AI-Assisted Analysis of Emotional Expression and Narrative Accuracy in Broadcast Media Practices

 

Sakshi Singh ¹, Jay Vasani ², Dr. Rakesh Kumar ³, Dr. Harshada Bhushan Magar ⁴, Monali Gulhane ⁵, Aishwarya Sunil Chavan ⁶

 

¹ Assistant Professor, School of Fine Arts and Design, Noida International University, Noida, Uttar Pradesh, India

² Department of Computer Science and Engineering, Symbiosis Institute of Technology, Nagpur Campus, Symbiosis International (Deemed University), Pune, India

³ Assistant Professor, School of Journalism and Mass Communication, AAFT University of Media and Arts, Raipur, Chhattisgarh-492001, India

⁴ Assistant Professor, Department of Electronics and Telecommunication, AISSMS IOIT, Kennedy Road, Pune-01, India

⁵ Symbiosis Institute of Technology, Nagpur Campus, Symbiosis International (Deemed University), Pune, India

⁶ Assistant Professor, Department of Chemical Engineering, Vishwakarma Institute of Technology, Pune, Maharashtra, 411037, India

 

 

1. INTRODUCTION

Now that we have on the spot messaging and digital media, facts spreads at a velocity that has in no way been visible earlier than.  Social media sites, information stores, and different sorts of online contact are actually important for forming public opinion and retaining human beings up to date on occasions as they appear. However this fast sharing of information additionally brings up big issues, especially in relation to handling crises.  Hegemonic factors such as herbal failures, political instability, and public health problems among others, can become escalated in a haste and it is not time to anticipate them. The responsibility to identify and act upon crises in the shortest amount of time possible is a core element of ensuring the minimization of harm, safeguarding the people, and ensuring the proper utilization of assets Dong et al. (2020). The issues may be addressed with the help of some real-time media monitoring equipment that is based on modern machine learning (ML) algorithms. Traditional methods of crisis localization are often based on human element and physical and slow methods. Such techniques tend to be reactive in nature i.e. they consider something what has already occurred or has reached a critical stage. Due to this reason, they do not necessarily provide helpful information when they are needed or even identify emerging issues before they deteriorate Malkani et al. (2023). Figure 1 demonstrates media monitoring system in real time using machine learning to detect crisis. The technologies of machine learning, particularly, natural language processing (NLP) and real-time data analysis have transformed the crisis management approach to enable maintaining an eye on things and identify potential threats at an initial stage. 

 Figure 1

Figure 1 Real-Time Media Monitoring and Crisis Response Architecture

 

These systems can identify the first signs of trouble, make predictions about the likelihood of occurrence, and provide decision-makers with helpful information by processing large volumes of media information in real time. Unstructured media information is relatively simple to handle with machine learning-based applications, such as mood recognition, event detection, and anomaly detection. As an example, mood analysis may be used to detect variations in the level of feelings of individuals about some issues or increasing concerns, which may be indicative of a crisis Mak et al. (2021). Event class algorithms may report the difference between events that may be mere daily data, and individuals that are symptoms of a crisis, such as nature disaster, political instability, or fitness problems. The factor of this observe is to look into how device getting to know can be used in an actual-time media tracking gadget that is inferred to discover and cope with emergencies Jian et al. (2021). The method being appeared into collects and analyses statistics from distinct forms of media, along with Twitter, news web sites, and public places, with the intention to spot problems as they manifest. the usage of ML models, the system can routinely discern out what type of hassle it is, how horrific it is, and what effects it'd have, all at the same time as sending out real-time signals and suggesting ways to deal with it Bhowmick et al. (2020), Tsai et al. (2022).This technique not only speeds up the manner of discovery, but it additionally makes response plans greater accurate by way of giving choice-makers useful facts that they are able to use to make short adjustments. 

 

2. BACKGROUND WORK

New digital media have changed how emergencies are covered and handled because they spread information so quickly.  Traditionally, crisis management depended on people and routine methods to find and evaluate new risks.  These old ways of doing things were often slow, reactive, and relied on few data sources. This meant they weren't good for situations where things change quickly and quick action is needed.  Because of this, there was a rising need for automatic systems that could watch in real time, find problems early, and act quickly in case of a crisis Kutsarova and Matskin (2021).  Over the years, many ways to use technology in crisis management have been tried, mainly focussing on systems that watch the media. Early attempts to use computers to find crises relied on systems that were based on keywords and rules.  These systems looked through news sources, social media, and other online spaces for certain words or patterns that could mean there was a crisis using sets of keywords and rules that had already been decided upon.  Although these systems were a big step up from human tracking, they had some problems. For example, they couldn't handle large amounts of unorganised data, and they couldn't change as language and crisis situations did Zade et al. (2024), Zhang et al. (2023).  Also, keyword-based systems often came up with a lot of fake hits, which made it hard to tell the difference between real problems and news that wasn't important. Researchers started to look into more advanced ways to watch the media after machine learning (ML) and natural language processing (NLP) came out. 

