YSYuliant Sibaroni

Yuliant Sibaroni

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Yuliant Sibaroni has a total of 70 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • A Comparative Study on Handling Imbalanced Data in Indonesian Hate Speech Detection Using FastText and BiLSTM

    by Akmal Muhamad Faza, Yuliant Sibaroni, Sri Suryani Prasetiyowati

    Online hate speech has become a serious threat to social harmony in Indonesia, with cases increasing significantly in recent years. This study develops and evaluates a system for detecting Indonesian hate speech using a Bidirectional Long Short-Term Memory (BiLSTM) deep learning model, complemented by FastText word embeddings. To address the common issue of data imbalance in hate speech datasets, this study implements and compares three oversampling techniques: Random Oversampler, Synthetic Minority Oversampling Technique (SMOTE), and Adaptive Synthetic Sampling (ADASYN). The research utilizes the Indonesian Hate Speech Superset, a dataset comprising 14,306 comments. The model's performance is evaluated using Stratified K-fold Cross-Validation, with metrics including Accuracy, Precision, Recall, and F1-score. Results, visualized using a Confusion Matrix to demonstrate that applying oversampling techniques enhances model performance, particularly by improving the Recall and F1-score metrics. These findings contribute to the development of hate speech classification systems that are fairer, more adaptive, and better suited to the unique characteristics of the Indonesian social media landscape.


  • untuk, dalam Training on The Use of AI to Increase Teacher Competency in Preparing The Learning Process

    by Diyas Puspandari, Sri Prasetiyowati, Yuliant Sibaroni

    Teachers, as educational personnel, are expected to enhance their competence in mastering information technology, which can later be utilized in classroom learning activities. However, in reality, teachers' knowledge and skills in mastering technology to support teaching and learning activities are still limited. Based on a survey conducted, teachers at SD Indriyasana, Baleendah, are eager to gain knowledge and skills in using open-source applications, such as Chat GPT, Gemini AI, and MajickPen AI, to support their tasks, both in teaching and administrative duties. Therefore, training and mentoring are needed in learning the use of these open-source applications to support teachers' activities. The aim of this training is to introduce and train teachers at SD Indriyasana, Baleendah, in using open-source applications based on artificial intelligence (AI), such as Chat GPT, Gemini AI, and MajickPen AI, to support teaching and administrative tasks. The method used involves training, direct practice by teachers, guided by speakers and instructors using AI-based open-source applications, namely ChatGPT, Gemini AI, and MajickPen. The implementation method includes the following steps: 1) user needs analysis, 2) literature study, 3) training module development, 4) module testing, 5) training implementation, 6) evaluation. The training results indicate that teachers acquired knowledge and practical experience in line with their needs, enabling them to use AI-based applications (Chat GPT, Gemini AI, and MajickPen AI) to support learning and administrative activities. It can be said that the training went well, as shown by the questionnaire results: 1) alignment of the training program with objectives (94%), 2) alignment of the program with partners' needs (71%), 3) adequacy of program implementation time (65%), 4) team’s ability to execute the program (88%), and 5) program sustainability (77%). This training activity is beneficial for both teachers and students because it can make learning more interesting and enjoyable for students. By utilizing AI-based applications, teachers can generate ideas for learning activities, create quizzes, or provide relevant and up-to-date teaching materials, thereby facilitating class preparation and easing teachers' administrative tasks.


  • Open Access

    Multilabel Hate Speech Classification in Indonesian Political Discourse on X using Combined Deep Learning Models with Considering Sentence Length

    by Revelin Angger Saputra, Yuliant Sibaroni

    Hate speech, as public expression of hatred or offensive discourse targeting race, religion, gender, or sexual orientation, is widespread on social media. This study assesses BERT-based models for multi-label hate speech detection, emphasizing how text length impacts model performance. Models tested include BERT, BERT-CNN, BERT-LSTM, BERT-BiLSTM, and BERT with two LSTM layers. Overall, BERT-BiLSTM achieved the highest (82.00%) and best performance on longer texts (83.20% ) with high and , highlighting its ability to capture nuanced context. BERT-CNN excelled in shorter texts, achieving the highest (79.80%) and an of 79.10%, indicating its effectiveness in extracting features in brief content. BERT-LSTM showed balanced and across text lengths, while BERT-BiLSTM, although high in r, had slightly lower on short texts due to its reliance on broader context. These results highlight the importance of model selection based on text characteristics: BERT-BiLSTM is ideal for nuanced analysis in longer texts, while BERT-CNN better captures key features in shorter content.


