Real-time Sentiment Analysis on Social Networks using Meta-model and Machine Learning Techniques

Scalable Computing: Practice and Experience, ISSN 1895-1767, http://www.scpe.org REAL-TIME SENTIMENT ANALYSIS ON SOCIAL NETWORKS USING META-MODEL AND MACHINE LEARNING TECHNIQUES Xiao ShiXiao ¥ , Mustafa Muwafak Alobaedy * , S. B. Goyal + , Sanjay Singla ~ , Sandeep Kang # , Raman Chadha† Abstract—Sentiment analysis is a critical task in social media analysis, enabling the understanding of user attitudes and opinions towards various topics. This paper proposes a real- time sentiment analysis system for social networks that utilizes a meta-model and machine learning techniques to accurately classify user sentiment. The proposed system integrates textual and visual data from social media posts to improve sentiment classification accuracy. The methodology includes data collection and preprocessing, feature extraction and selection, and the proposed meta-model for sentiment analysis. The system utilizes several machine learning techniques, including SVM, CNN, and LSTM networks. We evaluated the proposed system on a large- scale dataset and compared its performance with several state- of-the-art methods. The evaluation metrics, including accuracy, precision, recall, and F1-score, showed that our proposed system outperforms existing methods. The proposed system’s ability to handle multimodal data and achieve high accuracy in real- time makes it suitable for various applications, including social media monitoring and marketing analysis. The proposed system’s limitations provide opportunities for further research, such as developing more efficient algorithms and models that require less training data, and improving techniques for handling noisy and ambiguous data, such as sarcasm and irony. In conclusion, the proposed real-time sentiment analysis system using a meta-model and machine learning techniques provides a robust and efficient solution for sentiment analysis on social networks. The proposed system’s performance and potential applications demonstrate its importance in the field of social media analysis. Key words —Real-time, Sentiment Analysis, Social Networks, Machine Learning, Meta-Model INTRODUCTION. Introduction Social networks have become an integral part of our daily lives, with millions of users worldwide [1]. These platforms are a valuable source of user-generated content, which can provide insight into users’ opinions and needs. Sentiment analysis is a vital task in social media analysis, which aims to determine the polarity (positive, negative, or neutral) of the opinions expressed in social media posts. Traditional sentiment analysis methods mainly rely on rule- based or lexicon-based approaches, which have limitations such as low accuracy and failure to handle complex data. In recent years, machine learning techniques have shown promising results in sentiment analysis, where the accuracy of sentiment classification has been significantly improved [2]. However, most existing methods are not suitable for real- time analysis and cannot handle multimodal data, which is commonly found in social media. The main objective of this paper is to propose a real-time sentiment analysis system for social networks that utilizes a meta-model and machine learning techniques to accurately classify user sentiment. The proposed system takes advantage of both textual and visual data from social media posts to improve the accuracy of sentiment classification [3]. ¥ City University, Petaling Jaya, Malaysia ; Chengyi College Jimei University, Xiamen, China, (Chinaxiaoshixiao@jmu.edu.cn) * City University, Petaling Jaya,46100, Malaysia,( mustafa.theab@city.edu.my) + City University, Petaling Jaya,46100, Malaysia,(drsbgoyal@gmail.com) ~ Dept. Computer Science & Engg,Chandigarh University , Kharar , Punjab, India,(Dr.ssinglacs@gmail.com) # Dept. Computer Science & Engg , University,Kharar, Punjab, India,(ad1.cse@cumail.in) †Dept. Computer Science & Engg,, Chandigarh University, Kharar, Punjab, India,(Dr.ramanchadha@gmail.com) A. Background and Motivation. Sentiment analysis on social networks has been widely studied due to its importance in understanding public opinion on various topics. The task of sentiment analysis is to determine the polarity of the opinions expressed in social media posts, which can range from positive to negative to neutral. Accurate sentiment analysis can provide valuable insights for businesses and organizations to make informed decisions [4]. Traditional sentiment analysis methods mainly rely on rule-based or lexicon-based approaches, which have limitations such as low accuracy and failure to handle complex data [5]. In recent years, machine learning techniques have shown promising results in sentiment analysis, where the accuracy of sentiment classification has been significantly improved [6]. However most existing methods are not suitable for real-time analysis and cannot handle multimodal data, which is commonly found in social media. Realtime sentiment analysis is essential for businesses and organizations that need to monitor public opinion in real-time to make informed decisions. For example, a company can use real-time sentiment analysis to monitor customers’ reactions to a new product launch or a marketing campaign. Real-time sentiment analysis can also help detect and respond to negative 7opinions quickly, thereby preventing a crisis [7]. B. Research Objective The main objective of this paper is to propose a real-time sentiment analysis system for social networks that utilizes a meta-model and machine learning techniques to accurately classify user sentiment. The proposed system takes advantage of both textual and visual data from social media posts to improve the accuracy of sentiment classification. Specifically, the research objectives of this paper are as follows: 1. To develop a real-time sentiment analysis system for social networks. 2. To propose a novel meta-model that combines various machine learning models to improve the accuracy of sentiment classification. 3. To evaluate the proposed system on a large-scale dataset and compare it with existing state-of- the-art methods. C. Contribution of the Paper The contribution of this paper is twofold. First, we propose a novel meta-model that combines various machine learning models to improve the accuracy of sentiment classification. The meta-model takes advantage of both textual and visual data from social media posts to improve the accuracy of sentiment classification. Second, we evaluate the proposed system on a large-scale dataset and achieve superior performance compared to existing state-of-the-art methods. Our proposed system is able to handle a variety of data types and achieves high accuracy in real-time. In addition to experimental evaluations, real-world validation and user feedback play a crucial role in assessing the applicability and usability of sentiment analysis systems on social networks. This paper not only presents experimental results but also explores real-world scenarios and user perspectives to provide a comprehensive understanding of the proposed methodology. D. Handling Negative Results Challenges While we present promising results in this paper, it's important to acknowledge the possibility of facing challenges and limitations inherent in complex real-time sentiment analysis systems. Our commitment to transparency and rigorous evaluation led us to discuss these challenges openly in each section (if applicable) E. Organization of this Paper The remainder of this paper is organized as follows. Section 2 provides a brief overview of related work in sentiment analysis on social networks, machine learning techniques for sentiment analysis, and real-time sentiment analysis. Section 3 presents the proposed methodology for real-time sentiment analysis on social networks using meta-model and machine learning techniques. Section 4 describes the experimental setup, including the dataset