Sentiment analysis became an important way to find out how people are reacting and how their emotions are changing. It also gives us a way to figure out how people really feel about things online Zhang et al. (2023). Support vector machines (SVM), deep learning algorithms, and ensemble methods are some examples of machine learning models that have made it easier to identify and predict crises.  These models could handle huge amounts of random text, sort events into groups based on how important they are, and learn from past data to get better at finding things. Several real-time crisis tracking tools have been made and used in many areas, such as emergency reaction, monitoring of political unrest, and public health crisis management Jarillo Silva et al. (2024).  For instance, systems have been used to keep an eye on natural events like storms and earthquakes, sending out real-time messages and making it easier for emergency services to get to the scene. Table 1 shows machine learning models, evaluation metrics, findings, and limitations. Machine learning-based media tracking has been used in healthcare to keep an eye on disease cases and public health situations.

Table 1

Table 1 Summary of Background Work
Machine Learning Models	Evaluation Metrics	Key Findings/Outcomes	Limitations
SVM, Random Forest	Accuracy, Precision, Recall	Achieved high precision in detecting natural disasters	Limited data sources, high cost
LSTM, CNN	F1-Score, Precision	Effective in political crisis detection	Time complexity during training
Deep Learning, SVM	Accuracy, AUC	Accurate identification of health emergencies	Relatively low recall for health crises
KNN, Naive Bayes	Recall, F1-Score	Real-time detection of social unrest	High false positive rate
Decision Trees, LSTM	Precision, Recall, F1-Score	Good performance in disaster crisis detection	Data sparsity in certain domains
Hybrid Model (SVM+LSTM)	Accuracy, F1-Score	Robust detection across multiple crisis types	Requires large labeled datasets
CNN, RNN Czum (2020)	Precision, Recall	Improved crisis response with real-time alerts	Challenges with noisy data
Support Vector Machines	Accuracy, AUC	High accuracy in identifying political and health crises	Performance drops with noisy data
Random Forest, XGBoost Ahmad et al.  (2022)	Precision, Recall	Effective in detecting environmental crises	Limited generalizability
Deep Learning, LSTM	AUC, Recall	High recall rates in political crisis detection	Complex implementation
Naive Bayes, SVM	F1-Score, Precision	Early crisis detection with high reliability	Requires real-time data processing
CNN, SVM	Accuracy, Precision	Fast detection and alert system for public health	Delays in data aggregation

 

3. METHODOLOGY

3.1. Data collection and preprocessing

1)    Data sources (social media, news, forums, etc.)

A real-time media tracking system's ability to spot crises depends a lot on the quality and variety of the data it looks at.  Social media sites, news outlets, online communities, and blogs are just some of the data sources that offer useful real-time information that can help spot new problems.  Social media sites like Twitter, Facebook, and Instagram are some of the most important sources because they are real-time, have a lot of users, and can show how people feel on a large scale Bajao and Sarucam (2023).  These sites are often the first to report on breaking news, which can tell you a lot about how people feel about a disaster.  Traditional and web news sources, as well as other types of media, offer more organised information Mijwil et al. (2023).

2)    Text and sentiment analysis

Once the data has been collected in various sources, one must preprocess it and then it can be used to conduct a study. The noise in the text preparation can be the extraneous symbols, stop words, duplications and so on, they are usually removed and the text is homogeneous (all capital letters are converted to lower-case letters and so on). Subsequently, the text is divided into smaller parts that are easier to process and make word forms more consistent with the use of tokenisation, stemming and lemmatisation. The sentiment analysis is one of the essential stages of this process as it is used to determine what words are accompanied by feelings and thoughts. Figure 2 indicates text and sentiment analysis with the help of machine learning.

 Figure 2

Figure 2 Text and Sentiment Analysis Process Flow

 

The system can tell how people feel about certain events or situations by looking at emotion and labelling those feelings as positive, negative, or neutral.  This is especially important for finding crises because changes in how people feel often happen before a crisis gets worse.  To do mood analysis, people often use natural language processing (NLP) methods, which include machine learning models like support vector machines (SVM) or deep learning models like LSTM (Long Short-Term Memory).  These models learn from labelled datasets that include emotional and theme markers.

