.
4. EXPERIMENTAL RESULTS
Data Set
In this paper, the domain meeting transcripts has been focused. Meeting speech is significantly
different from written text and other speech data. For example, in meeting transcripts, many
people can participate, even the discussions are not organized well and the speech is unplanned
one .The speech contains disfluencies and also the sentences are not constructed well .The people
who are involved in the meeting ,speak different pronunciations and they use different types of
words. Each people can act as different roles and topics in the transcripts. So extracting keywords
from meeting transcripts is difficult one compared to the documents[9][10].
Many people can involve in this conversation. The proposed system validated using data set from
Convote debate dataset .The text file is similar to meeting transcripts format. In Convote dataset(
http://www.Cs.Cornell.Edu/Home/Llee/Papers/Tpl-Convote.Home.Html) It contains 17,940
conversations . Here, randomly conversations have been chosen from the Corpus. By applying
this proposed framework sentiment classifications has been done from these Corpus. These
inputs are tested by using both Existing systems and Proposed systems. The data sets are tested
with the existing systems such as JST , FRN ,ISCF and DIEE Methods and the performance of
the proposed system is compared with the existing systems. The results are shown that the fuzzy
logic based proposed system provide the results are very accurate one.

34
Existing Systems
Joint Sentiment-Topic (JST) Model
Joint Sentiment-Topic (JST) model is mainly used to detect sentiment and topic simultaneously
from text. It is mainly based on Latent Dirichlet allocation (LDA) method. LDA is a generative
probabilistic model of a corpus which is widely used in document analysis. It models the
semantic relationships between words based on their co-occurrences in documents[23]. Some
sentiment classifications methods often fail to produce satisfactory performance when shifting to
other domains but JST makes it highly portable to other domains. The LDA model has three
hierarchical layers, where topics are associated with documents, and words are associated with
topics. But in the Joint Sentiment-Topic (JST) model, an additional sentiment layer has been
added between the document and the topic layer. JST is basically a four-layer model, here
sentiment labels are associated with documents and topics are associated with sentiment labels
and words are associated with both sentiment labels and topics[ 4].
Topic Extraction
JST extracts the topics and evaluates the effectiveness of topic sentiment captured by the model.
Topics extracted from dataset under positive and negative sentiment labels. Using SVM
classifier classifies the dataset and the data set compared with training data set for identifying the
topic. Topic identification using SVM classifier is done for the best accuracy from the
conversation[4].Figure 3 shows the result of Sentiment classification using JST model.
Figure 3:Sentiment Classification using JST Model
Feature Relation Network (FRN)
A Rule-based multivariate text feature selection method is called Feature Relation Network that
considers semantic information and also leverages the syntactic relationships between N-gram
features. Feature Relation Network is intended for the inclusion of extended sets of heterogeneous
N-gram features for enhanced sentiment classification .It includes the following steps for
classifying the sentiments like N-Gram Feature generation, Feature extraction, Feature selection
and Relation between features[7].Figure 4 shows the result of sentiment classification using FRN
Model.

35
Difference between JST and FRN Model
Using FRN the user can detect sentiment classification only. It can’t detect topic of the particular
transcripts. It detect the sentiment classification using N-gram based methods .But in JST model
the sentiments are detected by using Svm classifier and training dataset.
Figure 4 :Sentiment Classification using FRN Model
Improved Sentiment Classification Framework (ISCF)
This framework proposes a method for improved selection of text attributes for enhanced
sentiment classification. The result shows whether the given transcript is a positive, negative and
neutral oriented topic. Word sense disambiguation is also done which governs the process of
identifying sensible words used in a sentence, when the word has multiple meanings. Our
proposed framework aims to identify sentiment and topic and also author classification from
meeting transcripts. By improving its performance, this framework will incorporate the features
of both JST and FRN method . Author classification method mainly used to identify both Author
identification and Author characterization from the transcripts and sentiments, topic and author
classification are extracted using SVM classifier. This method classifies the author ,the sentiment
of the topic and then groups the author based on the result. It benefits to identify the members of
the meeting transcripts and their opinion .The results are proved in JST, most of the features
match with human annotated than FRN. When comparison is done between JST outperforms and
FRN. This operations are performed in a single framework. This method reduces time and space
complexity[8].
Detecting Implicit Expression using Emotinet (DIEE)
In this method implicit emotions are detected from text using Emotinet. Emotinet is like a
knowledge base which is based on commonsense knowledge stored. It is mainly used to captures
and stores emotional reaction to real-world situations. Here commonsense knowledge plays a
significant role in the affective interpretation .This method can identify only seven implicit
expressions like joy, fear, anger, sadness, disgust, shame, and guilt are available in the emotinet .
Among these words, except “joy” all the other words are negative related words[3].

