Sentiment Analysis Using Lexical Approach and Fuzzy Logic

Sentiment Analysis Using Lexical Approach and Fuzzy Logic Renjith V. Ravi 1 ,S.B.Goyal 2( B ) , Xiao ShiXiao 2,3 , Mustafa Muwafak Alobaedy 2 , and Vladimir Kustov 4 1 Department of Electronics and Communication Engineering, M.E.A Engineering College, Malappuram, Kerala, India renjithravi@meaec.edu.in 2 City University, Petaling Jaya, Malaysia drsbgoyal@gmail.com, mustafa.theab@city.edu.my 3 Chengyi College Jimei University, Xiamen, China 4 Saint Petersburg Railway Transport University of Emperor Alexander I, Saint Petersburg, Russia Abstract.Understanding the feelings expressed in a statement is the goal of sentiment analysis. Depending on the thoughts expressed, a statement may be good, neutral, or negative. Positive, neutral, or negative feelings might be present in a statement. The reality is that the feeling expressed in each sentence whether it be positive, negative, or neutral is not always obvious. The examination of comparison, negation, intense, and sarcastic phrases, as well as the difficult work of dealing with grammatical errors, provide enormous hurdles to this process. The SA system is built in this study using a hybrid technique that combines dictionarybased approaches with fuzzy logic to address these issues. The suggested system’s outputs may be classified as either high positive, positive, neutral, negative, or high negative emotions. The analysis of customer satisfaction tweets about cloud services from Google, Amazon, and Microsoft revealed that the suggested method performs significantly, with results of 85.6% precision, 88.4% recall, and an 83.6% F-score. Keywords:Sentimental Analysis·Fuzzy Logic·SentiWordNet·SentiStrength 1 Introduction Socialmediahastransformedintoanothercontactroutebetweencustomersandorganisations. Users have a platform to post and express their feelings, opinions, and preferences regarding numerous subjects, people, goods, and services on social media platforms such as Facebook, Twitter, and Tumblr [1]. Normally, questions created by the researcher are used to gather text and reviews. This technique of data collection required a lot of time and was difficult to control [2]. With the development of the web and technology, people now use social media to post unstructured text reviews and comments. Views published on social media may be categorised to establish the orientation of submitted text (negative, positive, neutral). To establish the user’s opinion and emotion on a certain item or © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 S. Kadry and R. Prasath (Eds.): MIKE 2023, LNAI 13924, pp. 117–127, 2023. https://doi.org/10.1007/978-3-031-44084-7 _12 118R. V. Ravi et al. service, the sentiment intensity and strength of the post are assessed. Sentiment analysis is utilised to examine the content of both the text and the reviews [3]. Social networking sites must retain users to beat the competition [4]. Presenting and displaying items according to a client’s interests, alerting them to future sales, and suggesting new products that are similar to user preferences discovered via social networking sites are all ways to increase customer happiness [5]. The content of the social web is dynamic and constantly evolving to reflect the changing social and emotional states of its users. Sentiment analysis is a cutting-edge computer technique to enhance decision-making [6,7]. Views are always conveyed in free text as evaluations, judgements, or remarks. It bases its summary on quotes from the reviews that refer to pertinent items and their features or qualities. To determine the attitude of a person who wrote about a product or event, sentiment analysis can be utilised to mine these online remarks [8]. A method for handling imperfect and diverse information is fuzzy logic [9]. It is a kind of multi-valued logic that focuses on reasoning that is approximate rather than precise and fixed. Traditional logic often has values between true and false; however, fuzzy logic variables might have a truth value that is between 0 and 1. By creating membership functions, fuzzy logic is used to categorise the emotions. 