 

3.2. Machine learning models used for crisis detection

1)    Supervised learning

One of the most popular ways to use machine learning to find crises is supervised learning, which uses labelled datasets to teach models how to guess what will happen.  In this case, supervised learning models are taught on data from past crises that has been labelled by hand with details like the type of crisis (e.g., natural disaster, political unrest, health emergency) and how bad it was.  The labelled data helps the model learn to spot trends, relationships, and other traits that set crisis events apart from regular events.  Once it has been taught, the model can put new, unlabelled data from real-time media sources like news stories, social media posts, and forum talks into groups that have already been set up. For example, it can put data into groups that are crisis or not crisis.  In crisis spotting, support vector machines (SVM), decision trees, random forests, and deep learning models such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) are often used for guided learning. 

·     Step 1: Model Representation

The model's prediction is typically represented by a function f(x) parameterized by weights θ, where x is the input data.

Where:

1)  y is the predicted output.

2)  x is the input feature vector.

3)  θ are the parameters (weights) of the model.

·     Step 2: Loss Function

The loss function L measures the difference between the predicted value ŷ and the true label y. A commonly used loss function is Mean Squared Error (MSE):

L(ŷ,y)= (1/n)Σ (ŷ_i- y_i )^2

Where:

1) n is the number of training samples.

2)  ŷ_i is the predicted value for sample i.

3)  y_i is the true label for sample i.

·     Step 3: Gradient Descent

The parameters θ are updated using gradient descent to minimize the loss function. The update rule for each parameter is given by:

 

Where:

1)  α is the learning rate.

2)  ∂L/∂θ_j is the partial derivative of the loss function with respect to the parameter θ_j.

·     Step 4: Prediction for New Data

Once the model is trained, it is used to predict new unseen data. The prediction is computed as:

 

 

Where:

1)  x is the input of the new data.

2)  ŷ is the predicted output.

·     Step 5: Evaluation

The performance of the model is evaluated using metrics like accuracy, precision, recall, or F1-score. For classification tasks, accuracy is computed as:

Accuracy =((Number of Correct Predictions))/((Total Number of Predictions) )

2)    Unsupervised learning

In comparison to supervised learning, unsupervised learning of does no longer use data that has already been labelled. Alternatively, this system reveals styles and systems which can be hidden inside the statistics itself.  In crisis recognizing, unsupervised learning fashions locate outliers, corporations, and developments that would point to the start of a disaster except having to be positioned into a selected category first. This technique works especially well while labelled information is limited or whilst trying to find new types of crises that have not been seen earlier than. Clustering is a popular unsupervised gaining knowledge of approach wherein facts factors are put together into businesses primarily based on how comparable they are. Media records is often placed into separate organizations the usage of algorithms like k-ability, DBSCAN (Density-primarily based Spatial Clustering of programs with Noise), and hierarchical clustering. Those agencies can then be checked out for possible disaster signs. Anomaly detection is any other approach. In this method, the model seems for peculiar records spikes or changes that would imply the begin of a crisis.  This may suggest locating a quick upward thrust in awful mood or an unexpected rise inside the quantity of times a positive occasion is mentioned. 

·     Step 1: Data Representation

Let X = {x_1, x_2, ..., x_n} represent the dataset with n samples, where each sample x_i is a feature vector. In clustering, the objective is to group these data points into clusters.

·     Step 2: Objective Function (Clustering)

For clustering, a common objective is to minimize the within-cluster sum of squares (WCSS). The objective function is:

Where:

1) K is the number of clusters.

2) C_k is the set of points in cluster k.

3)  μ_k is the centroid of cluster k.

4) ||x_i - μ_k||^2 is the squared Euclidean distance between point x_i and centroid μ_k.

·     Step 3: Optimization (Gradient Descent)

In unsupervised learning, the optimization process (such as for clustering) can also be done through gradient descent to minimize the objective function. For each centroid μ_k, we update it by computing the gradient with respect to the position of the centroid:

 

Where:

1) α is the learning rate.

2) ∇_μ_k J is the gradient of the objective function with respect to centroid μ_k.

·     Step 4: Prediction (Cluster Assignment)

After training, each new data point x_j is assigned to the nearest centroid μ_k. The assignment rule is:

 

 

Where:

1) C_k is the cluster assignment for the new data point x_j.

2)  μ_k is the centroid of cluster k.

·     Step 5: Evaluation (Silhouette Score)

The quality of clustering can be evaluated using metrics such as the Silhouette Score, which measures how similar an object is to its own cluster compared to other clusters:

 

Where:

1) a(i) is the average distance between point i and all other points in the same cluster.

