264 Sentimental Analysis – A Survey of Some Existing Studies Prabakaran Thangavel, Ravi Lourduswamy, Sacred Heart College (Autonomous), Tirupattur(dt), Tamilnadu – 635 601, India. Abstract Sentimental Analysis is a process of computing and categorizing the expressed opinions of people about certain event, subject or product as positive, negative or neutral. The major objective of Sentimental Analysis is to help data-driven decisions using insights from replies in social media, surveys and product reviews. Sentiment Analysis can be done with words, sentences, documents, features or aspects, concepts, phrases, links, clauses and implications. Recently, there has been a lot of attention on sentiment analysis especially from researchers in the fields of text mining and natural language processing. But due to extreme absence of annotated datasets which are used to train models in various domains, the accuracy of sentiment analysis has been hindered. Many types of research have been done to confront the challenge and enhance sentiment analysis classification. Sentiment analysis is important as it helps in identifying the emotional and attitude states of people. Positive or negative feelings of people can be expressed in different ways. This research article talks about, in subtle terms, the different ways to deal with sentiment analysis mostly in Machine Learning, Lexicon-based, Hybrid and Ontology-based approaches. This research article gives point by point perspective of the distinctive applications and challenges of Sentiment Analysis. Keywords 1 Sentiment Analysis, Machine Learning, Lexicon-based, Corpus-based, Hybrid and Ontology 1. Introduction can be used likewise for opinion mining and Sentiment Analysis. Feature level classification Sentimental Analysis is a process of is done by identifying and extracting the determining whether the text, document and different product features from raw source data. media content as positive, negative or neutral. The feature analysis is done when a desired The foremost objective of Sentimental Analysis sentimental aspect or feature is to be got from a is to help data-driven decisions using review. This article is structured as follows: perceptions from responses in social media, section-2.presents background information surveys and product reviews. Sentiment related to the survey. Section-3.presents the Analysis can be done with words, sentences, related works conducted on various aspects of documents, features or aspects, concepts, sentimental analysis. Section-4.presents the phrases, links, clauses and implications. The methodology of the survey conducted. Detailed bag of words recovered from word cloud is used discussion on open issues and challenges of for word-level Sentiment Analysis. At the sentiment analysis is presented in section-5. sentence level Sentiment Analysis is done with Finally, this article is concluded in section-6. sentences in reviews and comments written by users. A document-level analysis is done by classifying the opinions expressed in an entire document into different sentiments. Features ISIC’21: International Semantic Conference, February 25-27, 2021, New Delhi, India EMAIL: prabagaran@shctpt.edu (Prabakaran Thangavel) ravi@shctpt.edu (Ravi Lourduswamy) ©️ 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). CEUR Workshop Proceedings (CEUR-WS.org) 265 2. Background same measurement. For content classification, SVM has been proved to be useful for viable learning calculations. 2.1 Machine Learning Approaches Machine Learning (ML) techniques have a Rule-based Technique When an ‘if-then’ variety of features that help in the construction rule is applied to a relationship that consists of of classifiers for the expressed sentiments in an antecedent and its corresponding texts. ML approaches make use of ML consequent, then it is a Rule-based Technique. algorithms for Sentimental Analysis to extract the linguistic and syntactic features and form Antecedent ->consequent standard classifications of those features. An antecedent pronounces a state and can be Another popular method used for sentimental symbolized as a single token or several tokens analysis is Deep-Learning which makes use of that are connected by the “∧” operator. A token neural network techniques. can be either “?” which symbolizes a proper Supervised learning: Supervised learning noun or a word or the token can be “#” that is a technique that makes use of labelled denotes the consequence of the state described datasets to create a model. A variety of by the antecedent. supervised models can be created depending on {token1 ∧ token2 ∧ ... ∧ token} ⇒ {+|−} the type of procedure that is employed to create the model. The three simple rules A, B and C relates to words describing three sentiments, each Decision tree classifiers: Decision tree denoting an antecedent. classifiers prepare the information space in different levels of decreasing order that {Good} ⇒ {Positive}… … (A) distinguishes data based on the estimation of {Bad} ⇒ {Negative}…..… (B) their credit. A recursive procedure is followed while creating the data division space that will {Ok} ⇒ {Neutral}……..