description, evaluation metrics, baseline methods, and experimental results. Section 5 provides a discussion on the proposed system and its potential appli- cations in various domains. Finally, Section 6 concludes the paper, summarizes the contributions of this work, and proposes directions for future research. RELATED WORK. In this section, we provide a detailed overview of related work in sentiment analysis on social networks, machine learning techniques for sentiment analysis, and real-time sentiment analysis. A. Sentiment Analysis on Social Networks Sentiment analysis on social networks has been an active research area for over a decade. It has gained importance due to its potential applications in various domains, including mar- keting, politics, and public opinion monitoring. Researchers have proposed several approaches to tackle the challenges of sentiment analysis on social networks, including rule-based, lexicon-based, and machine learning-based methods [8-11]. Rule-based methods rely on handcrafted rules and heuristics to classify sentiment. These methods are straightforward to implement and can achieve reasonable accuracy, but they have limitations in handling complex data and may require manual adjustments for different domains. As an example, proposed a rule-based sentiment analysis system that utilized a set of predefined rules and a sentiment lexicon to classify tweets [12]. Lexicon-based methods rely on pre-defined sentiment lexicons to classify sentiment. These methods can achieve high accuracy in certain domains, but they may struggle to handle sarcasm, irony, and other forms of figurative language. As an example, proposed a lexicon-based sentiment analysis system that utilized an Arabic sentiment lexicon to classify tweets in Arabic [13]. Machine learning-based methods have shown promising results in sentiment analysis on social networks. These methods use machine learning algorithms to learn patterns and features from data and make predictions. Some of the most popular machine learning algorithms used in sentiment analysis on social networks include Support Vector Machines (SVM), Na ̈ıve Bayes [14- 16], and Deep Learning- based approaches such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). As an ex- ample, proposed a CNN-based sentiment analysis system that achieved state-of-the-art performance on a dataset of Arabic tweets [17-19]. Table 1 summarizes some of the most relevant studies in sentiment analysis on social networks and their key contributions. B. Machine Learning Techniques for Sentiment Analysis Machine learning techniques have shown great potential in sentiment analysis, and several studies have explored different approaches to improve the accuracy of sentiment classification. One of the key challenges in sentiment analysis is to handle the variability of language in different domains and contexts. Several studies have proposed the use of domain adaptation techniques to improve the performance of sentiment analysis in different domains. Domain adaptation techniques aim to transfer knowledge from a source domain to a target domain to improve the performance of sentiment classification. As an example, proposed a domain adaptation approach that utilized a shared latent variable model to transfer knowledge between different domains [20]. Another popular approach is to use feature selection and extraction techniques to identify the most informative features for sentiment classification. Table 1: Summary of studies in sentiment analysis on social networks. Study Purpose Focus Conclusion Challenges Future Scope [8] [10] [12] Develop a rulebased sentiment analysis system Utilize a set of predefined rules and a sentiment lexicon to classify tweets Rule-based sentiment analysis can achieve reasonable accuracy Limitations in handling complex data and may require manual adjustments Explore the use of machine learning techniques in sentiment analysis [13] [14] Develop a lexicon-based sentiment analysis system Utilize an Arabic sentiment lexicon to classify tweets in Arabic Lexicon-based sentiment analysis can achieve high accuracy in certain domains Struggle to handle sarcasm, irony, and other forms of figurative language Investigate the use of machine learning techniques in Arabic sentiment analysis [15] [16] [17] Develop a deep learning-based sentiment analysis system Utilize a CNN to classify Arabic tweets Deep learning-based sentiment analysis can achieve state-ofthe-art performance Dependence on large amounts of labeled data and computational resources Investigate the use of transfer learning and semi-supervised learning in sentiment analysis Feature selection aims to select a subset of relevant features from the original feature space, while feature extraction aims to transform the original feature space into a new space that better represents the data. As an example, proposed a feature selection approach that utilized a genetic algorithm to select the most informative features for sentiment classification on Chinese microblogs [21]. Deep learning-based approaches such as CNN and RNN have also been applied in sentiment analysis, with promising results. These methods can capture the context and semantics of the data and have been shown to outperform traditional machine learning algorithms in sentiment classification tasks. As an example, proposed a CNN-based sentiment analysis system that achieved state- of-the-art performance on a dataset of English tweets [22]. Table 2 summarizes some of the most relevant studies in machine learning techniques for sentiment analysis and their key contributions. Table 2: Summary of studies in machine learning techniques for sentiment analysis. Study Purpose Focus Conclusion Challenges Future Scope [20] Develop a domain adaptation approach for sentiment classification Utilize a shared latent variable model to transfer knowledge between different domains Domain adaptation can improve the performance of sentiment classification Limited availability of labeled data in target domains Investigate the use of transfer learning in domain adaptation [8] [21] Develop a feature selection approach for sentiment classification Utilize a genetic algorithm to select the most informative features for sentiment classification on Chinese microblogs Feature selection can improve the performance of sentiment classification High dimensionality of the feature space Investigate the use of deep learning-based feature extraction in sentiment analysis [6] [22] Develop a CNN-based sentiment analysis system Utilize a CNN to classify English tweets Deep learning-based sentiment analysis can outperform traditional machine learning algorithms Dependence on large amounts of labeled data and computational resources Investigate the use of transfer learning and multimodal data in sentiment analysis. Table 2: Summary of studies in machine learning techniques for sentiment analysis. Study Purpose Focus Conclusion Challenges Future Scope [20] Develop a domain adaptation approach for sentiment classification Utilize a shared latent variable model to transfer knowledge between different domains Domain adaptation can improve the performance of sentiment classification Limited availability of labeled data in target domains Investigate the use of transfer learning in domain adaptation [8] [21] Develop a feature selection approach for sentiment classification Utilize a genetic algorithm to select the most informative features for sentiment classification on Chinese microblogs Feature selection can improve the performance of sentiment classification High dimensionality of the feature space Investigate the use of deep learningbased feature extraction in sentiment analysis [6] [22] Develop a CNNbased sentiment analysis system Utilize a CNN to classify English tweets Deep learning-based sentiment analysis can outperform traditional machine learning algorithms Dependence on large amounts of labeled data and computational resources Investigate the use of transfer learning and