2 Related Works Sentimental Analysis research is the subject of several new studies that are released annually. Regarding the creation, adaptation, or use of new sentimental lexicons [10–14], the use of traditional machine learning algorithms like nave Bayes [15,16], regression techniques [17,18], decision trees [17,18], clustering [18,19], ensemble classifiers and genetic algorithms [19,20], or transfer learning [21], various new contributions can be found. This study aims to organise the current understanding of the many uses of fuzzy logic in Sentimental Analysis. This study mainly focuses on the many activities that fall within the purview of sentiment analysis and other functions that rely on sentiment assessments to function. In each of these instances, fuzzy logic must act as the main approach used to carry out a crucial Sentimental Analysis-related activity, phase, or application in order to satisfy the criteria of this review; however, other techniques may also be used as part of the whole procedure or application. 3 Materials and Methods Using a fuzzy logic framework, the suggested method divides tweets into five groups by pulling out and combining different kinds of information [22]. “High negative,” “negative,” “neutral,” “positive,” and “high positive” are the four categories. The suggested technique is composed of four processing steps, as shown in Fig.1: data gathering, pre-processing, extraction of features, and model creation. Sentiment Analysis Using Lexical Approach and Fuzzy Logic119 Fig. 1.Proposed Approach for Sentimental Analysis 3.1 Features of Dataset The official Twitter accounts of Google, Microsoft, and Amazon provided the dataset for the suggested research project [22,23]. Users who are pleased with the web services that these organisations provide have posted 9421 tweets about them. Punctuation, stop words, URLs, digits, foreign words, and Twitter keywords are all eliminated during the pre-processing phase. Then, the content is tokenized. Then there are words that make no sense together, hashtags, emojis, negative words, adverbs, and intensive words. In feature extraction, theSentiWordNetdictionary polarity of words, theSentiStrengthdictionary polarity of words, the polarity of emojis, adverb, intensive, negation, and hashtag, exclamation, and retweets are all retrieved from the input text. The SentiWordNetphrase “polarity” is symbolised by a number in the range [−1–1]. It has a linear impact on the output outcomes. Because of this, an output with a polarity of -1 is more likely to fall under the category of “very negative.” The polarity corresponds to a result in the “high positive” class with a value of 1, and so on. The word “polarity” fromSentiStrengthalso refers to a value between−5 and 5. Moreover, it has a linear influence on the outcomes. The six groups of emoji are delight, grin, savouring, smirking, confuse, and wink. Each group has an impact on the output that may be favourable, negative, or neutral, as will be made clear throughout the model’s design. The intense words, such as intensive, adverb, exclamation, and negation, each have an impact on the output results that may be either good, negative, or neutral. The effect of a hashtag may be positive, negative, or neutral, while the influence of a retweet might be positive, negative, or neutral. The percentage of uplifting tweets to total tweet volume is used to calculate the effect of a hashtag. More than 60% of the tweets using the positive hashtag are positive. In contrast, the negative hashtag contains more than 120R. V. Ravi et al. or equal to 60% of tweets that are critical, with the remaining tweets being neutral. Likes and retweets can be positive, negative, or neutral. When a tweet with a polarity of positive words has 200 retweets or likes, it is considered positive. In comparison, a tweet with a polarity of negative words is considered negative if it receives more than 200 retweets or likes; otherwise, it is considered neutral. 