… (C) result in the leaf hubs having the base quantities of the records used as the end goal of Probabilistic classifiers: Probabilistic characterization. classifiers also known as the generative classifier makes use of a mix of models that Linear classification: Linear classifiers are generatively examines each section for a of different types. The Sustenance Vector particular term using the mix and classifies the machines are one kind of linear classifiers that sections accordingly. The three probabilistic make use of distinct direct separators for classifiers are Naïve Byes, Maximum Entropy classification. Support Vector Machines (SVM) and Bayesian Network. which is used in neural networks makes use of supervised learning method to create decision Naive Bayes Classifier (NB): NB is a planes that provide the decision boundaries probabilistic classifier that relies on upon Bayes specification. Decision planes consist of sets of hypothesis with solid and innocent freedom objects that have different class membership. A suppositions. It is a champion among the most linear classifier is a line that separates the sets principal content order procedures with of objects classified according to their different applications such as archive corresponding spheres. If the partitioning is classification, email spam discovery, individual more than two dimensions, the classifier is a email sorting, and slant recognition and dialect curve that is known as the hyperplane classifier. identification. It performs better in numerous The SVM produces 0 or 1 which is positive or perplexing genuine issues. The Naive Bayes negative from the organized input in a vector Variations are Bernoulli Naive Bayes, space using its portrayed information. The Binarized Multinomial Naive Bayes and the content that is archived in a unique space cannot Multinomial Naive Bayes. Each framework be used for learning. It has to be configured for passes on a very surprising result since they use calculation using ML algorithms. The pre- a unique model. When the different events of preparation of the content archive involves the the words matter a great deal in the transformation of every word to measurement characterization issue multinomial naive bayes and indistinguishable words will relate to the is utilized. Binarized multinomial naive bayes 266 is used when the regularities of the words don't learning makes use of both the labelled and accept a key part in the arrangement. some unlabelled data in its training sets. So the goal of semi-supervised learning could be to Bayesian Network (BN): The Naïve Bayes either to predict the values of the unlabelled classifier is the independence of the features. data in the training set in which case it would Assumption of Naïve Bayes is to expect that be called transductive semi-supervised learning every one of the components is completely or to find the values of the test sets in which dependent. This prompts to the BN which case it would be called the inductive semi- shows a coordinated acyclic graph and whose supervised learning. nodes correspond random variables, and edges that represent conditional dependencies. BN is viewed as an entire factor with their association. 2.2 Lexicon based Approaches In this way, an entire joint probability This approach depends on sentiment distribution over each one of the elements is Lexicons. Lexicon is considered as an resolved for a model. The computation important indicator for sentiment, which is complexity of the BN is exceptionally costly in called opinion word. Lexicon can be separated text mining. So, BN is used very little. BN is into the dictionary approach and corpus-based utilized to consider a true issue. approach Maximum Entropy (MaxEnt) Classifier: Dictionary-based approach: An MaxEnt classifiers are feature-based models arrangement of sentiment word is gathered that are most preferred when it is required to manually with known instructions. The uniformly satisfy a given constraint. Unlike the conclusion set is created by looking in the Naïve Bayes model, MaxEnt does not make any prominent repository WordNet for their autonomous assumption of the features. So proportionate word and antonyms. The next MaxEnt allows the addition of features like iteration starts when the words are added into bigrams and phrases without the problem of the seed list. In the absence of finding any new feature overlapping. The MaxEnt principle is words, the iterative process stops. After the explicitly important in the case of information system completes, a manual appraisal can be that needs to be examined to determine if a done to evacuate or amend errors. Dictionary given distribution is consistent with the testable approach has an inconvenience, that is, the information. MaxEnt or its variations are useful inability to discover feeling words with space because of their accuracy. Consistency results and setting specific introductions. of the algorithm are another major advantage of Corpus-based approach: Corpus-based it. The ability of the classifier is to work with a method deals with the issue of finding huge amount of data describes its performance sensitivity words with setting specific or its efficiency. When it comes to handling presentation. Corpus construct techniques different types of data in a single platform and depend on syntactic illustrations or examples classify them according to