36
Fuzzy Logic based Sentiment Classification Framework (FL-SCF)
Fuzzy logic can handle the problems with imprecise and incomplete data and it can model
nonlinear functions of arbitrary complexity [24].Fuzzy logic technique is introduced to improve
the accuracy of the keyword list and also it includes the benefits of simplicity and flexibility. The
proposed fuzzy logic method is mainly used to classify the sentiments both implicit and explicit
expression of emotion from the Dataset. In this proposed framework along with emotinet , the
word net dataset has been used. This method can able to detect any type of implicit emotions that
are available in the various datasets compare to existing method. The output of sentiment
classification like list of implicit and explicit emotions words weights are identified using
Sentiwordnet. The average of the weights are calculated .threshold weight are assigned .The
weight which is having greater then or equal to threshold value are considered as output of this
method. Again this output will be sent to the clustering step. Here Fuzzy C-means algorithm used
to reduce the emotions by applying cluster method. In this cluster method similar words are
grouped and it is able to group the emotions using cluster centroid. Based on this list of words, it
is concluded whether the input is based on the Positive conversation, Negative conversation or
Neutral conversation .It can able to find the accurate implicit and explicit emotions from the
sentiment classification .This method is able to find the Topic and also identify both Author
identification and Author characterization of the particular inputs. Finally the output will be
reduced two times namely fuzzy logic method and fuzzy C-means algorithm method.
Proposed Algorithm for Fuzzy based Sentiment Classification
Input: Convote debate dataset
//Data Preprocessing
1.Read Input Data
2.Collect Datasets D={D1,D2,…..Dn}
3. Remove Stop words from the inputs.
//Sentiment Classification
4. Read Preprocessed input.
5. Foreach (i = 1 to n) (n= number of inputs)
6. Seperate the words based on explicit, implicit and neutral database.
if input match with explicit database words
move into explicit group
else
if input match with implicit database words
move into implicit group
else

37
if input match with neutral database words
move into neutral group.
End if
End if
End if
7. end Foreach.
8. Repeat steps 5 to 7 for all inputs.
//Fuzzy based Sentiment classification
9. Read the sentiment classification output as input.
10. Each explicit, implicit database words assign weight from 1-10.
11. Foreach (i= 1 to n)
12. Find the average weight.
13. end Foreach.
14. Repeat steps 11 to 13 for each list of words in implicit and explicit.
15.Assign the threshold value and Calculate the words that is equal or greater than threshold for
each list.
16.Display the results
//Topic Detection
17. Read Preprocessed input.
18. Foreach (i= 1 to n)
19. Count no of occurrences of each word.
20. end Foreach.
21. Find the word has maximum count.
22. Display the word (topic)
23. If more then one word has the same maximum count then
display word one by one.
end if

38
// Fuzzy C-means Clustering
24. Read fuzzy output as input.
25. Initialize three clusters namely 4 ,7 ,10
26. Foreach (i=1 to n)
27.Cluster the words based on its associated weight
if the word weight is 5
move it to cluster group 4
end if
28. end Foreach
29.Repeat steps 25 to 28 for all words.
30.Recalculate the clusters name by the list of words it containing.
31.Find the average weight of the list (called centroid)
32.Recalculate the cluster based on centroid value.
33.Repeat steps 30 to 32 for all clusters.
34. Display the result
Output : Reduced the Sentiment classification list
In this algorithm the process is done step by step for the best result in extraction of sentiment
classification . Here Convote datasets has been taken as input. Steps 1 to 3 data preprocessing has
been implemented for this given text file. In step 4 the data preprocessing output will be sent
to input to the sentiment classification .In step6 the sentiments are classified like implicit and
explicit expression of positive ,negative and neutral emotions compared with training dataset.
In step 9 the output of sentiment classification will be sent to fuzzy based sentiment
classification step. In step 10 to 15 the weights are assigned for each word and the weights are
calculated from SentiWordNet. Based on the weight of the word, the “Threshold” value is
calculated. The threshold value is calculated based on the average of each listed word. Finally, the
list of positive, negative and neutral words are listed which is greater than or equal to the
Threshold value. In step 17 the pre processed data will be sent to the topic detection step. In step
19 count the number of occurrences of each word in the given input. The word which has a
maximum weight is selected as a topic for the given conversation. In step 23 , If two or more
words have the same maximum weight, then all the words are listed as a topic separated by
numbers which will be useful to select any one. In step 24 the fuzzy based sentiment output will
be sent to the fuzzy C-means clustering. In step 25 initialize the three clusters namely 4,7, 10. In
step 27 Cluster the words based on its associated weight .for example if the word weight is 5 then
the word is move it to cluster group 4.In step 30 assigning membership to each data point
corresponding to each cluster center ,on the basis of distance between the cluster center and the
data point. Most of the data is nearer to the cluster center and its membership is towards the
particular cluster center and finally summation of membership of each data point should be equal