3.2 Model Creation Model creation is done as a method of fuzzy logic for two reasons: the fuzzy impact of the input parameter on the output and the capacity of fuzzy systems to accommodate inputs of diverse types. As a result, the Takagi-Sugeno-Kang (TSK) fuzzy logic decision-making process is used as the model for classifying emotions in this research. TSK provides greater flexibility than necessary in the proposed emotional analytic system in comparison to other systems based on fuzzy logic. The proposed framework is created in four phases: initialization of parameters, fuzzification, evaluation of rules, and consolidation. In the parameter initialization, six parameters are selected: average polarity depending onSentiWordNet, word polarity depending onSentiStrength, emoji, intensive, hashtag, and retweet. The fuzzy set for each attribute is determined based on the words stated earlier. The membership functions are developed around the values given earlier and employ a trial-and-error technique in which the final sets and processes are finished as the experiments are done. The fuzzification method is accomplished based on the TSK system by turning the crisp value for inputs having crisp values, such as polarity and hashtags, into fuzzy words. The criteria used in the rule assessment stage are produced via a trial-and-error method. A set of basic rules is constructed, as presented in Table1. Nevertheless, the set gets refined as the trials are done. The aggregation procedure is based on using the rule’s output with the greatest confidence. Table 1.Rules for Fuzzy Controller P\NNegativeHigh NegativePositiveHigh Positive NegativeHNHNNNNN High NegativeHNPPHNNN PositiveNNNNHPPP High PositiveNNNNPPHP 3.3 SentiWordNet SentiWordNet[24] is a lexical resource for opinion mining and sentiment analysis. It is an extension of WordNet, a large lexical database of English language words that are grouped into sets of synonyms called synsets. InSentiWordNet, eachsynsetis assigned a score for three sentiment categories: positivity, negativity, and objectivity. These scores indicate the degree to which a word Sentiment Analysis Using Lexical Approach and Fuzzy Logic121 is associated with positive or negative sentiment, or if it is neutral and objective. The scores range from 0 to 1, with higher values indicating stronger sentiment. SentiWordNetis often used in natural language processing (NLP) applications such as sentiment analysis, opinion mining, and text classification. It provides a valuable resource for analyzing text data and extracting sentiment information that can be used to make more informed decisions in various fields, including marketing, politics, and customer service. The algorithm ofSentiWordNetis shown in Algorithm 1. Algorithm 1: Algorithm for SentiWordNet[9] 3.4 SentiStrength SentiStrength[25–27] is a sentiment analysis and natural language processing program that detects the sentiment of a given text on a scale of−5to+5. Mike Thelwall, Kevan Buckley, and Georgios Paltoglou created it at the University of Wolverhampton in the United Kingdom. SentiStrengthuses a mix of linguistic rules and machine learning to figure out the tone of a piece of text. It can look at both short messages like tweets and long ones like news articles or book reviews. The program is very good at recognizing sarcasm and other types of subtle emotions. SentiStrengthemploys the binary method of sentiment analysis, analysing the existence and intensity of positive and negative sentiment separately. A statement containing both positive and negative emotion, for example, will earn two different ratings, one for positive sentiment and one for negative sentiment. SentiStrengthis frequently utilized in academic and commercial research, notably in marketing, customer service, and social media analysis. It’s accessible as both a free web tool and a software package for local installation. 