it, the flexibility of that are found together with a semblance of the classifier is almost perfect. assessment words to find other inferencing Unsupervised Learning: Unsupervised words in a broad corpus. The objectives are for learning technique applies a comparison conjunctions like AND, OR, BUT, EITHER E- formula between sentiment values in a lexicon XOR. The conjunction in AND case that are with the components of the text at hand. The conjoined descriptive words, for the most part, words in the lexicon have predefined values and have the comparative introduction. This they are applied to similar words in the text. contemplation is called notion consistency, Two most commonly used unsupervised which is for the most part not solid for all techniques are hierarchical clustering and predictable practicality. There are also partial clustering. adversative expressions on account of BUT they are shown as opinion changes to figure out Semi-Supervised Learning: Semi- whether two conjoined descriptors are of the supervised learning involves the use of both the same or distinctive introductions. The corpus- unsupervised and supervised techniques in its based approach uses factual or semantic classification models. The semi-supervised techniques to discover assumption extremity. 267 2.3 Hybrid Approaches Entities and interrelationship between the A hybrid approach is a combination of ML entities that can be interpreted by machines. An and Lexicon approaches used in the ontology model incorporates the entities, Sentimental Analysis. Although the hybrid concepts, classes or objects, their properties and approach may not be used frequently, they are the relation between them. For the sharing of known to produce results better than the knowledge among researchers about a approaches mentioned earlier. [1] particular domain, the ontology provides common vocabulary definitions. Such 2.4 Ontology-based Approaches ontological definitions not only help to share a Ontology applies the hierarchy of concepts common knowledge understanding among to represent domain knowledge. An ontology people and software agents but also allows the can be defined as an unambiguous, machine- reusability of domain knowledge. Storage of readable representation of a common knowledge in an ontology model is done using formulation of concepts. An Ontology is used either OWL/XML or RDF/XML format. The to represent knowledge formally by modelling merging of two ontology models which have the terms of a particular domain and by the same domain knowledge is possible. capturing the semantic relationship between the Information retrieval from an ontology model is terms. The relationship between terms is done by querying it. [2] particularly important for aspect-level sentiment analysis more specifically in product In the concept of sentiment analysis portraits reviews since product reviews are usually different levels, various approaches, many qualified by their aspects. Ontological classifications and evaluation methods are approaches are generally used to capture the displayed in figure 2.1. relation between the product and their properties in a hierarchical order. Such ordering is done using definitions of fundamental Figure 2.1 Levels of sentiment analysis, approaches, classifications and method 268 and explained that the theoretical type of 3. Related Works sentiment challenges is the growing area of research. By proving that with the growth in research in a sentiment analysis the average Various sentimental approaches are accuracy has reduced, the inference from the surveyed [3-8] and these approaches are average of accuracy based on the number of compared based on the criteria such as issues researches in each challenge has been addressed, techniques, datasets used, accuracy established. The compassion circle could be and limitation [3]. Sentiment analysis tools are extended to new research literature for further also surveyed [6]. work. [12]. The tools for Sentiment Analysis that are used in different domains were also studied analyzing its features and performance. 4. Methodology SentiWordNet, LIWC, EMOTIONS, SenticNet, Happiness Index, AFINN, PANAS- The Preferred Reporting Items for t, Sentiment140 Dataset, NRC, EWGA, FRN Systematic Reviews and Meta-Analyses [6], Mahout and Weka [9] are the sentiment (PRISMA) are a method of reporting a analysis tools reported in the literature. minimum set of items grounded on shreds of Twenty four articles are compared in the evidence in systematic reviews and meta- survey to study sentiment analysis for its analyses. The PRISMA flow diagram in figure importance, effects and the challenges in 4.2 exhibits how information moves along the sentiment evaluation. The sentiment review different phases in systematic reviews. The structure and the challenges in sentiment number of records identified, included and analysis are initially compared. The domain- excluded, and the reasons for exclusions are dependence of the sentiment challenges is an charted out in the PRISMA flow diagram. [13]. essential feature that is revealed by the The following procedures of the systematic comparison. The popularity of the negation literature review were carried out to survey the challenges in the structure of all the reviews current