39
to one. In step 32 cluster the similar words and it is able to group the emotions based on the
cluster centroid. Finally this method will given the reduced and accurate emotions list.
Metrics Considered for Evaluation
The performance of the proposed framework is measured in terms of the quality measures
namely Precision, Recall and F-Measure.
Precision
Precision is the fraction of retrieved emotions that are relevant
Precision = {Number of Relevant emotions} ∩{Number of Retrieved emotions}
{Number of Retrieved emotions}
Recall
Recall is the fraction of the emotions that are relevant to the query that are successfully retrieved.
Recall = {Number of Relevant emotions} ∩{Number of Retrieved emotions}
{Number of emotions Keywords}
F-Measure
F-Measure computes both precision and recall as the test to compute the score. Here precision is
the number of correct emotions divided by number of all returned emotions. Recall is the number
of correct emotions divided by the number of emotions.
Figure 5. Precision achieved

40
Figure 6. Recall achieved
It is observed from Figure 5 that the FL-SCF achieves the best Precision and from Figure 6 the
best Recall .It is also seen that the use of Fuzzy logic improves the performance of all the
techniques with SCF performing better than ISCF.
Figure 7. F-Measure
It is observed from Figure 7 that the FL-SCF achieves the best F-Measure. It is also seen that the
use of Fuzzy Logic improves performance of all the techniques with FL-SCF performing better
than other techniques.Figure 8 shows the Classification Accuracy obtained from the various
Techniques.

41
Figure 8. Classification Accuracy
It is observed from Figure 8 that the FL–SCF achieves the best Classification Accuracy .It is
also seen that the Fuzzy Logic based SCF method improves the performance compare with ISCF
Clustering is a technique of automatically grouping related data into clusters. A wide variety of
distance functions and similarity measures have been used for clustering, such as Euclidean
distance, cosine similarity, Jaccard distance, Pearson Correlation distance and Kullback –Leibler
Divergence. Fuzzy C means algorithm is operated in this process by assigning membership to
each data point which is corresponding to each cluster center from the basis of distance between
the cluster center and the data point. This proposed framework is used to compare , analyze and
evaluate the effectiveness of five similarity measure functions with Fuzzy C means clustering
approach. Here purity and entropy validity measures are used to estimate the optimal number of
clusters . The results are compared subsequently and shown that Euclidean similarity measure
function provides better and faster results as compared to the other distance functions with
Fuzzy C means clustering approach .Figure 9 shows the purity performance with different
techniques. Figure 10 shows Entropy performance with different techniques.
Figure 9: Purity values with different techniques using Convote debate dataset

42
Figure 10: Entropy values with different techniques using Convote debate dataset
Data set validation
The data set was validated using 10 fold cross validation method. Split the data into 10 samples .
Fit a model to the training for 9 samples and use the test 1 sample .Repeat the process for the
next sample, until all samples have been used to either train or test the model. This table 1 shows
the overall mean accuracy values for convote debate dataset. A confidence interval is a range
around a measurement that conveys how precise the measurement is .Table 2 shows the 10 fold
cross validation performed on 95% confident Interval on convote data set.
Table 1:Mean Accuracy values for different K values
Table 2:10 fold cross validation performed on convote dataset with 95%confidence interval.
In the above table 2.T value represents difference between mean values . Z.95 can be found using
t-distribution table based on df and α values. where Z.95 is the number of standard deviations
extending from the mean of a normal distribution required to contain 0.95 of the area. Df gives
degree is of freedom. P denotes null hypothesis. lower and upper values denotes 95% confidence