multimodal data in sentiment analysis C. Real-Time Sentiment Analysis Real-time sentiment analysis is essential for businesses and organizations that need to monitor public opinion in real-time to make informed decisions. Several studies have proposed real-time sentiment analysis systems for social networks, with varying degrees of success. One of the key challenges in real-time sentiment analysis is to handle the high volume and velocity of data in social networks. Several studies have proposed the use of distributed computing and streaming algorithms to process and analyze social media data in real- time. As an example, proposed a real-time sentiment analysis system that utilized Apache Storm and Hadoop to process and analyze tweets in real-time [23-24]. Another challenge is to handle multimodal data, which is commonly found in social media. Multimodal data includes not only text but also visual and audio data. Several studies have proposed the use of deep learningbased approaches to handle multimodal data in real-time sentiment analysis. As an example, proposed a multimodal deep learningbased sentiment analysis system that utilized both textual and visual data to improve the accuracy of sentiment classification [25- 26]. Table 3 summarizes some of the most relevant studies in real-time sentiment analysis and their key contributions. Table 3: Summary of studies in real-time sentiment analysis. Study Purpose Focus Conclusion Challenges Future Scope [23] [24] Develop a real-time sentiment analysis system Utilize Apache Storm and Hadoop to process and analyze tweets in real-time Realtime sentiment analysis can be achieved using distributed computing and streaming algorithms Dependence on high computational resources and large storage capacity Investigate the use of deep learning-based approaches in real-time sentiment analysis [25] [26] Develop a multimodal deep learning-based sentiment analysis system Utilize textual and visual data to improve the accuracy of sentiment classification Multimodal data can improve the performance of sentiment classification in real- time Difficulties in handling multimodal data and large-scale datasets Investigate the use of transfer learning and reinforcement learning in multimodal sentiment analysis. PROPOSED METHODOLOGY. This section presents the proposed methodology for real- time sentiment analysis on social networks using a meta- model and machine learning techniques [27-28]. In this sec- tion, we describe the various steps involved in the proposed methodology, including data collection and pre-processing, feature extraction and selection, the proposed meta-model for sentiment analysis, and the machine learning techniques used in the proposed system. The proposed methodology leverages both textual and visual information from social media posts to improve the accuracy of sentiment classification. The use of a meta-model and ensemble learning techniques helps to combine the strengths of multiple machine learning models and improve the overall performance of the system. Transfer learning techniques are used to overcome the limitations of limited labeled data in target domains. The proposed method- ology is evaluated through experiments and compared with several state-of-the-art methods in the next section. A. Meta-Model for Sentiment Analysis. The proposed system utilizes a meta-model for sentiment analysis on social networks. A meta-model is a model that combines the outputs of multiple machine learning models to generate a more accurate prediction. In the context of sentiment analysis, the metamodel combines the outputs of various classifiers to generate a final sentiment classification for a given social media post. The proposed meta-model con- sists of multiple classifiers, including Support Vector Machines (SVM), Na ̈ıve Bayes, and Convolutional Neural Networks (CNN). The output of each classifier is combined using an ensemble method to generate the final sentiment classification. Ensemble learning is a machine learning technique that combines the outputs of multiple classifiers to generate a more accurate prediction. Ensemble learning is particularly useful in situations where individual classifiers may have high variance or may perform poorly on certain types of data. In the proposed system, we use an ensemble method called majority voting to combine the outputs of the individual classifiers. In majority voting, the final prediction is determined by the class that receives the most votes from the individual classifiers. The figure 1 provides a clear visual representation of the proposed meta-model for sentiment analysis on social networks, including the individual classifiers, the ensemble method used to combine their outputs, and the meta-learning techniques used to optimize the weights of the ensemble. The diagram is a helpful tool for understanding the proposed methodology and how the various components of the meta- model work together to generate a more accurate prediction of sentiment classification. We provide a detailed explanation of our innovative meta-model designed for real-time sentiment analysis. The meta-model comprises several key components, each contributing to its overall functionality and accuracy. Below, we outline the structural aspects and components of our meta-model, highlighting how they work in concert to improve sentiment analysis accuracy. To optimize the ensemble weights and improve the performance of the meta-model, we use meta-learning techniques. Metalearning is a subfield of machine learning that deals with learning how to learn. In the context of ensemble learning, meta-learning techniques are used to learn how to combine the outputs of multiple classifiers to generate a more accurate prediction. In the proposed system, we use a technique called stacked generalization to optimize the ensemble weights. Stacked generalization involves training a second- level classifier on the outputs of the individual classifiers. The second-level classifier learns how to combine the outputs of the individual classifiers to generate a more accurate prediction. Figure 1: proposed meta-model for sentiment analysis on social networks The individual classifiers in the meta-model are trained using supervised learning techniques. Supervised learning is a machine learning technique that involves learning from labeled data. In the proposed system, we use both textual and visual features to train the individual classifiers. The textual features include bag-of-words, n-grams, and word embeddings, while the visual features include image features such as color histograms and texture features. The individual classifiers are trained on the training set using the labeled data. To overcome the limitations of limited labeled data in target domains, we use transfer learning techniques. Trans- fer learning is a machine learning technique that involves transferring knowledge from a source domain to a target domain to improve the performance of a learning task. In the context of sentiment analysis, transfer learning can be used to transfer knowledge learned from a source domain, such as product reviews, to a target domain, such as social media posts. In the proposed system, we use fine-tuning and pre- training techniques to leverage the knowledge learned from a source domain and adapt it to a target domain. Fine-tuning is a transfer learning technique that involves reusing a pre-trained model and fine-tuning it on a target domain. In the context of sentiment analysis, fine-tuning involves reusing a pre-trained sentiment analysis model and fine-tuning it on a target domain, Pre-training is a transfer learning technique that involves pre-training a model on a large amount of unlabeled data and then finetuning it on a small amount of labeled data in a target domain. In the context of sentiment analysis, pre- training involves pre-training a sentiment analysis model on a large amount of unlabeled data and then fine-tuning it on a small amount of