study and research in and around the types differed in its implicit or explicit subject of sentiment analysis. meaning. The result of such comparison provided a capacity to measure the effects of 4.1 Search Process each sentiment challenge based on the structure The search process included the search for of the review types. The topic, nature and the pertinent research literature on the internet structure of the review regulate the appropriate through websites, online libraries and databases challenges for the assessment of reviews on like Springer, IEEE Xplore, Web of Science, sentiment analysis. The next comparison that is Scopus, Science Direct, ACM Digital Library, done between the relevance of the challenges in Elsevier and Google Scholar, etc.. (Listed in sentiment analysis relevant with the accuracy Table 4.1). The vital terms or strings that were rate. From this comparison, two things were used for search are ‘‘survey on sentiment evident, namely, the status of the sentiment analysis”, ‘‘survey on opinion mining”, and challenges for evaluation of the sentiments and “Art of survey on sentimental analysis”. It was the method of finding the fitting challenge to found that searching the references of the improve accuracy. The relation between the research articles of the various studies with the amount of theoretical and technical use of three key terms yielded more articles. sentiment techniques to resolve sentiment challenges is also found. It was also established Table 4.1 List of research articles searched in different databases. Google ACM Databases IEEE Web of Science Scholar Digital Springer Xplore Science Scopus Direct /Proceedings Elsevier Library No. of Research Articles 8 16 4 12 8 7 2 9 269 6 5 4 4 4 4 3 3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 2004 - 2010 2011 - 2013 2014 - 2015 2016 - 2017 2018 - 2019 Figure 4.2: List of research articles searched in different databases – Year wise. 4.2 The criterion for including or research articles that were not relevant to above the keyword search or not relevant to the survey excluding articles and ‘white’ articles as mentioned in table 4.2. The number of literature available on sentimental analysis is quite vast. To set the number of literature to that which can be 4.3 Document Retrieval and managed or those with a focus, certain criteria Bibliography Management were devised in the selection of the literature for review. The search process described above was i) Criterion for Including – Only those used to identify the relevant literature and these literature on sentiment analysis or on tools and were then checked by their title and abstract techniques for sentiment analysis that was using the criteria of including or excluding published during the period 2005–2019 literature. Once all the relevant research articles (mentioned in figure 4.2) are considered. If the that were identified for the review process, they same or similar content is published in more were downloaded extracting their data and to than one journal or conference proceeding, the analyse them further. In Figure 4.2 is a most complete version of the literature was PRISMA flowchart to illustrate the process of chosen. the search that was done for the selection for ii) Criteria for Excluding – Works literature for review [13]. published in unknown conferences or journals, 270 Figure 4.2 PRISMA flowchart for sentiment review of literature Inclusion and exclusion criteria based on PRISMA Model as listed in the table 4.2. Table 4.2 Findings and methodology of selected research literature Reference Data Content Modality Sentiment Findings (Research Issues) Methodology No Measureme /Approach nt [1] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Language Dependency [2] Social Media Text/Image No - Detection of Fake Reviews - Lexicon Content and Spams - Language Dependency [3] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Language Dependency 271 [4] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Language Dependency [5] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Language Dependency [6] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Limitations in Classification Filtering - Domain Dependency [7] User Text No - Domain Dependency - Lexicon Generated - Comparisons Data - Negations - Sarcasm [8] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Language Dependency [9] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Language Dependency [10] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Domain Dependency - Sarcasm - Interrogative Sentences - Sentiment without sentiment words - Conditional sentences [11] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Language Dependency [12] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Language Dependency [14] User Text No - Detection of Fake Reviews - Lexicon Generated and Spams Data - Language Dependency [15] Facial Image No - Lexicon Extraction (biometrics) [16] Online News Text and Yes - Lexicon Galleries Image [17] Micro- Text and Yes - Lexicon Blogging Image Contents. Like Twitter and Messaging Excluded in the PRISMA Systems Model [18] User Review Text Yes - Lexicon on Social Content and Geo Located Map [19] Audio and Audio, Video No - Lexicon Video signals: and Text Emotion Categorization 