43
interval between mean accuracy values. The results obtained from this work obviously shows that
the proposed algorithms achieving better accuracy when trained with fuzzy logic .
5. CONCLUSION
This paper proposes a method for improved selection of text attributes for enhanced sentiment
classification. It shows whether the given conversation is a Positive or Negative or Neutral
oriented topic. The proposed framework is able to identify both implicit and explicit expression of
emotions and topic detection from the datasets using Fuzzy Logic technique and also applied for
the Fuzzy C-means clustering algorithm for clustering the words. Author identification and
characterization are also performed. Hence, this method classifies the author and the topic
sentiment and then it groups the author. To adopt more accuracy this framework has been
implemented. It is performed in a single framework, so it reduces time and space complexity. It
benefits to identify the members of the conversation and their opinions. It also enhances the
process of classification sentiments from the corpus which are compared to the existing systems.
Here Fuzzy logic balances the importance of all the features and also returns the accuracy of the
all important emotions from the inputs compared to existing methods. The data set was validated
using 10-fold cross validation method and observed 95% confidence interval .The proposed fuzzy
logic method produced more than 85 % accurate results and error rate is very less compared to
existing sentiment classification techniques.
REFERENCES
[1] Zied Kechaou , Mohamed Ben Ammar, (2011)“Improving elearning with sentiment analysis of users
opinions”, 2011 IEEE Global Engineering Education Conference (EDUCON) "Learning
Environments and Ecosystems in Engineering Education", Amman, Jordan, April 4 - 6,
2010,pp.1032-1038.
[2] Danushka Bollegala, David Weir,(2012) “Cross-Domain Sentiment Classification using a Sentiment
Sensitive Thesaurus”, IEEE transactions on knowledge and data engineering.
[3] Alexandra Balahur , Jesús M. Hermida,(2012) “Detecting implicit expressions of emotion in text: A
comparative analysis”, Decision support system ,Elsevier journal, pp.1-12.
[4] Chenghua Lin, Yulan He, Richard Everson,(2012) "Weakly-Supervised Joint Sentiment-Topic
Detection from Text," IEEE Transactions on Knowledge and Data Engineering, Vol. 24, No. 6, June
2012, pp .1134-1145.
[5] Giacomo Inches, Fabio Crestani, (2011)“Online Conversation Mining for Author Characterization
and Topic Identification” ,PIKM’11, October 28,ACM, pp. 19-26.
[6] Yulan He Chenghua Lin, Harith Alani,(2011) “Automatically Extracting Polarity-Bearing Topics for
Cross-Domain Sentiment Classification”. Proceedings of the 49th Annual Meeting of the Association
for Computational Linguistics, June 19-24, pp. 123–131.
[7] Ahmed Abbasi, Stephen France, Zhu Zhang, Hsinchun Chen,(2011) "Selecting Attributes for
Sentiment Classification Using Feature Relation Networks," IEEE Transactions on Knowledge and
Data Engineering, Mar. 2011,vol. 23, no.3, pp. 44-42.
[8] J.I.Sheeba , Dr.K.Vivekanandan,(2012) “Improved Sentiment Classification From Meeting
Transcripts”, IJCSI International Journal of Computer Science Issues, Vol. 9, Issue 5, No 3,
September 2012.
[9] Feifan Liu, Deana Pennell, Fei Liu,(2009) “Unsupervised Approaches for Automatic Keyword
Extraction Using Meeting Transcripts”, ACM .
[10] Fei Liu, Feifan Liu,(2011) “A Supervised Framework for Keyword Extraction From Meeting
Transcripts”, IEEE Transactions On Audio, Speech, And Language Processing, vol. 19, No. 3, March
2011,pp.538-548.
[11] http://www.support-vector.net.
[12] Veselin Stoyanov , Claire Cardie,(2008) “Topic Identification for Fine-Grained Opinion Analysis”,
International Conference on Computational Linguistics (Coling 2008), Manchester, August 2008, pp.
817–824.
[13] Eibe Frank ,Gordon W.Paynter, “Domain specific Keyphrase extraction”.