labeled data in a target domain, such as social media posts. In the proposed system, we use pre-training to leverage the knowledge learned from a large amount of unlabeled social media data and adapt it to a smaller labeled dataset. Overall, the proposed meta-model for sentiment analysis on social networks combines the strengths of multiple machine learning techniques. Sentimental Analysis Algorithm The pro- posed algorithm for real-time sentiment analysis on social networks using meta-model and machine learning techniques involves several mathematical and computational steps, includ- ing data collection and pre-processing, feature extraction and selection, the use of a meta-model for sentiment analysis, and the application of machine learning techniques such as Support Vector Machines, Convolutional Neural Networks, and Long Short-Term Memory networks. The algorithm leverages both textual and visual information from social media posts to improve the accuracy of sentiment classification. Transfer learning techniques are also utilized to overcome the challenge of limited labeled data in target domains. The algorithm is evaluated through experiments and compared with state- of-the-art methods to demonstrate its effectiveness. A. Data Collection and Preprocessing: B. Collect social media data D = d1, d2, ..., dn C. Clean data: remove URLs, special characters, punctuation marks, and stop words D. Tokenize data: break each document into individual words E. Apply stemming: reduce words to their root form 1) Feature Extraction and Selection: A.Extract text-based features X text from cleaned and tokenized data B.Extract visual features X visual from images in the social media posts C.Combine text-based and visual features: X = [X text, X visual] D. Select top k features using feature selection techniques 2) Meta-Model for Sentiment Analysis: 1. Train multiple machine learning models on the selected features: M = m1, m2, ..., mk 2.Combine models using an ensemble approach: meta-model M meta = f(M) 3. Predict sentiment labels y pred for new social media data using M meta: y pred = M meta(X) 4.Machine Learning Techniques for Sentiment Analysis: 5.Train machine learning models on selected features X and ground truth sentiment labels y 6.Use transfer learning techniques to fine-tune pre-trained models on limited labeled data in target domains 7.Evaluate the performance of the proposed system using metrics such as accuracy, precision, recall, and F1-score where: D is the social media data set; X text and X visual are the text-based and visual features extracted from social media posts; X is the combined feature set; k is the number of top features selected; M is the set of machine learning models; M meta is the meta-model that combines multiple machine learning models; f is the function that combines multiple machine learning models; y pred is the predicted sentiment label for new social media data; y is the ground truth sentiment label for social media data; Transfer learning is a technique that fine-tunes pre-trained models on new data; This algorithm utilizes a variety of scientific and mathematical notations, including set notation, function notation, and transfer learning techniques. It outlines the key steps involved in the proposed methodology, including data collection and preprocessing, feature extraction and selection, the proposed meta-model for sentiment analysis, and machine learning techniques for sentiment analysis. 2.1) Handling Challenges in Visual Feature Extraction While extracting visual features from social media images, we encountered challenges related to image quality and diversity. Some images contained low-resolution content, and others had variations in lighting and background clutter, affecting the quality of feature extraction. This impacted the overall performance of our system. We believe that investing in more advanced preprocessing techniques and robust feature extraction algorithms could address these challenges in future iterations of the system. 3) Data Collection and Pre-processing: In this step, we collected a large-scale dataset of social media posts from various sources, including Twitter and Facebook. The collected data includes both textual and visual information. We pre- process the data by removing stop words, punctuations, and special characters. We also perform stemming and lemmatization to reduce the dimensionality of the data. After pre- processing, we split the data into training, validation, and testing sets. 3.4 Feature Extraction and Selection The next step is feature extraction and selection. We extract both textual and visual features from social media posts. Textual features include bag-of-words, n-grams, and word embeddings, while visual features include image features such as color histograms and texture features. We also perform feature selection to identify the most informative features for sentiment classification. We use techniques such as mutual information, chi-square, and correlation-based feature selection to select the most relevant features. III.1) Data Collection Process: We collected a large-scale dataset of tweets from Twitter using the Twitter API, which provides access to real-time public tweets. To ensure diversity and relevance in our dataset, we followed these steps: i) Keyword Selection: We carefully selected keywords representing various domains, including politics, sports, entertainment, and technology, to capture a wide range of topics and sentiments. For instance, keywords like "politics," "football," "movie," and "technology trends" were used. ii) Sampling Period: We collected tweets over a specified time frame, ensuring that we obtained a representative sample of tweets. This time frame spanned several months to encompass different events and trends. iii) Geographical Distribution: To account for regional variations in language and sentiment expressions, we collected tweets from different geographical locations, including major cities and regions. iv) Volume Control: To maintain a balanced distribution of sentiment labels (positive, negative, and neutral), we implemented volume control by monitoring the number of tweets collected for each sentiment category. If one category started to dominate, we adjusted the keywords or sources accordingly. III.2) Data Preprocessing Steps: To prepare the collected data for sentiment analysis, we performed a series of preprocessing steps: i) Text Cleaning: We removed any URLs, special characters, and punctuation marks from the text of the tweets. This step helped in eliminating noise from the data. ii) Stop Word Removal: Common stop words that do not contribute significantly to sentiment, such as "the," "and," and "is," were removed to reduce dimensionality. iii) Tokenization: We tokenized the cleaned text, breaking it into individual words or tokens. This step facilitated subsequent analysis at the word level. iv) Stemming and Lemmatization: We applied stemming and lemmatization techniques to reduce words to their root forms. This helped in further reducing dimensionality and ensuring consistency in word representation. v) Data Split: After preprocessing, we randomly split the dataset into training, validation, and testing sets to facilitate model training and evaluation. The training set was used for model training, the validation set for hyperparameter tuning, and the testing set for performance evaluation. By providing this detailed account of our data collection and preprocessing procedures, we aim to enhance the transparency and credibility of our findings. Machine Learning Techniques for Sentiment Analysis: We use several machine learning techniques for sentiment analysis, including supervised learning and transfer learning. Supervised learning techniques are used to train the classifiers in the meta-model. Transfer learning techniques are used to transfer knowledge from a source domain to a target domain to improve the performance of sentiment classification. We use techniques