272 [20] Multimedia Text Yes - Lexicon Content (Product Reviews) [21] User Posts Text Yes - Detection of Fake Reviews - Lexicon (Reviews on and Spams Social Media - Language Dependency Content) [22] Reviews, Text Yes - Detection of Fake Reviews - Lexicon forum and Spams discussions, - Language Dependency blogs and social networks [23] Video Video No - Lexicon Streaming Excluded in the PRISMA (Content- Model Based Retrieval) [24] User generated Text No - Detection of Fake Reviews - Lexicon data and Spams - Language Dependency - Sentiment without sentiment words [25] Social Media Emoji No - Lexicon Contents Symbols on Excluded in the PRISMA User Reviews Model [26] User- Text Yes - Lexicon generated data [27] Research Text Yes - Detection of Fake Reviews - Lexicon Articles and Spams - Language Dependency [28] User Reviews Text Yes - Lexicon on the Internet [29] User Reviews Text Yes - Lexicon on Social Media [30] Extracting the Text No - Lexicon Text on Image Excluded in the PRISMA on Social Model Media [31] User- Text No - Lexicon generated data [32] Multimedia Text, Audio No - Lexicon Contents and Video [33] Twitter Data Text Yes - Detection of Fake Reviews - Lexicon and Spams - Language Dependency [34] Healthcare Text and No - Lexicon Review Image Information [35] Multimedia Image and No Excluded in PRISMA - Lexicon Signals Video Model [36] User opinions Text Yes - Lexicon on Social Media 273 [37] User Audio No - Lexicon communicatio n (Speech information) [38] User Reviews Text Yes - Lexicon (Based on Text, Speech and Visual) [39] User generated Text, Image No - Lexicon data [40] User generated Text, Audio No - Lexicon content [41] Audio Fusion Audio No - Lexicon Information [42] User generated Text No - Lexicon content [43] Multimedia Text, Image No - Lexicon content [44] User generated Image and No - Lexicon content Video [45] Image Dataset Image No - Lexicon [46] User generated Image No - Lexicon content [47] User generated Text No - Lexicon data [48] Multimedia Text and No - Lexicon Contents Image [49] User generated Text No - Detection of Fake Reviews - Lexicon content and Spams - Language Dependency [50] Twitter Data Text Yes - Detection of Fake Reviews - Lexicon and Spams - Language Dependency - Sentiment without sentiment words [51] YouTube - Text Yes - Detection of Fake Reviews - Lexicon User generated and Spams Reviews - Language Dependency [52] User generated Text No - Lexicon content [53] Textual and Text and No - Lexicon Excluded in the PRISMA Visual Data in Image Model Social Media [54] User generated Text No - Lexicon content [55] User generated Text No - Detection of Fake Reviews - Lexicon content and Spams - Language Dependency [56] User generated Text and No - Lexicon content on Visual Data Web [57] Image content Image No - Lexicon in social media Excluded in the PRISMA Model [58] Microblog Image No - Lexicon Image content [59] User generated Text, Image No - Lexicon content 274 [60] User generated Text Yes - Lexicon content [61] User generated Image No - Lexicon Image [62] Twitter Data Emoji Yes - Lexicon [63] Text-based Text No - Detection of Fake Reviews - Lexicon Survey and Spams - Language Dependency - Sentiment without sentiment words [64] User generated Image No - Lexicon Images Excluded in the PRISMA [65] User generated Text No Model - Lexicon Review Text [66] Pre – Processing the text-based of tokenization 5. Issues, Challenges and Research Gap Having studied the contents of the literature on challenging to find the highly positive or highly Sentimental Analysis, the following issues and negative rating based on the intensity of opinion challenges were evident: that is given. It is called a degree of polarity. [7] Finding Fake Reviews and Spams: Internet Negations: If the negations are not handled contents especially in social media comprises properly can give completely wrong results. [7] both authentic and spam contents. Therefore, it is necessary to remove the spam and fake Sarcasm: Identify and analyze emotions contents before pre-processing. [4] [6] [10] voiced in the text at a more fine-grained level. [7] [10] Limitations in Classification Filtering: Unrelated opinions are removed to establish the Interrogative Sentences: When dealing with most popular opinion. The classification question type sentences, the sentence itself may filtering techniques do not provide the expected not contain positive or negative sentiments but results. [6] the keywords used in such a sentence might express positive or negative sentiments. [10] Language Dependency: Maximum of the work focused only on the English text based Sentiment without sentiment words: At times content and thus, most of the resources are there could be sentences without any of the available in English. [4] keywords that express sentiments such good, better, best, worst, bad and so on but the Domain Dependency: Opinion mining sentences itself might be used to express depends on the domain text used. [6] [10] positive or negative feedback about some particular product, service or policy. [10] Word Sense Disambiguation: Exact meaning of a word based on the context needs to be Conditional sentences: Conditional sentences extracted as words can have different meanings create problems similar to interrogative for different fields. [7] sentences and therefore are a challenge in sentimental analysis. [10] Comparisons: To decide the polarity for relative sentences can be a challenge. It is 275 The following two important issues were not and approach.” 