Sentiment Analysis is a difficult problem, where the users can freely express their opinions and feelings on different facility and topics
In
A fuzzy system is a powerful approach for modeling systems, which is introduced by (Zadeh, 1965). The basis of fuzzy logic is to consider the inputs and outputs of the system as fuzzy sets and build the fuzzy rules. Each fuzzy set contains elements that have varying degrees of membership
In this paper the following technique was used, firstly, the text was pre-processed to get useful information by removing the unnecessary information. Because the students post additional information when express toword there's sentiment by using abbreviations, URLs, emoticons and hashtags. Secndly, the fuzzy Logic with two lexicans (SentiWordNet and AFINN lexicons beside the confirmation parameter as inputs) to process the students's comments was applied. The used technique was applied in future university in Yemen.
In this section, the details of the proposed fuzzy logic algorithm were presenting to analyze the sentiment of the Students toward the university.
Based on the polarity of the words, which can be positive or negative, the outputs are assigned either (positive, very positive, natural, negative, and very negative), however, polarities of the output are increased or decreased with the presence of adverbs (In this work, we express about this as input to our proposed system as confirmation parameter), such as very, not, absolutely and never and are reversed with the present of negative words. To build this form, a set of rules is developed to evaluate input and produce proper output. The results show the satisfactory performance of the proposed algorithm and more accuarcy with respect to comparing methods.
Users always post information that depicts sentiment, besides that; they post additional information using abbreviations, emoticons, slang, or URLs. Thus, the information which they posted needs to be pre-processed to get useful information by removing the additional data. So, we have eliminated all of the additional data which they don’t carry any sentiment.
A sentiment lexicon is a list of words which are generally labeled according to their semantic orientation as either positive or negative. In the article, two different sentiment lexicons have been used beside to construct our own Opinion Words Lexicon: SentiWordNet
In the SentiWordNet three numerical scores relating to positive, negative, and neutral were annotated. Each score ranges from 0 to 1, and their sum is 1. For example, in-text "I am very impressed from her thinking method", the word "very" gives an amount of assurance, and the word "impressed" is an opinion word. The SentiWordNet method obtains the scores of each word from this lexicon and each word has positive and negative scores these scores are computed as the following:
The AFINN lexicon is a list of English terms rated for valence with an integer between -5 negative and + 5 positive. The AFINN method fetches the score of each word, if it is greater than 0 it is a positive word, and if less than 0 it is a negative word.
To build the fuzzy model for Analyzing Student sentiment the popularly Mamdani fuzzy model was used. The Mamdani style fuzzy inference process is performed in four steps: Fuzzification of input variables, Rule evaluation (inference), Aggregation of the rule outputs, and Defuzzification.
The inputs of the fuzzy model are Confirmation, AFINN score; theSentiWordNetpositive and negative score of each post obtained from the above phase is fuzzified using Pi-shaped membership function. When the Pi-shaped membership function is used, each linguistic term T involves four key points, x, a, b, c, d associated with the change of pattern of the fuzzy membership. The Pi-shaped membership functions of fuzzy model inputs were graphically presented in
The fuzzy rules for Analyzing Student Satisfaction were described in
Experts consider each variable as a linguistic variable (LV), define a set <αi, T(αi), Xi, Gi, Mi>, i = 1 , n , where αi– name of LV; T(αi) - term-set LV αi; Xi – domain LV αi , Gi - syntactic rule; Mi - semantic rule
In this work, the selected parameters will be determined at the verbal level by the following linguistic variables (LV):
- LP1 – the SentiWordNet positive score has base-sets Т(LP1)={
- LP2 – the SentiWordNet negative score has base-sets Т(LP2)={
- LP3 – AFINN score has base-sets Т(LP3)={
- LP4 – Confirmation has base-sets Т(LP4)={
Rule |
AFINN score |
SentiWordNetpositive |
SentiWordNetnegative |
Confirmation |
Sentiment |
|
Positive |
HighPos |
LowNeg |
Pos |
very positive |
|
Positive |
MediumPos |
LowNeg |
Pos |
very positive |
|
Positive |
LowPos |
LowNeg |
Pos |
Positive |
|
Positive |
HighPos |
LowNeg |
Neg |
Negative |
5 |
Negative |
LowPos |
HighNeg |
Pos |
VeryNegative |
|
Negative |
LowPos |
MediumNeg |
Pos |
VeryNegative |
|
Natural |
LowPos |
LowNeg |
|
Natural |
|
|
…. |
…. |
|
…… |
Finally, after the inputs and output are determined, the fuzzy rules are builded. The program to analyze the sentiment of Students toward the university is created as shown in
The user enters the Sentences which express about students’ sentiment toward the university and then the program analyze these sentiments and classify it.
The program results for sentiment analysis should be evaluated by Precision criteria. Therefore, in this case, the application of expert assessments is desirable and gives more reliable estimates.
|
|
Our work |
0.891 |
Support Vector Machines (SVM) |
0.694 |
Naïve Bayes |
0.456 |
Tthe Fuzzy Inference to analyze the sentiment with their own lexicon Opinion Words Lexicon and classified the sentiment into 2 class- positive, negative |
0.721 |
The Fuzzy Inference to analyze the sentiment with only SentiWordNet lexicon |
0.788 |
As shown in
This study proposes a fuzzy model for Sentiment Analysis of Student Satisfaction toward the Future University in Yemen. The novelty is the formulation of few fuzzy rules to evaluate the sentiment class of tweets, and the proposed model can be adapted to any lexicon. The proposed fuzzy model was implemented using two different lexicons SentiWordNet, AFINN as inputs beside the confirmation input. Comparison with other methods (Support Vector Machines (SVM), Naïve Bayes , Tthe Fuzzy Inference to analyze the sentiment with their own lexicon and The Fuzzy Inference to analyze the sentiment with only SentiWordNet lexicon) reveals that the proposed fuzzy rule-based model performs consistently the best.
This work employed the fuzzy rules due to the fact that the Fuzzy systems can deal with ambiguity. The fuzzy rules incorporate the fuzziness of positive and negative scores and deal with reasoning and give closer views to the exact sentiment values.
In the future, this work can be implemented in other domains like evaluation the consumer satisfaction with products, etc. this study can extend by incorporating fuzzy inferencing into deep neural network models.