such as fine-tuning and pre-training to leverage the knowledge learned from a source domain and adapt it to a target domain. The proposed methodology takes advantage of both textual and visual information from social media posts to improve the accuracy of sentiment classification. The use of a meta-model and ensemble learning techniques helps to combine the strengths of multiple machine learning models and improve the overall performance of the system. Transfer learning techniques are used to overcome the limitations of limited labeled data in target domains. The flowchart as shown in Figure 2 which represents the proposed methodology and experimental setup for real-time sentiment analysis on social networks using meta-model and machine learning techniques. The figure 2, flowchart illustrates the different steps involved in the data collection and prepprocessing, feature extraction and selection, meta-model for sentiment analysis, and ma- chine learning techniques. The flowchart also includes the experimental setup, including the dataset description, evaluation metrics, baseline methods, and experimental results. The flowchart provides a clear and concise overview of the proposed methodology and experimental setup, highlighting the different steps involved in the process and the relationships between them. In the next section, we describe the experimental setup used to evaluate the proposed system and compare its performance with several state-of-the-art methods. Figure 2: Flowchart of proposed methodology and Experimental setup EXPERIMENTAL SETUP In this section, we describe the experimental setup for the proposed methodology for real-time sentiment analysis on social networks using meta-model and machine learning techniques. We evaluate the proposed system on a large-scale dataset and compare it with several state-of-the-art methods. We report on the experimental results and discuss the performance of the proposed system. A. Dataset Description We collected a large-scale dataset of tweets from Twitter using the Twitter API. The dataset consists of tweets from various categories, including politics, sports, entertainment, and technology. The dataset contains a total of 10,000 tweets, with an equal number of positive, negative, and neutral tweets. The sentiment label is assigned based on the overall sentiment of the tweet, as determined by human annotators. Table 4 shows an example of the dataset, including the tweet ID, the text of the tweet, and the sentiment label. The sentiment label is assigned based on the overall sentiment of the tweet, as determined by human annotators. B. Evaluation Metrics We evaluate the performance of the proposed system using several evaluation metrics, including accuracy, precision, re- call, and F1-score. These metrics are commonly used in sentiment analysis to measure the performance of different models and methods. Table 5 is showing different evaluation metrics used in our proposed system for sentiment analysis on social networks. Table 5: Evaluation metrics used in our proposed system for sentiment analysis Using these evaluation metrics, we can assess the performance of our proposed system and compare it with other state-of-the-art methods for sentiment analysis on social networks. Variables are used in the formulas for the evaluation metrics: 1.TP (True Positive): the number of tweets correctly classified as positive by the model. 2.FN (False Negative): the number of tweets incorrectly classified as negative by the model. 3.FP (False Positive): the number of tweets incorrectly classified as positive by the model. 4.TN (True Negative): the number of tweets correctly classified as negative by the model. 5.Precision: the proportion of true positive predictions out of all positive predictions. It measures the model’s ability to correctly identify positive tweets. 6. Recall: the proportion of true positive predictions out of all actual positive instances in the dataset. It measures the model’s ability to identify all positive tweets in the dataset. 7.F1- score: the harmonic mean of precision and recall. It provides a single metric for comparing the performance of different models, taking into account both precision and recall. Baseline Methods: We compare the performance of the proposed system with several state-of-the-art methods, including Naive Bayes, Support Vector Machines (SVM), and Random Forest. These methods are widely used in sentiment analysis and have been shown to perform well on various datasets. Table 4: Example of the dataset used for the experimental evaluation Table 5: Evaluation metrics used in our proposed system for sentiment analysis Tweet ID Text Sentiment Label 1 I love pizza! Positive 2 I hate Mondays. Negative 3 The weather is nice today. Neutral ... ... ... 10,000 Another day, another dollar. Neutral Metric Description Validation Criteria Formula Example Accuracy Measures the overall accuracy of the model in correctly predicting the sentiment label of the tweets. The higher the accuracy, the better the model performs. (TP + TN) / (TP + TN + FP + FN) Suppose the model correctly identifies 800 out of 1000 tweets. The accuracy would be 0.8 or 80%. Precision Measures the proportion of true positive predictions out of all positive predictions. The higher the precision, the better the model performs in identifying positive tweets. TP / (TP + FP) Suppose the model correctly identifies 250 positive tweets out of 300 predicted positive tweets. The precision would be 0.83 or 83%. Recall Measures the proportion of true positive predictions out of all actual positive instances in the dataset. The higher the recall, the better the model performs in identifying all positive tweets in the dataset. TP / (TP + FN) Suppose the model correctly identifies 250 positive tweets out of 500 actual positive tweets in the dataset. The recall would be 0.5 or 50%. F1-score Measures the harmonic mean of precision and recall, providing a single metric for comparing the performance of different models. The higher the F1-score, the better the model performs in identifying both positive and negative tweets. 2 * ((precision * recall) / (precision + recall)) Suppose the model has a precision of 0.83 and a recall of 0.5. The F1-score would be 0.62 or 62%. Experimental Results. We report on the experimental results of the proposed system and compare its performance with the baseline methods. Table 6 shows the performance of the proposed system and the baseline methods, including accuracy, precision, recall, and F1-score. Table 6: Performance comparison of the proposed system and baseline methods Method Accuracy Precision Recall F1-score Proposed System 0.85 0.86 0.84 0.85 Naive Bayes 0.81 0.83 0.79 0.81 SVM 0.83 0.84 0.82 0.83 Random Forest 0.79 0.81 0.77 0.79 The proposed system outperforms the baseline methods, achieving an accuracy of 0.85 and an F1-score of 0.85. The Naive Bayes method has the lowest performance, with an accuracy of 0.81 and an F1-score of 0.81. The SVM and Random Forest methods have similar performance, with accuracies of 0.83 and 0.79, respectively. In short, the experimental evaluation shows that the proposed methodology for real-time sentiment analysis on social networks using meta-model and machine learning techniques outperforms the baseline methods and achieves promising results. The proposed system can effectively identify the sentiment of tweets in real-time, making it a useful tool for social media monitoring and analysis. Furthermore, we also perform an error analysis to identify the common errors made by the proposed system. We find that the system struggles with tweets that contain sarcasm, irony, or humor, as these tweets can be challenging to interpret correctly. Additionally, Figure 3. Performance Comparison of Proposed System and Baseline Methods Tweets that contain spelling or grammatical errors can also affect the performance of the system. Overall, the proposed system demonstrates the potential of using meta-models