2016 Eighth tackled in the studies that were done so far: International Conference on Advanced Computing (ICoAC) (2017): 72-76. Multimedia Content – so far the above survey [4] Jain, S. and P. Singh. “Systematic pointed out the 99% of content taken for the Survey on Sentiment Analysis.” 2018 sentimental analysis based on text. Few articles First International Conference on only tell about the multimedia content like Secure Cyber Computing and audio, image and video-based sentiment Communication (ICSCCC) (2018): analysis based on lexicon approach. 561-565. Machine Learning Approach – different [5] Kaur, Harpreet & Mangat, Veenu & approaches proposed for sentiment analysis Nidhi,. (2017). A survey of sentiment (like Lexicon based, Machine learning based analysis techniques. 921-925. and Ontology-based), but no implementation 10.1109/I-SMAC.2017.8058315. based on multimedia content using machine [6] S. Rajalakshmi, S. Asha and N. learning. Pazhaniraja, "A comprehensive survey on sentiment analysis," 2017 Fourth 6. Conclusion International Conference on Signal Processing, Communication and There has been a lot of attention recently on Networking (ICSCN), Chennai, 2017, sentiment analysis especially from researchers pp.1-5, doi: from the fields of text mining and natural 10.1109/ICSCN.2017.8085673. language processing. But due to extreme absence of annotated datasets which is used to [7] Kolkur, Seema et al. “Study of train models in various domains, the accuracy Different Levels for Sentiment of sentiment analysis has been hindered. Analysis.” (2015). Several kinds of research have been done out to [8] K.H. Rizwana, Dr. B. Kalpana, “A confront the challenge and enhance sentiment Survey on Sentiment Analysis and analysis classification. Sentiment analysis is Opinion Mining”. International Journal important as it helps in identifying the of Pure and Applied Mathematics, emotional and attitude states of people. Volume 118, No.18, 2681-2688, ISSN: Positive or negative feelings of people can be 1311-8080 (printed version), ISSN: expressed in different ways. This research 1314-3395 (online version), 2018. article talks about, in subtle terms, the different [9] Jyoti and Seema Rao. 2016. A Survey ways to deal with sentiment analysis mostly in on Sentiment Analysis and Opinion Machine Learning, Lexicon-based, Hybrid and Mining. Article 53, 1–5. Ontology-based approaches. This research DOI:https://doi.org/10.1145/2979779. article gives a point by point perspective of the 2979832 distinctive applications and challenges of [10] M., Vidisha & Vala, Jay & Balani, Sentiment Analysis. Prem. (2016). A Survey on Sentiment Analysis Algorithms for Opinion References Mining. International Journal of Computer Applications. 133. 7-11. [1] Megha Joshi, Purvi Prajapati, Ayesha 10.5120/ijca2016907977. Shaikh and Vishwa Vala. “A Survey on [11] S. Vasanthraj, A. Martin, A. Sentiment Analysis”. International Meiappane, J. Madhusudnan, “A Journal of Computer Applications Survey on Sentiment Analysis - 163(6):34-38, April 2017. Applied in Opinion Mining”, Journal [2] Thakor, Pratik and S. Sasi. “Ontology- of Network Communication and based Sentiment Analysis Process for Emerging Technologies (JNCET), Social Media Content.” INNS Volume 1, Issue 1, ISSN: 2395-5317, Conference on Big Data (2015). 2015. [3] Abirami, A. and V. Gayathri. “A [12] Doaa Mohey El-Din Mohamed survey on sentiment analysis methods Hussein,A survey on sentiment 276 analysis challenges, Journal of King 3-14, ISSN 0262-8856, Saud University - Engineering https://doi.org/10.1016/j.imavis.2017 Sciences, Volume 30, Issue 4, 2018, .08.003. Pages 330-338, ISSN 1018-3639, [21] Lin Yue, Weitong Chen, Xue Li, https://doi.org/10.1016/j.jksues.2016. Wanli Zuo, and Minghao Yin. 2019. 04.002. A survey of sentiment analysis in [13] PRISMA Model - social media. Knowl. Inf. Syst. 60, 2 http://www.prisma- (August 2019), 617–663. statement.org/PRISMAStatement/Fl DOI:https://doi.org/10.1007/s10115- 018-1236-4 owDiagram. [22] Joshi, N. and Suhasini A. Itkat. “A [14] T. Al-Moslmi, N. Omar, S. Abdullah Survey on Feature Level Sentiment and M. Albared, "Approaches to Analysis.” (2014). Cross-Domain Sentiment Analysis: A Systematic Literature Review," in [23] Calic, Janko & Campbell, N. & IEEE Access, vol. 5, pp. 16173- Dasiopoulou, Stamatia & 16192, 2017, doi: Kompatsiaris, Ioannis. (2005). A 10.1109/ACCESS.2017.2690342. Survey on Multimodal Video [15] Zheng, Yufeng & Elmaghraby, Adel. Representation for Semantic (2011). A brief survey on Retrieval. 1. 135 - 138. multispectral face recognition and 10.1109/EURCON.2005.1629877. multimodal score fusion. 543-550. [24] Hussein, D.-M.E.D.M. “A survey on 10.1109/ISSPIT.2011.6151622. sentiment analysis challenges”. [16] Helen Caple & John S. Journal of King Saud University – Knox (2015) A framework for the Engineering Sciences (2016), multimodal analysis of online news http://dx.doi.org/10.1016/j.jksues.20 galleries: What makes a “good” 16.04.002. picture gallery?, Social [25] G. Varni, G. Volpe, and A. Camurri. Semiotics, 25:3, 292321, DOI: 10.10 2010. A System for Real-Time 80/10350330.2014.1002174 Multimodal Analysis of Nonverbal [17] Baecchi, C., Uricchio, T., Bertini, Affective Social Interaction in User- M. et al. A multimodal feature Centric Media. Trans. Multi. 