Word Embedding Fasttext Techniques

Oct 13, 2025 by recoveryshake.com 148 views

Introduction

The rapid development of communication and information technology has led to the widespread use of social media platforms, including X (Twitter). While social media offers numerous benefits, such as real-time information dissemination and discussion forums, it also poses potential problems, including online harassment. This issue is particularly concerning in Indonesia, where online harassment can compromise user safety and comfort. Therefore, identifying and classifying online harassment tweets on social media X is crucial to protect users and prevent the spread of violent and harassing content.

Background

Online harassment is a pervasive issue on social media platforms, including X. It can take various forms, such as hate speech, cyberbullying, and harassment. The consequences of online harassment can be severe, including emotional distress, anxiety, and even physical harm. In Indonesia, online harassment is a significant concern, particularly among young people who are active on social media. The need to address online harassment has become increasingly pressing, and the development of effective classification systems is essential to mitigate this issue.

Methodology

This study employed the Gated Recurrent Unit (GRU) algorithm and Word Embedding Fasttext techniques to classify online harassment tweets on social media X. The data used consisted of 7,006 tweets collected using a Tweet Harvest Tool in the X application. The data was then processed and analyzed using the GRU algorithm to identify and classify content that could potentially contain harassment.

The Gated Recurrent Unit (GRU) Algorithm

The GRU algorithm is a type of Recurrent Neural Network (RNN) that is designed to handle sequential data, such as text. It is particularly effective in modeling long-term dependencies in data and has been widely used in natural language processing tasks, including text classification. In this study, the GRU algorithm was used to classify online harassment tweets on social media X.

Word Embedding Fasttext Techniques

Word Embedding Fasttext techniques were used to improve the performance of the GRU algorithm. This technique allows a better understanding of the context of words in sentences, thereby increasing classification accuracy. By understanding the context of the word, the algorithm can be more appropriate in determining whether a tweet is included in the category of harassment or not.

Evaluation Results

The evaluation results using a confusion matrix showed that the accuracy rate of the model used reached 91%. This indicates that the algorithm applied is quite effective in recognizing and classifying tweets that contain harassment. This achievement is a step forward in an effort to tackle online harassment, which is an important issue in digital society today.

Conclusion

The application of the GRU algorithm and Word Embedding Fasttext techniques in classifying online harassment tweets on social media X has shown promising results. The accuracy rate of 91% indicates that the algorithm is effective in recognizing and classifying tweets that contain harassment. This achievement is a significant step forward in addressing online harassment, which is a pressing issue in digital society today. The use of this method not only provides a positive contribution in creating a safer social media environment but also offers a foundation for the development of more sophisticated systems in detecting and handling online harassment on other platforms.

Future Directions

This study provides a foundation for the development of more sophisticated systems in detecting and handling online harassment on other platforms. Future research can build upon this study by exploring the use of other machine learning algorithms and techniques, such as deep learning and natural language processing. Additionally, the development of more effective classification systems can be achieved by incorporating additional features, such as user behavior and sentiment analysis.

Recommendations

Based on the findings of this study, the following recommendations are made:

Development of more sophisticated classification systems: The development of more sophisticated classification systems can be achieved by incorporating additional features, such as user behavior and sentiment analysis.
Use of other machine learning algorithms and techniques: Future research can explore the use of other machine learning algorithms and techniques, such as deep learning and natural language processing.
Incorporation of additional features: The incorporation of additional features, such as user behavior and sentiment analysis, can improve the accuracy of classification systems.
Development of more effective detection and handling systems: The development of more effective detection and handling systems can be achieved by incorporating additional features and using more sophisticated machine learning algorithms and techniques.

Limitations

This study has several limitations, including:

Limited dataset: The dataset used in this study consisted of 7,006 tweets, which may not be representative of the entire population of online harassment tweets on social media X.
Limited features: The features used in this study were limited to text-based features, which may not capture the complexity of online harassment.
Limited evaluation metrics: The evaluation metrics used in this study were limited to accuracy, which may not capture the full range of performance metrics.

Future Research Directions

Q: What is online harassment, and why is it a significant issue on social media?

A: Online harassment refers to any form of hostile or threatening behavior that occurs online, including hate speech, cyberbullying, and harassment. It is a significant issue on social media because it can compromise user safety and comfort, and can have severe consequences, including emotional distress, anxiety, and even physical harm.

Q: What is the Gated Recurrent Unit (GRU) algorithm, and how is it used in this study?

A: The GRU algorithm is a type of Recurrent Neural Network (RNN) that is designed to handle sequential data, such as text. It is particularly effective in modeling long-term dependencies in data and has been widely used in natural language processing tasks, including text classification. In this study, the GRU algorithm is used to classify online harassment tweets on social media X.

Q: What is Word Embedding Fasttext, and how does it improve the performance of the GRU algorithm?

A: Word Embedding Fasttext is a technique that allows a better understanding of the context of words in sentences, thereby increasing classification accuracy. By understanding the context of the word, the algorithm can be more appropriate in determining whether a tweet is included in the category of harassment or not. This technique improves the performance of the GRU algorithm by providing a more nuanced understanding of the language used in the tweets.

Q: What are the evaluation results of this study, and what do they indicate?

A: The evaluation results using a confusion matrix showed that the accuracy rate of the model used reached 91%. This indicates that the algorithm applied is quite effective in recognizing and classifying tweets that contain harassment. This achievement is a step forward in an effort to tackle online harassment, which is an important issue in digital society today.

Q: What are the limitations of this study, and how can they be addressed in future research?

A: This study has several limitations, including a limited dataset, limited features, and limited evaluation metrics. Future research can address these limitations by using larger and more diverse datasets, incorporating additional features, and using more comprehensive evaluation metrics.

Q: What are the implications of this study for the development of more sophisticated systems in detecting and handling online harassment?

A: This study provides a foundation for the development of more sophisticated systems in detecting and handling online harassment on other platforms. Future research can build upon this study by exploring the use of other machine learning algorithms and techniques, such as deep learning and natural language processing. Additionally, the development of more effective classification systems can be achieved by incorporating additional features, such as user behavior and sentiment analysis.

Q: What are the recommendations for future research and development in this area?

A: Based on the findings of this study, the following recommendations are made:

Development of more sophisticated classification systems: The development of more sophisticated classification systems can be achieved by incorporating additional features, such as user behavior and sentiment analysis.
Use of other machine learning algorithms and techniques: Future research can explore the use of other machine learning algorithms and techniques, such as deep learning and natural language processing.
Incorporation of additional features: The incorporation of additional features, such as user behavior and sentiment analysis, can improve the accuracy of classification systems.
Development of more effective detection and handling systems: The development of more effective detection and handling systems can be achieved by incorporating additional features and using more sophisticated machine learning algorithms and techniques.

Q: What are the potential applications of this research in real-world settings?

A: The potential applications of this research in real-world settings include:

Development of more effective online harassment detection systems: The development of more effective online harassment detection systems can be achieved by incorporating additional features and using more sophisticated machine learning algorithms and techniques.
Improvement of user safety and comfort on social media: The improvement of user safety and comfort on social media can be achieved by developing more effective classification systems and detection systems.
Development of more effective online harassment handling systems: The development of more effective online harassment handling systems can be achieved by incorporating additional features and using more sophisticated machine learning algorithms and techniques.