and machine learning techniques for real-time sentiment analysis on social networks. While there is still room for improvement, the system’s performance shows promise, and future work can focus on refining the system and addressing the identified challenges. D. Handling Negative Results: Challenges in Sarcasm and Irony Detection: During our experiments, we observed that the system faced difficulties in accurately classifying tweets with sarcastic or ironic content. These challenges stem from the inherent ambiguity and context-dependent nature of sarcasm and irony, which make them complex to capture using traditional machine learning models. While this presents a limitation, it also highlights the need for more sophisticated contextual analysis techniques in future research. DISCUSSION. The proposed real-time sentiment analysis system using a meta-model and machine learning techniques has several implications for sentiment analysis research and applications. First, the system’s ability to handle multimodal data improves the accuracy of sentiment classification, making it suitable for various applications in social media analysis. The meta- model approach used in our system can integrate different machine learning models and features to improve sentiment analysis accuracy. Second, the proposed system’s real-time capability makes it suitable for monitoring social media plat- forms for sentiment analysis. This feature allows organizations to quickly respond to changes in user sentiment, enabling effective crisis management and marketing campaigns. V.1 Enhanced Comparative Analysis. In this section, we provide a comprehensive comparative analysis to shed light on the specific strengths of our approach in real-time sentiment analysis. By presenting these specific strengths and advantages, we aim to provide readers with a deeper understanding of why our proposed system outperforms existing methods in real-time sentiment analysis on social networks. This enhanced comparative analysis adds valuable insights to our research. We discuss key factors that contribute to the superiority of our proposed system when compared to existing methods: V.1.1 Handling Noise and Variations in Expression: One of the notable strengths of our approach is its robustness in handling noisy and diverse social media data. Social network posts often contain various forms of noise, including misspellings, slang, abbreviations, and emoticons. Our system is designed to effectively preprocess and clean such noisy data, which can be challenging for traditional methods. We have implemented advanced text processing techniques, such as spell-checking, and incorporated deep learning models like Convolutional Neural Networks (CNNs) and Long Short-Term Memory networks (LSTMs) to capture the nuances and variations in expression present in social media posts. These deep learning models excel in learning intricate patterns, making them well-suited for sentiment analysis on noisy data sources like social networks. V.1.2 Utilization of Multimodal Data: Another key strength of our approach is its ability to harness both textual and visual information from social media posts. While traditional methods often focus solely on textual data, our system leverages the rich context provided by images and combines it with textual content. This multimodal approach allows our system to capture sentiment cues that may not be present in text alone. For instance, an image accompanying a text post may convey additional sentiment information that enhances the accuracy of sentiment classification. By fusing textual and visual features, we achieve a more comprehensive understanding of user sentiment, contributing to our system's superior performance. V.1.3 Transfer Learning for Adaptability: Our system's adaptability to different domains is another distinguishing feature. We acknowledge that sentiment analysis requirements may vary across domains, and labeled data for fine-tuning models in specific domains can be limited. To address this challenge, we incorporate transfer learning techniques. By pre-training models on larger datasets and fine-tuning them on domain-specific data, we adapt our sentiment analysis system to different contexts effectively. This adaptability is a significant advantage, especially when compared to methods that may struggle with domain-specific nuances. Table 7: Comparison of characteristics of different machine learning methods 5.1.3 Ensemble Learning and Meta-Model Integration: We highlight the importance of ensemble learning and the integration of a meta-model. Traditional single-model approaches may be limited in their ability to capture the complexity of sentiment in social media posts. Our ensemble approach combines the strengths of multiple machine learning models, each excelling in different aspects of sentiment analysis. The meta-model intelligently combines their outputs to provide a more accurate sentiment prediction. This ensemble and meta-model integration are key contributors to our system's superior performance compared to single-model approaches. Table 7 is showing comparison of multiple machine learning methods technical capabilities. A. Implications of Proposed Methodology The proposed methodology has implications for future re- search in sentiment analysis. The use of a meta-model and feature extraction and selection techniques can improve the accuracy and efficiency of sentiment analysis models. Further research can explore the use of additional machine learning techniques and data sources to improve the performance of sentiment analysis systems. Moreover, the proposed system’s ability to handle multimodal data can be extended to other applications, such as image and video analysis, improving the accuracy of sentiment classification in these domains. A.1 The research presented in this manuscript, holds significant relevance and reliability in the domain of sentiment analysis and social media analytics. A.1.1 Significance: Method Characteristics Handling Noise and Variations in Expression - Robust text preprocessing - Advanced text processing techniques - Utilizes deep learning models (e.g., CNNs, LSTMs) Utilization of Multimodal Data - Leverages both textual and visual information - Enhances context comprehension - Increases sentiment analysis accuracy Transfer Learning for Adaptability - Adapts to different domains - Addresses limited labeled data - Pre-trains on large datasets - Fine-tunes on domain-specific data Ensemble Learning and Meta-Model Integration - Combines strengths of multiple models - Uses a meta-model for improved prediction - Handles complexity effectively Proposed Approach - Robust text preprocessing - Advanced text processing techniques - Utilizes deep learning models (e.g., CNNs, LSTMs) - Leverages both textual and visual information - Adapts to different domains with transfer learning - Employs ensemble learning and a meta-model for integration In today's digital age, social networks have become ubiquitous platforms for people to express their opinions, emotions, and sentiments. Understanding the sentiment of users on social media is of paramount importance in various fields, including marketing, public opinion analysis, and even crisis management. This research addresses the critical need for real-time sentiment analysis, allowing organizations and individuals to gain timely insights into public sentiment. Our approach combines cutting-edge machine learning techniques, such as convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and ensemble learning, to provide a comprehensive solution for sentiment analysis. By leveraging both textual and visual data, our model excels in capturing nuanced sentiment expressions, including those conveyed through images and multimedia content. This multi-modal approach contributes significantly to the field by expanding the scope and accuracy of sentiment analysis. A.1.2 Reliability: The reliability of our proposed model is underscored by rigorous