122R. V. Ravi et al. Algorithm 2: Algorithm of SentiStrength[9] 3.5 Fuzzification The first intitial phase in a fuzzy logic system is to find out what the inputs and outputs of the variables are. After the input variables and membership function are determined, the rule-base model (or matrix analysis of the fuzzy base of knowledge) must be comprised of expert IF-THEN rules [2,9]. These regulations convert the variables used as inputs into an output. This will show if the likelihood of operational issues is normally low, typical, or high. Figure2depicts the fuzzification technique. Fig. 2.Fuzzification Process 3.6 Membership Function Design The Membership Function (MF) converts the input values of the linguistic variables into the supplied fuzzy sets [28–30]. Appropriate membership functions translate the Sentiment Analysis Using Lexical Approach and Fuzzy Logic123 inputs into the degree of membership for the partitions of linguistic variables. The many membership roles are varied. The right membership functions must be used for fuzzy sets in order for a fuzzy system to accurately reflect fuzzy modelling. In the proposed study, the categorization labels for the degree of semantic orientation and the sentence’s positive or negative polarity are: •High Negative (HN) •Negative (NN) •Positive (PP) •High Positive (SP) The sentence could be classified as objective or subjective. So, let’s say it’s seen as subjective. If so, it may be either negative or positive, or it might be either very positive, highly positive, or highly negative [1]. By carefully examining the lexical scores offered inSentiWordNet, trapezoidal and triangular membership functions are applied for the relevant fuzzy words in this study. The MF can be defined using Eqs. (1) and (2) when the MF are trapezoids (in this case, ‘Strong’ or ‘Weak’). μ weak (trapezoid) ( x;q 0 ,q 1 ,q 2 ,q 3 ) =max ( min ( x−q 0 q 1 −q 0 ,1, q 3 −x q 3 −q 2 ) ,0 ) (1) μ strong (trapezoid) ( x;q 3 ,q 4 ,q 5 ,q 6 ) =max ( min ( x−q 3 q 4 −q 3 ,1, q 6 −x q 6 −q 5 ) ,0 ) (2) The MF may be specified in the equation - if the MF are triangles (in this instance, “Positive” or “Negative”) (3). μ Positive/Negative (triangle) ( x,q 2 ,q 3 ,q 4 ) =max ( min ( x−q 2 q 3 −q 2 , q 4 −x q 4 −q 3 ) ,0 ) (3) 3.7 Defuzzification A fuzzy quantity is defuzzified, or changed into a known amount. The area of the generated figures on every MF is determined in three phases [2,9]. Given that these areas are frequently triangles and trapezoids. If the membership degree isn’t one, the area won’t be in the shape of a triangle. Instead, it will be in the shape of a trapezium. The Algorithm 3 shows the rules for defuzzification: Algorithm 3: Rules for defuzzification[9] 124R. V. Ravi et al. 4 Results and Discussion As previously indicated, the information was gathered from the Twitter accounts of three different firms, and each message was individually annotated. In the studies, a total of 500 tweets, 100 in each category, were used. The outcomes of the suggested method are assessed using F-Score, recall, and precision. Table2contains the suggested approach’s confusion matrix, whereas Table3contains the F-Score, recall, and precision [28–30]. Table 2.Confusion Matrix of the Proposed Approach High PositivePositiveNeutralNegativeHigh Negative Neutral64952425 Negative002690 High Negative000065 Positive6963710 High Positive880000 Total100100100100100 Table 3.Values of F-Score, recall, and precision from the Proposed Approach CategoriesPrecisionRecallF-score High Positive88%100%93% Positive96%88%91% Neutral95%60%75% Negative78%96%80% High Negative71%98%79% Average85.6%88.4%83.6% The proposed approach has given an average precision value of 85.6, Recall of 88.4% and an F-Score of 83.6%. There was significantly reduced recall for the “positive” group (88%), and somewhat less recall for the “negative” category (96%). Also, the proposed approach had given a high recall value of 100% and 98% percentage in High Positive and High Negative category. There was somewhat less Recall for the “negative” group (96%), and slightly less for the “positive” category (88%). With just 55% of recall the “neutral” category, the suggested technique is seen as resilient since it was necessary to find both positive and negative comments when studying sentiment in this area in order to enhance the services. As a result, having poor “neutral” recall has no effect on how usable the suggested strategy is for SA of cloud service providers since it belongs to a less advantageous category. For all of the categories, the accuracy was good, except for “very negative,” where some of the “neutral” categories were mistakenly labelled as Sentiment Analysis Using Lexical Approach and Fuzzy Logic125 negative by the proposed method. This is seen as a problem with the technique that was suggested, and it will be taken into account in future research. 