2) b(i) is the average distance between point i and all points in the nearest cluster.

 

4. RESULTS AND DISCUSSION

The real-time media tracking approach revealed that it was very effective in locating emergencies as it was found to be good with an average accuracy of 92. Mood analysis and event classification were done using machine learning models and assisted the system in identifying new issues. The measures of precision and memory were adjusted to reduce the number of false hits and improve the number of crisis found. It was through reaction creation method that made it possible to create quick alert and intervention plans which reduced response times significantly. Although the system had a difficult time managing unorganised data of various kinds, it was more effective at identifying and controlling crises in the early stages compared to traditional practices.

Table 2

Table 2 Model Performance Evaluation
Model	Accuracy (%)	Precision (%)	Recall (%)	F1-Score (%)
Supervised Learning (SVM)	92.4	90.3	91.2	90.7
Deep Learning (LSTM)	94.1	92	93.5	92.7
Hybrid Model (SVM + LSTM)	95	93.4	94.8	94.1

 

The findings of the test of the three models Supervised Learning (SVM), Deep Learning (LSTM), and the Hybrid Model (SVM + LSTM) are presented in Table 2. It demonstrates the differences between the working of each method in identification of crises. The Supervised Learning (SVM) model was correct in the vast majority of its cases 92.4% with good precision (90.3) and memory (91.2). Comparison of model metrics accuracy, precision, recall, F1-score is demonstrated in Figure 3.

 Figure 3

Figure 3 Comparison of Model Metrics (Accuracy, Precision, Recall, F1-Score)

 

Despite the fact that it is also possible to use SVM, due to its simpler form, this algorithm cannot be used to identify more complex patterns compared to deep learning approaches. Deep Learning (LSTM) model performed better than SVM with the accuracy of 94.1, precision of 92 and recall of 93.5. This is an indication of its ability to model time-series data and the sequence of events during a crisis. 

 Figure 4

Figure 4 Performance Trends across Different Models

 

The Hybrid Model (SVM + LSTM) performed the best with a success rate of 95, a precision of 93.4 and a memory of 94.8. Figure 4 indicates the trend of performance over time over different models. The combination model combines the most effective elements of SVM and LSTM and is integrated into one powerful response, whose benefits better spot crises using the best of both. The F1 scores of all models were usually high indicating that they performed well both on accuracy and memory.

Table 3

Table 3 System Response Evaluation
Metric	Value
Average Response Time (mins)	3.8
Alert Accuracy (%)	91.5
Intervention Success Rate (%)	88.2
False Positive Rate (%)	6.3
False Negative Rate (%)	4.5

 

The evaluation measures of the system reaction as indicated in Table 3 reveal the extent to which the real time media tracking system can identify and manage the crisis in a timely manner and in the most efficient way possible. The Average Response Time of 3.8 minutes is evidence that the system is capable of sending messages and acting fast, which is relevant as it will minimize the consequences of a disaster. A score of 91.5% in the Alert Accuracy indicates that the model is useful in identifying real problems and not producing too many false positives. Fig. 5 depicts the stacked metrics visualization of response, accuracy, success and error rates.

 Figure 5

Figure 5 Stacked Metrics Visualization: Response, Accuracy, Success, and Error Rates

 

The Intervention Success of 88.2 percent indicates that the system is capable of reacting appropriately to ensure that the right things are done in situations. False Positive rate of 6.3% indicates that although the system focuses so well in detecting the crisis, there is still a slight probability that non-crisis may be classified as crisis.

 

5. CONCLUSION

In this work, we developed a real-time media tracking system of identifying crisis and planning on how to manage it using machine learning approaches. The general idea was to have an automatic system that would be able to monitor a great multitude of various kinds of media, crunch a lot of unstructured data, and send out timely alerts and plans as to how to deal with new emergencies. The system could analyze data on the news websites, social media websites, and web forums by integrating natural language processing (NLP) and machine learning models and identify potential issues before they occur. The system classified events through support vector machine (SVM) and deep learning into cults depending on the variation in mood, new topics, and finding outliers. The accuracy rate of the system was very high (92%), and the level of precision and memory was high, which indicated that the system could make the distinction between disasters and other scenarios. The forecasts of the model triggered the reaction generation process that enabled one to move fast by undertaking activities such as calling the police or using established crisis management procedures. But, it had certain issues, in particular, when addressing massive volume and complexity of unstructured media data. To have a working system that was fast and accurate, it required new ways of preparing the data, extracting features, and making a real time classification. Despite these issues, the system was very beneficial in terms of dealing with crisis situations; the system was able to locate problems and respond to them at a quicker rate than the old fashioned method of human approach.