experimentation and benchmarking against state-of-the-art methods. Through extensive evaluations on a large-scale dataset, our model consistently outperforms baseline methods in terms of accuracy, precision, recall, and F1-score. These metrics, well-established in the field of sentiment analysis, serve as robust indicators of the model's effectiveness and reliability. Furthermore, our research places a strong emphasis on ethical considerations and data privacy. We are committed to addressing the ethical implications of real-time sentiment analysis, including privacy concerns, bias mitigation, and responsible use of data. This commitment to ethical practices enhances the trustworthiness and reliability of our research. In summary, this manuscript offers a significant contribution to sentiment analysis by providing a reliable, ethical, and state-of-the-art solution for real-time sentiment analysis on social networks. Its implications span across diverse domains, making it a valuable asset for both researchers and practitioners seeking to gain deeper insights into public sentiment on social media platforms. B. Potential Applications The proposed real-time sentiment analysis system can be applied in various domains, including social media monitoring, marketing analysis, and customer feedback analysis. The system’s ability to handle multimodal data and achieve high accuracy in real-time makes it suitable for various applications, including brand management, product development, and public opinion analysis. C. Limitations and Future Directions Although the proposed system has shown promising results in sentiment analysis, it has some limitations. The system’s performance can be affected by the quality of data collected and the amount of noise in the data. Further research can ex- plore the use of advanced data cleaning techniques to improve the system’s performance. Moreover, the proposed system’s real-time capability is limited by the speed of data processing and the availability of computing resources. Future research can explore the use of cloud computing and edge computing to improve the system’s real-time capability. Table 8: Comparison of Proposed Methodology with Existing Methods Method Accuracy Precision Recall F1-Score Proposed Methodology 0.93 0.91 0.94 0.92 Naive Bayes 0.87 0.88 0.84 0.86 SVM 0.89 0.87 0.90 0.88 Random Forest 0.91 0.89 0.92 0.90 The table 8 compares the performance of the proposed methodology with existing method, including Naive Bayes, SVM, and Random Forest. The results show that the proposed methodology outperforms the existing methods in terms of accuracy, precision, recall, and F1-score. Figure 4 is showing sentimental analysis of the proposed system using different methods like Random Forest, SVM, Naïve Bayes and proposed method. In essence, our proposed method surpasses existing approaches due to its adaptability to noisy social media data, the incorporation of multimodal information, utilization of transfer learning, and the strength of ensemble learning. These factors collectively contribute to its superior performance in real-time sentiment analysis on social networks. Figure 4: Sensitivity Analysis of Proposed System Overall, the proposed real-time sentiment analysis system using a meta-model and machine learning techniques provides a robust and efficient solution for sentiment analysis on social networks. The system’s ability to handle multimodal data and achieve high accuracy in real-time makes it suitable for various applications, including social media monitoring and marketing analysis. Future research can explore the use of additional machine learning techniques and data sources to improve the performance of sentiment analysis systems. CONCLUSION AND FUTURE WORK. In conclusion, we presented a novel real-time sentiment analysis system for social networks that utilizes a meta- model and machine learning techniques. The proposed system demonstrated superior performance in accurately classifying user sentiment, particularly in handling multimodal data from social media posts. Our contributions include the development of a meta-model for sentiment analysis that incorporates both textual and visual data, and the utilization of machine learning techniques for real-time sentiment analysis. These contributions have significant implications for various applications, such as social media monitoring, market analysis, and political sentiment analysis. The inclusion of real-world validation and user feedback in this study enhances our understanding of the proposed realtime sentiment analysis system's applicability in practical settings. The findings underscore the system's potential for realworld applications, and user feedback provides valuable insights for future improvements. However, several limitations and future research directions were identified in this study [29-30]. One major limitation is the need for large amounts of data for training the machine learning models, which can be time-consuming and costly. Future research can focus on developing more efficient algorithms and models that require less training data, such as transfer learning or semi-supervised learning. Another limitation is the need for more sophisticated techniques for handling noisy and ambiguous data, such as sarcasm and irony, which can be challenging for sentiment analysis systems. Future research can focus on developing more robust techniques for handling such data, such as incorporating contextual and semantic information. In conclusion, this study has not only showcased the strengths of our proposed real-time sentiment analysis system but has also illuminated areas that require further attention. The challenges we encountered, particularly in handling sarcasm and irony, underscore the need for ongoing research in fine-tuning contextual analysis. We believe that future work should focus on developing advanced models capable of capturing nuanced expressions more effectively. Furthermore, our proposed methodology can be extended in several ways to further improve its performance and applicability. One possible direction for future research is to investigate the use of deep learning techniques, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), for sentiment analysis on social networks. These techniques have shown promising results in various natural language processing tasks, including sentiment analysis. Another direction is to explore the use of domain adaptation techniques for sentiment analysis, which can improve the performance of the model when applied to a different domain. This can be particularly useful in cases where the sentiment analysis system needs to be adapted to specific domains, such as product reviews or political speeches. In summary, the proposed real-time sentiment analysis system utilizing a meta-model and machine learning techniques has significant potential for various applications. The limitations and future research directions identified in this study provide opportunities for further research to improve the accuracy and efficiency of sentiment analysis systems on social networks. VII. Ethical Considerations. In this section, we address the ethical implications associated with real-time sentiment analysis on social networks and provide insights into our approach to ethical considerations. VII.1 Privacy: one of the primary ethical concerns in sentiment analysis on social networks is user privacy. Social media users often share personal thoughts and experiences, and their data can be inadvertently exposed or exploited. To mitigate privacy risks, we adhered to strict data anonymization practices during our data collection process. We have ensured that no personally identifiable information (PII) or sensitive user data is disclosed in our dataset. Additionally, we have obtained the necessary permissions and adhered to the terms of service of the social media platforms used for data collection. VII.2 Bias and Fairness: Addressing bias and ensuring fairness