5 Conclusion In this essay, a novel approach to studying human emotions is proposed. It is based on a fuzzy logic system with many extracted characteristics. The suggested method, which can be categorized as a mix of machine learning and dictionary-based methods, combines the polarities of the emoji, hashtag, and intense words in fuzzy logic after extracting the polarities of the words in the input text using two dictionaries. As a result, the implementation of a fuzzy system involves three steps: fuzzification, rule evaluation, and aggregation. For each input, the system may produce one of the following five outputs: high positive, positive, neutral, negative, or high negative. Three wellknown cloud service providers, Amazon, Google, and Microsoft, were included in a dataset measuring consumers’ satisfaction with providers. The suggested approach’s performance on the gathered data revealed that it had an F-score of 83.6%, an precision of 85.6%, and a recall of 88.4%. The outcomes also showed that 100% of the time, the suggested method correctly remembered both “high positive” and “high negative” categories. The recall for the “negative” group was somewhat lower (96%), while the recall for the “positive” category was slightly lower (88%). Despite the poor recall of the “neutral” category (60%), the suggested technique is regarded as robust since it was necessary to discover negative and sometimes positive feedback when studying the mood of such a domain in order to enhance their services. As the less advantageous category in the suggested method, having a poor “neutral” recall has no effect on the usefulness of the proposed technique for SA of cloud service providers. All categories, with the exception of “strong negative,” had excellent results in terms of accuracy. Several of the “neutral” classes were given a negative label by the suggested technique, which is seen as a drawback and will be taken into account in further work. References 1. Vashishtha, S.: Design And Development of Fuzzy Logic Based Sentiment Analysis System for Online Reviews & Social Media Posts (2022) 2. Mary, A., Jothi, J., Arockiam, L.: A framework for aspect based sentiment analysis using fuzzy logic. ICTACT J. Soft Comput.8, 1611–1615 (2018) 3. Verma, B., Thakur, R.S.: Sentiment analysis using lexicon and machine learning-based approaches: a survey. In: Tiwari, B., Tiwari, V., Chandra Das, K., Kumar Mishra, D., Bansal, J.C. (eds.) Proceedings of International Conference on Recent Advancement on Computer and Communication, pp. 441–447. Springer Singapore, Singapore (2018).https://doi.org/10. 1007/978-981-10-8198-9_46 4. Rodr ́ıguez-Penagos, C., Grivolla, J., Codina-Filba, J.: A hybrid framework for scalable opinion mining in social media: detecting polarities and attitude targets. In: Proceedings of the Workshop on Semantic Analysis in Social Media (2012) 5. Appel, O., Chiclana, F., Carter, J., Fujita, H.: A hybrid approach to sentiment analysis. In: 2016 IEEE Congress on Evolutionary Computation (CEC) (2016) 126R. V. Ravi et al. 6. Wankhade, M., Rao, A.C.S., Kulkarni, C.: A survey on sentiment analysis methods, applications, and challenges. Artif. Intell. Rev.55, 5731–5780 (2022) 7. Ligthart, A., Catal, C., Tekinerdogan, B.: Systematic reviews in sentiment analysis: a tertiary study. Artif. Intell. Rev.54, 4997–5053 (2021) 8. Cui, J., Wang, Z., Ho, S.