12, 6 learning approach for sentiment (October 2010), 576–590. analysis of social network DOI:https://doi.org/10.1109/TMM.2 multimedia. Multimed Tools 010.2052592. Appl 75, 2507–2525 (2016). [26] D'Andrea, Alessia & Ferri, Fernando https://doi.org/10.1007/s11042-015- & Grifoni, Patrizia & Guzzo, Tiziana. 2646-x (2015). Approaches, Tools and [18] Evangelia Polymerou, Despoina Applications for Sentiment Analysis Chatzakou, and Athena Vakali. 2014. Implementation. International Journal EmoTube: A Sentiment Analysis of Computer Applications. 125. 26- Integrated Environment for Social 33. 10.5120/ijca2015905866. Web Content. Article 20, 1–6. DOI:https://doi.org/10.1145/2611040 [27] Toqir A. Rana and Yu-N Cheah. .2611081 2016. Aspect extraction in sentiment [19] Sikandar, M.. “A Survey for analysis: comparative analysis and Multimodal Sentiment Analysis survey. Artif. Intell. Rev. 46, 4 Methods.” (2014). (December 2016), 459–483. [20] Mohammad Soleymani, David DOI:https://doi.org/10.1007/s10462- Garcia, Brendan Jou, Björn Schuller, 016-9472-z. Shih-Fu Chang, Maja Pantic, A [28] M., Vidisha & Vala, Jay & Balani, survey of multimodal sentiment Prem. (2016). A Survey on Sentiment analysis, Image and Vision Analysis Algorithms for Opinion Computing, Volume 65, 2017, Pages 277 Mining. International Journal of [36] Shah, Rajiv Ratn. (2016). Multimodal Computer Applications. 133. 7-11. Analysis of User-Generated Content 10.5120/ijca2016907977. in Support of Social Media [29] Siti Salwa Hasbullah, Diana Applications. 423-426. Maynard, Rita Zaharah Wan Chik, 10.1145/2911996.2912032. Farahwahida Mohd, and Myzan [37] Nicu Sebe, Ira Cohen, Theo Gevers , Noor. 2016. Automated Content and Thomas S. Huang, “Multimodal Analysis: A Sentiment Analysis on Approaches for Emotion Malaysian Government Social Recognition: A Survey”, SPIE 5670, Media. Article 30, 1–6. Internet Imaging VI, (17 January DOI:https://doi.org/10.1145/2857546 2005); doi: 10.1117/12.600746 .2857577 [38] Cummins, Nicholas & Amiriparian, [30] R. Ji, D. Cao and D. Lin, "Cross- Shahin & Ottl, Sandra & Gerczuk, Modality Sentiment Analysis for Maurice & Schmitt, Maximilian & Social Multimedia," in 2015 IEEE Schuller, Björn. (2018). Multimodal International Conference on Bag-of-Words for Cross Domains Multimedia Big Data (BigMM), Sentiment Analysis. Beijing, China, 2015 pp. 28-31. doi: 10.1109/ICASSP.2018.8462660. 10.1109/BigMM.2015.85. [39] Atrey, P.K., Hossain, M.A., El [31] Zhang, L, Wang, S, Liu, B. Deep Saddik, A. et al. Multimodal fusion learning for sentiment analysis: A for multimedia analysis: a survey. survey. WIREs Data Mining Knowl Multimedia Systems 16, 345–379 Discov. 2018; 8:e1253. (2010).https://doi.org/10.1007/s0053 https://doi.org/10.1002/widm.1253 0-010-0182-0. [32] Poria S, Cambria E, Howard N, [40] Jaimes A., Sebe N. (2005) Huang G & Hussain A (2016) Fusing Multimodal Human Computer audio, visual and textual clues for Interaction: A Survey. In: Sebe N., sentiment analysis from multimodal Lew M., Huang T.S. (eds) Computer content, Neurocomputing, 174 (A), Vision in Human-Computer pp. 50-59. Interaction. HCI 2005. Lecture Notes [33] Giachanou, Anastasia & Crestani, in Computer Science, vol 3766. Fabio. (2016). Like It or Not: A Springer, Berlin, Survey of Twitter Sentiment Analysis Heidelberg.https://doi.org/10.1007/1 Methods. ACM Computing Surveys. 1573425_1. 49. 1-41. 10.1145/2938640. [41] Multimodal information fusion of [34] R. Pang, A. Baretto, H. Kautz and J. audio emotion recognition based on Luo, "Monitoring adolescent alcohol kernel entropy component analysis use via multimodal analysis in social zhibing xie and ling guan multimedia," 2015 IEEE International Journal of Semantic International Conference on Big Data Computing 2013 07:01, 25-42. (Big Data), Santa Clara, CA, 2015, [42] Dumas B., Lalanne D., Oviatt S. pp. 1509-1518, doi: (2009) Multimodal Interfaces: A 10.1109/BigData.2015.7363914. Survey of Principles, Models and [35] Camurri A., Mazzarino B., Ricchetti Frameworks. In: Lalanne D., Kohlas M., Timmers R., Volpe G. (2004) J. (eds) Human Machine Interaction. Multimodal Analysis of Expressive Lecture Notes in Computer Science, Gesture in Music and Dance vol 5440. Springer, Berlin, Performances. In: Camurri A., Volpe Heidelberg.https://doi.org/10.1007/9 G. (eds) Gesture-Based 78-3-642-00437-7_1. Communication in Human-Computer [43] Tadas Baltrusaitis, Chaitanya Ahuja, Interaction. GW 2003. Lecture Notes and Louis-Philippe Morency. 2019. in Computer Science, vol 2915. Multimodal Machine Learning: A Springer, Berlin, Heidelberg. Survey and Taxonomy. IEEE Trans. https://doi.org/10.1007/978-3-540- Pattern Anal. Mach. Intell. 41, 2 24598-8_3. (February 2019), 423–443. 278 DOI:https://doi.org/10.1109/TPAMI. https://doi.org/10.1016/j.asej.2014.0 2018.2798607. 4.011. [44] Shah, Rajiv Ratn. (2016). [50] Kharde, Vishal & Sonawane, Sheetal. Multimodal-based Multimedia (2016). Sentiment Analysis of Twitter Analysis, Retrieval, and Services in Data: A Survey of Techniques. Support of Social Media International Journal of Computer Applications. Applications. 