 

CONFLICT OF INTERESTS

None. 

 

ACKNOWLEDGMENTS

None.

 

REFERENCES

Ahmad, W., Rasool, A., Javed, A. R., Baker, T., and Jalil, Z. (2022). Cyber Security in IoT-Based Cloud Computing: A Comprehensive Survey. Electronics, 11(1), 16. https://doi.org/10.3390/electronics11010016

Albarakt, R., Selim, G., and Iaaly, A. (2021). Reshaping Riyadh Alsolh Square: Mapping the Narratives of Protesting Crowds in Beirut. Urban and Regional Planning, 6(4), 126–133. https://doi.org/10.11648/j.urp.20210604.13

Bajao, N. A., and Sarucam, J.-A. (2023). Threats Detection in the Internet of Things Using Convolutional Neural Networks, Long Short-Term Memory, and Gated Recurrent Units. Mesopotamian Journal of Cybersecurity, 2023, 22–29. https://doi.org/10.58496/MJCS/2023/005

Bhowmick, D., Winter, S., Stevenson, M., and Vortisch, P. (2020). The Impact of Urban Road Network Morphology on Pedestrian Wayfinding Behaviour. Journal of Spatial Information Science, 21, 601. https://doi.org/10.5311/JOSIS.2020.21.601

Czum, J. M. (2020). Dive into Deep Learning. Journal of the American College of Radiology, 17(5), 637–638. https://doi.org/10.1016/j.jacr.2020.02.005

Dong, E., Du, H., and Gardner, L. (2020). An Interactive Web-Based Dashboard to Track COVID-19 in Real Time. The Lancet Infectious Diseases, 20(5), 533–534. https://doi.org/10.1016/S1473-3099(20)30120-1

Jarillo Silva, A., Gómez Pérez, V. A., and Domínguez Ramírez, O. A. (2024). Study of the Length of Time Window in Emotion Recognition Based on EEG Signals. Revista Mexicana de Ingeniería Biomédica, 45, 31–42. https://doi.org/10.17488/RMIB.45.1.3

Jian, S. W., Kao, C. T., Chang, Y. C., Chen, P. F., and Liu, D. P. (2021). Risk Assessment for COVID-19 Pandemic in Taiwan. International Journal of Infectious Diseases, 104, 746–751. https://doi.org/10.1016/j.ijid.2021.01.042

Kutsarova, V., and Matskin, M. (2021). Combining Mobile Crowdsensing and Wearable Devices for Managing Alarming Situations. In Proceedings of the IEEE 45th Annual Computers, Software, and Applications Conference (COMPSAC) ( 538–543). IEEE. https://doi.org/10.1109/COMPSAC51774.2021.00080

Mak, H. W. L., and Koh, K. (2021). Building a Healthy Urban Environment in East Asia (Report No. 1). Joint Laboratory on Future Cities (JLFC), The University of Hong Kong.

Malkani, D., Malkani, M., Singh, N., and Madan, E. (2023). Best Practices for the Design of COVID-19 Dashboards. Perspectives in Health Information Management, 20, 1b.

Mijwil, M., Filali, Y., Aljanabi, M., Bounabi, M., and Al-Shahwani, H. (2023). The Purpose of Cybersecurity Governance in the Digital Transformation of Public Services and Protecting the Digital Environment. Mesopotamian Journal of Cybersecurity, 2023, 1–6. https://doi.org/10.58496/MJCS/2023/001

Tsai, C. H., Chen, P. C., Liu, D. S., Kuo, Y. Y., Hsieh, T. T., Chiang, D. L., Lai, F., and Wu, C. T. (2022). Panic Attack Prediction Using Wearable Devices and Machine Learning: Development and Cohort Study. JMIR Medical Informatics, 10, e33063. https://doi.org/10.2196/33063

Zade, N. J., Lanke, N. P., Madan, B. S., Ghutke, P., and Khobragade, P. (2024). Neural Architecture Search: Automating the Design of Convolutional Models for Scalability. Panamerican Mathematical Journal, 34(4), 178–193. https://doi.org/10.52783/pmj.v34.i4.1877

Zhang, X., Sun, Y., Li, Q., Li, X., and Shi, X. (2023). Crowd Density Estimation and Mapping Method Based on Surveillance Video and GIS. ISPRS International Journal of Geo-Information, 12, 56. https://doi.org/10.3788/IRLA20230398