-B., Cambria, E.: Survey on sentiment analysis: evolution of research methods and topics. Artif. Intell. Rev.56, 8469–8510 (2023) 9. Mary, A.J.J., Arockiam, L.: ASFuL: aspect based sentiment summarization using fuzzy logic. In: 2017 International Conference on Algorithms, Methodology, Models and Applications in Emerging Technologies (ICAMMAET) (2017) 10. Xing, F.Z., Pallucchini, F., Cambria, E.: Cognitive-inspired domain adaptation of sentiment lexicons. Inform. Process: Manag.56, 554–564 (2019) 11. Bernabé-Moreno, J., Tejeda-Lorente, A., Herce-Zelaya, J., Porcel, C., Herrera-Viedma, E.: A context-aware embeddings supported method to extract a fuzzy sentiment polarity dictionary. Knowl.-Based Syst.190, 105236 (2020) 12. Wang, Y., Yin, F., Liu, J., Tosato, M.: Automatic construction of domain sentiment lexicon for semantic disambiguation. Multimed. Tools Appl.79, 22355–22373 (2020) 13. Ahmed, M., Chen, Q., Li, Z.: Constructing domain-dependent sentiment dictionary for sentiment analysis. Neural Comput. Appl.32, 14719–14732 (2020) 14. Mowlaei, M.E., Abadeh, M.S., Keshavarz, H.: Aspect-based sentiment analysis using adaptive aspect-based lexicons. Expert Syst. Appl.148, 113234 (2020) 15. Bi, J.-W., Liu, Y., Fan, Z.-P.: Representing sentiment analysis results of online reviews using interval type-2 fuzzy numbers and its application to product ranking. Inf. Sci.504, 293–307 (2019) 16. Li, Z., Li, R., Jin, G.: Sentiment analysis of Danmaku videos based on naïve bayes and sentiment dictionary. IEEE Access8, 75073–75084 (2020) 17. Bawa, V.S., Kumar, V.: Emotional sentiment analysis for a group of people based on transfer learning with a multi-modal system. Neural Comput. Appl.31, 9061–9072 (2018) 18. Al-Smadi, M., Al-Ayyoub, M., Jararweh, Y., Qawasmeh, O.: Enhancing Aspect-Based Sentiment Analysis of Arabic Hotels’ reviews using morphological, syntactic and semantic features. Inform. Process. Manag.56(2), 308–319 (2019).https://doi.org/10.1016/j.ipm.2018. 01.006 19. Riaz, S., Fatima, M., Kamran, M., Nisar, M.W.: Opinion mining on large scale data using sentiment analysis and k-means clustering. Clust. Comput.22, 7149–7164 (2017) 20. López, M., Valdivia, A., Martínez-Cámara, E., Victoria Luzón, M., Herrera, F.: E2SAM: evolutionary ensemble of sentiment analysis methods for domain adaptation. Inform. Sci. 480, 273–286 (2019) 21. Saad, S.E., Yang, J.: Twitter sentiment analysis based on ordinal regression. IEEE Access7, 163677–163685 (2019) 22. Alharbi, J.R., Alhalabi, W.S.: Hybrid approach for sentiment analysis of twitter posts using a dictionary-based approach and fuzzy logic methods. Int. J. Semant. Web Inf. Syst.16, 116–145 (2020) 23. Alharbi, J.R., Alhalabi, W.S.: Sentimental analysis using fuzzy logic for cloud service feedback evaluation. Int. J. Inform. Comput. Technol.8, 1–10 (2018) 24. Husnain, M., Missen, M.M.S., Akhtar, N., Coustaty, M., Mumtaz, S., Prasath, V.S.: A systematic study on the role of SentiWordNet in opinion mining. Front. Comp. Sci.15, 154614 (2021) 25. Gouthami, S., Hegde, N.P.: Automatic sentiment analysis scalability prediction for information extraction using sentistrength algorithm. In: Brahmananda Reddy, A., Nagini, S., Balas, V.E., Srujan Raju, K. (eds.) Proceedings of Third International Conference on Advances in Computer Engineering and Communication Systems: ICACECS 2022, pp. 21–30. Springer Nature Singapore, Singapore (2023).https://doi.org/10.1007/978-981-19-9228-5_3 Sentiment Analysis Using Lexical Approach and Fuzzy Logic127 26. Khaira, U., Johanda, R., Utomo, P.E.P., Suratno, T.: Sentiment analysis of cyberbullying on twitter using SentiStrength. Indonesian J. Artif. Intell. Data Mining3, 21–27 (2020) 27. Sari, S., Khaira, U., Pradita, P.E.P.U., Tri, T.S.: Analisis sentimen terhadap komentar beauty shaming di media sosial twitter menggunakan algoritma sentistrength: sentiment analysis against beauty shaming comments on twitter social media using sentistrength algorithm. Indonesian J. Inform. Res. Softw. Eng.1, 71–78 (2021) 28. Jefferson, C., Liu, H., Cocea, M.: Fuzzy approach for sentiment analysis. In: 2017 IEEE international conference on fuzzy systems (FUZZ-IEEE) (2017) 29. Wang, Y., Subhan, F., Shamshirband, S., Asghar, M.Z., Ullah, I., Habib, A.: Fuzzy-based sentiment analysis system for analyzing student feedback and satisfaction. Comput. Mater. Continua62, 631–655 (2020) 30. Haque, M., et al.: Sentiment analysis by using fuzzy logic. arXiv preprintarXiv:1403.3185 (2014)