44
[14] Jorge Ropero, Ariel Gomez (2012)“ A Fuzzy logic intelligent agent for information extraction
introducing a new Fuzzy logic-based term weighting scheme”, Expert systems with
applications39(2012),Elsevier journal, pp. 4567-4581
[15] J. I. Sheeba , K. Vivekanandan,(2012) “Low Frequency Keyword and Keyphrase Extraction from
Meeting Transcripts with Sentiment Classification using Unsupervised Framework”, CCSEIT-2012,
October 26~28, Coimbatore, Tamilnadu, India. ACM .pp.212-216, ISBN 978-1-4503-1310-0.
[16] Long Thanh Ngo and Dinh Sinh Mai,(2012) “GPU-based Acceleration of Interval Type-2 Fuzzy C-
Means Clustering for Satellite Imagery Land-Cover Classification”, 12th International Conference on
Intelligent Systems Design and Applications (ISDA), IEEE, pp. 992-997.
[17] Nayana Mariya Varghese and Jomina John,(2012) “ Cluster Optimization for Enhanced Web Usage
Mining using Fuzzy Logic”, World Congress on Information and Communication
Technologies,IEEE, pp.948-952
[18] M. Mir and G. Tadayon Tabrizi ,(2012) “Improving Data Clustering Using Fuzzy Logic and PSO
Algorithm”, 20th Iranian Conference on Electrical Engineering, (ICEE2012), May 15-17,Tehran,
Iran. IEEE, pp .784-788
[19] Tushar,Dilip Kumar Pratihar,(2009)“Design of cluster-wise optimal fuzzy logic controller to model
input-output relationships of some manufacturing processes”, Int J.of Data Mining, Modelling and
Management,vol.1,no2 pp.178-205,DOI :10.1504/IJDMMM 2009.026075.
[20] S. Selva Kumar; H. Hannah Inbarani ,(2013)“Analysis of mixed C-means clustering approach for
brain tumour gene expression data”, Int. J. of Data Analysis Techniques and Strategies,Vol.5, No.2,
pp.214 – 228, DOI: 10.1504/IJDATS.2013.053682.
[21] Maciej Piasecki; Michał Marcińczuk; Radosław Ramocki; Marek Maziarz,(2013) “WordNetLoom: a
WordNet development system integrating form-based and graph-based perspectives”, Int. J. of Data
Mining, Modelling and Management, Vol.5, No.3, pp.210 – 232, DOI: 10.1504 / IJDMMM. 2013.
055861.
[22] Earl cox(2005)“Fuzzy modeling and Genetic algorithms for data mining and exploration”, Published
by Elsevier, Morgan Kaufmann Publishers. ISBN No :0-12-194275-9.
[23] Blei ,D.M.,NG, A.Y., And Jordan ,M.I,(2003) Latent Drichlet Allocation Math.Learn.Res.3,pp.993-
1022.
[24] Innocent P.R and John, R.I.(2004), “Computer Aided Fuzzy Medical Diagnosis”, Information
sciences ,Vol .162,No.2,pp.81-104.
BIBLIOGRAPHY
J.I.Sheeba received her B.E in Computer Science and Engineering from Bharathidasan University and M.E
in Computer Science and Engineering from Anna University. She is currently pursuing her Ph.D in
Computer Science and Engineering, from Pondicherry Engineering College affiliated to Pondicherry
University. Presently she is working as Assistant Professor in Department of Computer Science and
Engineering, Pondicherry Engineering College. Her research interest includes Data mining and Fuzzy
Logic.
Dr.K.Vivekanandan received his B.E from Bharathiyar University, M.Tech from Indian Institute of
Technology, Bombay and Ph.D from Pondicherry University. He has been the faculty of Department of
Computer Science and Engineering, Pondicherry Engineering College from 1992. Presently he is working
as Professor in the Department of Computer Science and Engineering. His research interest includes
Software Engineering, Object Oriented Systems, Information Security and Web Services. He has
coordinated two AICTE sponsored RPS projects on “Developing Product Line Architecture and
Components for e-Governance Applications of Indian Context” and “Development of a framework for
designing WDM Optical Network”.

A fuzzy logic based on sentiment

Recommended

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to A fuzzy logic based on sentiment (20)

Recently uploaded (20)

A fuzzy logic based on sentiment