139. 5-15. 10.1145/2964284.2971471. 10.5120/ijca2016908625. [45] Tao Chen, Felix X. Yu, Jiawei Chen, [51] Asghar, Dr. Muhammad & Ahmad, Yin Cui, Yan-Ying Chen, and Shih- Shakeel & Marwat, Afsana & Kundi, Fu Chang. 2014. Object-Based Visual Fazal. (2015). Sentiment Analysis on Sentiment Concept Analysis and YouTube: A Brief Survey. Brist. 3. Application. In Proceedings of the 1250-1257. 22nd ACM international conference [52] Yuan J., You Q., Luo J. (2015) on Multimedia (MM '14). Association Sentiment Analysis Using Social for Computing Machinery, New Multimedia. In: Baughman A., Gao York, NY, USA, 367–376. J., Pan JY., Petrushin V. (eds) DOI:https://doi.org/10.1145/2647868 Multimedia Data Mining and .2654935 Analytics. Springer, Cham. [46] Yan-Ying Chen, Tao Chen, Winston https://doi.org/10.1007/978-3-319- H. Hsu, Hong-Yuan Mark Liao, and 14998-1_2 Shih-Fu Chang. 2014. Predicting [53] Quanzeng You. 2016. Sentiment and Viewer Affective Comments Based Emotion Analysis for Social on Image Content in Social Media. In Multimedia: Methodologies and Proceedings of International Applications. In Proceedings of the Conference on Multimedia Retrieval 24th ACM international conference (ICMR '14). Association for on Multimedia (MM '16). Association Computing Machinery, New York, for Computing Machinery, New NY, USA, 233–240. York, NY, USA, 1445–1449. DOI:https://doi.org/10.1145/2578726 DOI:https://doi.org/10.1145/2964284 .2578756. .2971475 [47] Swagatam Das, Sayan Maity, Bo- [54] K. Schouten and F. Frasincar, Yang Qu, P.N. Suganthan, Real- "Survey on Aspect-Level Sentiment parameter evolutionary multimodal Analysis," in IEEE Transactions on optimization — A survey of the state- Knowledge and Data Engineering, of-the-art, Swarm and Evolutionary vol. 28, no. 3, pp. 813-830, 1 March Computation, Volume 1, Issue 2, 2016, doi: 2011, Pages 71-88, ISSN 2210-6502, 10.1109/TKDE.2015.2485209. https://doi.org/10.1016/j.swevo.2011 [55] M. Desai and M. A. Mehta, .05.005. "Techniques for sentiment analysis of [48] H. Wang, A. Meghawat, L. Morency Twitter data: A comprehensive and E. P. Xing, "Select-additive survey," 2016 International learning: Improving generalization in Conference on Computing, multimodal sentiment analysis," 2017 Communication and Automation IEEE International Conference on (ICCCA), Noida, 2016, pp. 149-154, Multimedia and Expo (ICME), Hong doi: 10.1109/CCAA.2016.7813707. Kong, 2017, pp. 949-954, doi: [56] Louis-Philippe Morency, Rada 10.1109/ICME.2017.8019301. Mihalcea, and Payal Doshi. 2011. [49] Walaa Medhat, Ahmed Hassan, Hoda “towards multimodal sentiment Korashy, Sentiment analysis analysis: harvesting opinions from algorithms and applications: A the web”. Association for Computing survey, Ain Shams Engineering Machinery, New York, NY, USA, Journal,Volume 5, Issue 4, 2014, 169–176. Pages 1093-1113, ISSN 2090-4479, DOI:https://doi.org/10.1145/2070481 .2070509. 279 [57] J. Islam and Y. Zhang, "Visual [64] Eric C. Yang, “Development of an sentiment analysis for social images integrated multimodal optical using transfer learning approach", pp. imaging system with real-time image 124-130,doi:10.1109/BDCloud- analysis for the evaluation of oral SocialCom-SustainCom.2016.29. premalignant lesions, J. Biomed. Opt. [58] Cao, D., Ji, R., Lin, D. et al. “Visual 24(2) 025003 (21 February 2019) sentiment topic model based https://doi.org/10.1117/1.JBO.24.2.0 microblog image sentiment analysis”. 25003. Multimed Tools Appl 75, 8955–8968 [65] Sankar, H, Subramaniyaswamy, V, (2016).https://doi.org/10.1007/s1104 Vijayakumar, V, Arun Kumar, S, 2-014-2337-z. Logesh, R, Umamakeswari, A. [59] J.Kehrer and H.Hauser, Intelligent sentiment analysis "Visualization and Visual Analysis of approach using edge computing‐ Multifaceted Scientific Data: A based deep learning technique. Softw: Survey," in IEEE Transactions on Pract Exper. 2020; 50: 645– Visualization and Computer 657.https://doi.org/10.1002/spe.2687 Graphics, vol. 19, no. 3, pp. 495-513, [66] Vijayarani, S., Ms J. Ilamathi, and Ms March2013,doi:10.1109/TVCG.2012 Nithya. "Preprocessing techniques for .110. text mining-an overview." [60] Q. Fang, C. Xu, J. Sang, M. S. International Journal of Computer Hossain and G. Muhammad, "Word- Science & Communication Networks of-Mouth Understanding: Entity- 5.1 (2015): 7-16. Centric Multimodal Aspect-Opinion Mining in Social Media," in IEEE Transactions on Multimedia, vol. 17, no. 12, pp. 2281-2296, Dec. 2015, doi: 10.1109/TMM.2015.2491019. [61] Christian Otto, Matthias Springstein, Avishek Anand, and Ralph Ewerth. 2019. Understanding, Categorizing and Predicting Semantic Image-Text Relations. In Proceedings of the 2019 on International Conference on Multimedia Retrieval (ICMR '19). Association for Computing Machinery, New York, NY, USA, 168–176. DOI:https://doi.org/10.1145/3323873 .3325049. [62] Ziad Al-Halah, Andrew Aitken, Wenzhe Shi, Jose Caballero, “Smile, be Happy: Emoji Embedding for Visual Sentiment Analysis”, Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019. [63] Saxena Diksha, Gupta Shubham, Jos eph Joy, Prof. Mehra Richa, “Sentiment Analysis”, International Journal of Engineering, Science and Mathematics Vol. 8 Issue 3, March 2019,ISSN:2320-0294 Impact Factor: 6.765.