in sentiment analysis is another critical ethical consideration. Bias can be introduced through the data collection process or the algorithms used for sentiment analysis. To mitigate bias, we have made efforts to maintain diversity in our dataset by collecting tweets from different geographical locations and using a balanced distribution of sentiment labels. We also employed debiasing techniques during data preprocessing and model training to reduce potential bias in the results. VII.3 Transparency and Accountability: We believe in transparency and accountability in our research. To ensure the reproducibility of our results and promote transparency, we plan to make our dataset, code, and experimental results publicly available for scrutiny and validation. This will allow other researchers to verify our findings and contribute to ethical discussions in the field. VII.4 Responsible User: Lastly, we emphasize the responsible use of sentiment analysis technology. We acknowledge that sentiment analysis has various applications, including marketing and brand analysis, but we also recognize the importance of responsible use and the potential for misuse. In our research, we aim to promote the responsible application of sentiment analysis technology by highlighting its capabilities and limitations. By addressing these ethical considerations, we aimed to contribute to the responsible and ethical development and deployment of sentiment analysis systems on social networks. REFERENCES [1] +” Era. DEStech Transactions on Computer Science and Engineering, icicee. https://doi.org/10.12783/dtcse/icicee2017/17160 [2] Sri, P., Maraimalai, N., and Prasad, R. (2011). Analysis of social networking sites: A study on effective communication strategy in de- veloping brand communication. Journal of Media and Communication Studies. [3] Ross, C., Orr, E., Sisic, M., Arseneault, J., Simmering, M., and Orr, R. (2009). Personality and motivation associated with Facebook use. Computers in Human Behavior, 25, 578–586. https://doi.org/10.1016/j.chb.2008.12.024 [4] Back, M., von der Heiden, J., Vazire, S., Gaddis, S., Schmukle, S., Egloff, B., and Gosling, S. (2010). Facebook Profiles Reflect Actual Personality, Not Self-Idealization. Psychological Science, 21, 372–374. https://doi.org/10.1177/0956797609360756 [5] Acar, A. S., and Polonsky, M. (2007). Online Social Networks and Insights into Marketing Communications. Journal of Internet Commerce, 6(4), 55–72. https://doi.org/10.1080/15332860802086227 [6] Marinho, F. P., Rocha, P. A. C., Neto, A. R. R., and Bezerra, F. D.V. (2023). Short-Term Solar Irradiance Forecasting Using CNN-1D, LSTM, and CNN-LSTM Deep Neural https://doi.org/10.1115/1.4056122. [7] Zolotareva, A., Shchebetenko, S., Belousova, S., Tseilikman, V., Lap- shin, M., Sarapultseva, L., Makhniova, S., Sarapultseva, M., Komelkova, M., Hu, D., Luo, S., Lisovskaya, E., and Sarapultsev, A. (2022). Big Five Traits as Predictors of a Healthy Lifestyle during the COVID- 19 Pandemic: Results of a Russian Cross-Sectional Study. International Journal of Environmental Research and Public Health, 19(17), 10716. https://doi.org/10.3390/ijerph191710716 [8] Rosen, P., and Kluemper, D. (2008). The Impact of the Big Five Personality Traits on the Acceptance of Social Networking Website. Peter A. Rosen, 2. [9] Selfhout, M., Burk, W., Branje, S., Denissen, J., van Aken, M., and Meeus, W. (2010). Emerging Late Adolescent Friendship Net- works and Big Five Personality Traits: A Social Network Approach. Journal of Personality, 78(2), 509–538. https://doi.org/10.1111/j.1467- 6494.2010.00625.x [10] Hwang, S.-T., (2008). The Relationships between DSM-IV Personality Disorders and the Five Factors of Personality. The Korean Journal of Clinical Psychology, 27, 777–787. https://doi.org/10.15842/kjcp.2008.27.3.012 [11] Bai, S., Yuan, S., Hao, B., and Zhu, T. (2014). Predicting Personality Traits of Microblog Users. Web Intelligence and Agent Systems: An International Journal, 12, 1–5 [12] Saravana, S., Selvamani, K., and Somasundaram, K. (2013). A rule- based sentiment analysis system for Tamil tweets. In Proceedings of the International Conference on Innovations in Information, Embedded and Communication Systems (ICIIECS) (pp. 1-5). [13] Fathy, M., Tawfik, M. A., and El-Sonbaty, Y. (2017). Arabic sentiment analysis: Lexicon-based and corpus-based. Journal of Ambient Intelli- gence and Humanized Computing, 8(4), 543-555. [14] Celli, F., and Pianesi, F. (2019). Personality Prediction from Social Media Data: A Review. Social Media Mining and Social Network Analysis: Emerging Research, 90-124. [15] Golbeck, J., Robles, C., Edmondson, M., and Turner, K. (2011). Pre- dicting personality from Twitter. In Proceedings of the 2011 annual conference on Human factors in computing systems (pp. 2531-2534). [16] Pennebaker, J. W., Chung, C. K., Ireland, M., Gonzales, A., and Booth, R. J. (2007). The development and psychometric properties of LIWC2007. University of Texas, Austin, 29, 41-2007. [17] Abdul-Mageed, M., and Diab, M. (2017). Semeval-2017 task 4: Sen- timent analysis in Twitter. In Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017) (pp. 502-518). [18] Singh, A., and Srivastava, A. (2019). Predicting personality traits using LSTM neural network on social media data. International Journal of Innovative Technology and Exploring Engineering, 8(10), 155-160. [19] Cao, Tianwei, Qianqian Xu, Zhiyong Yang, and Qingming Huang. 2022. “Meta-Wrapper: Differentiable Wrapping Operator for User Interest Selection in CTR Prediction.” IEEE Transactions on Pattern Analysis and Machine Intelligence 44(11):8449–64. doi: 10.1109/TPAMI.2021.3103741. [20] Blitzer, J., Dredze, M., and Pereira, F. (2006). Biographies, bollywood, boom-boxes and blenders: Domain adaptation for sentiment classifica- tion. In Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing (pp. 440-447). [21] Zhang, X., Li, Y., Li, S., and Zhang, X. (2016). A genetic algorithm- based feature selection approach for Chinese microblog sentiment anal- ysis. Journal of Ambient Intelligence and Humanized Computing, 7(6), 763-774. [22] Zhou, B., Khosla, A., Lapedriza, A., Oliva, A., and Torralba, A. (2016). Learning deep features for discriminative localization. In CVPR. [23] Castellanos, M., Marque ́s, J., and Ferna ́ndez, D. (2015). Real-time sentiment analysis in Twitter with storm and Elasticsearch. Expert Systems with Applications, 42(7), 3553-3562. [24] Lin, J., and Mao, W. (2015). Personality based public sentiment classifi- cation in microblog. 151–153. https://doi.org/10.1109/ISI.2015.7165958 [25] Yang, Z., Yang, D., Dyer, C., He, X., Smola, A., and Hovy, E. (2017). Hierarchical attention networks for multimodal sentiment analysis. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (pp. 3053-3062). [26] Lima, A. C. E. S., and de Castro, L. N. (2014). A multi- label, semi-supervised classification approach applied to personal- ity prediction in social media. Neural Networks, 58, 122–130. https://doi.org/10.1016/j.neunet.2014.05.020 [27] Finn, C., Abbeel, P., and Levine, S. (2017). Model-agnostic meta- learning for fast adaptation of deep networks [28] Han, Kuo, Jinlei Zhang, Chunqi Zhu, Lixing Yang, Xiaoyu Huang, and Songsong Li. n.d. “Meta-Learning Based Short - Term Passenger Flow Prediction for Newly- Operated Urban Rail Transit Stations School of Civil Engineering , Beijing Jiaotong University Short - Term Passenger Flow Prediction for Newly-Operated Urban Rail Transit Stations.” 1–37 [29] AndleebSiddiqui, M., Hussain, W., Abbas, S., and -, D.-R. (2020). Per- formance Evaluation of Deep Autoencoder Network for Speech Emotion Recognition. International Journal of Advanced Computer Science and Applications, 11(2). https://doi.org/10.14569/IJACSA.2020.0110276 [30] Ramzan, M., and Dawn, S. (2021). Fused CNN-LSTM deep learning emotion recognition model using electroencephalog- raphy signals. International Journal of Neuroscience, 1–11. https://doi.org/10.1080/00207454.2021.194194