  • Open Access

    Implementation of BiLSTM and IndoBERT for Sentiment Analysis of TikTok Reviews

    by Azziz Fachry Al Farizi, Yuliant Sibaroni

    The significant increase in users on TikTok has led to a notable rise in the number of reviews in the form of opinions given to the application. The large number of opinions can be analyzed to identify the prevailing sentiment among the community towards the application. The sentiment analysis method employing machine learning is particularly well-suited to this problem due to its practicality and efficiency. The objective of this research is to develop a model that can be utilized as a sentiment analysis tool with a high degree of accuracy. In this research, the BiLSTM algorithm, combined with IndoBERT, a pre-trained model, is employed. The BiLSTM can comprehend the interrelationships between words within a sentence in a bidirectional manner. IndoBERT is pertinent to this research as it is a model that has been fine-tuned using Indonesian language datasets from various sources on the Internet. To support this research, a scenario was created by considering various aspects when adding methods as an optimization scheme until the optimal model was identified. The outcomes of experimentation demonstrate that sentiment analysis using the BiLSTM+IndoBERT method achieved the highest accuracy, reaching 81% in the classification report test and an average accuracy of 92.03% in cross-validation testing with a total of 10 folds.


  • Open Access

    Effectiveness of Bi-GRU and FastText in Sentiment Analysis of Shopee App Reviews

    by Rayhan Fadhil Rahmanda, Yuliant Sibaroni, Sri Suryani Prasetiyowati

    E-commerce is proof of evolution in the economic field due to its flexibility to shop for various necessities of life anytime and anywhere. Shopee is one of the e-commerce platforms in demand by people from varied circles in Indonesia. Multiple reviews are shed publicly by Shopee users on the Google Play Store regarding shopping experiences, which can be positive or negative. This condition affects the decision of other users to shop at Shopee, thus impacting the increase or decrease in profits from Shopee itself. Therefore, user sentiment analysis is needed as a form of effort to maintain user trust in Shopee. This research aims to build a system to classify the sentiment of Shopee application users through reviews in the Google Play Store by utilizing the Bidirectional Gated Recurrent Unit (Bi-GRU) deep learning model. The dataset contains 9,716 reviews, including 3,937 positive and 5,779 negative sentiments. Several test scenarios were conducted to achieve the highest peak of performance, utilizing TF-IDF feature extraction, FastText feature expansion, and optimization using the Cuckoo Search Algorithm. Additionally, SMOTE resampling was utilized to correct the dataset’s uneven distribution. The combined test scenarios mentioned significantly improved the accuracy by 1.03% and F1-Score by 1.04% from the baseline, with the highest accuracy reaching 90.48% and the highest F1-Score of 90.16%.


  • Open Access

    WORD EMBEDDING OPTIMIZATION IN SENTIMENT ANALYSIS OF REVIEWS ON MYTELKOMSEL APP USING LONG SHORT-TERM MEMORY AND SYNTHETIC MINORITY OVER-SAMPLING TECHNIQUE

    by Muhammad Raffif Haziq, Yuliant Sibaroni, Sri Suryani Prasetyowati

    Telkomsel is one of the internet service provider companies that has a mobile-based application called MyTelkomsel which functions to facilitate users in conducting online services independently. Users of the application certainly have their own responses about the application, so that users can provide responses to the application. Therefore, sentiment analysis can be one of the solutions to find out public sentiment towards the application. In this research, the author builds a system for sentiment analysis using word embedding Word2vec, GloVe, FastText to get word representation in vector form with classification using Long Short-Term Memory (LSTM) combined with Synthetic Minority Over-sampling Technique (SMOTE) which can handle data imbalance. The data used comes from user reviews of the MyTelkomsel application found on the Google Play Store. This study compares the performance of several word embedding in LSTM and LSTM-SMOTE classifiers. The results showed the results show that the performance of three-word embedding on the LSTM model is superior compared to the LSTM-SMOTE model. Overall, it was found that the combination of FastText and LSTM gave the best performance compared to the other five combinations with an accuracy value of 89.11%.


  • Open Access

    HATE SPEECH DETECTION USING GLOVE WORD EMBEDDING AND GATED RECURRENT UNIT

    by Aulia Riefqi Ardana, Yuliant Sibaroni

    Social media has become a tool that makes it easier for people to exchange information. The freedom to share information has opened the door for increased incidents of hate speech on social media. Hate speech detection is an interesting topic because with the increasing use of social media, hate speech can quickly spread and trigger significant negative impacts, discrimination, and social conflict. This research aims to see the effect of GRU method, GloVe word embedding and word modifier algorithm in detecting hate speech. GRU and GloVe are used in this research for the hate speech detection system, where deep learning with a Gated Recurrent Unit (GRU) and Word Embedding with the Global Vector model (GloVe) converts words in text into numerical vectors that represent the meaning and context of the words. GRU is chosen due to its ability to capture long-term dependencies in textual data with higher computational efficiency compared to Long Short-Term Memory (LSTM). Gated Recurrent Unit (GRU) model processes the sequence of words to understand the sentence structure. GRU model processes the sequence of words to understand the sentence structure. The evaluation results for the classification of hate speech using GRU and GloVe are 90.7% accuracy and 91% F1 score. With the combination of informal word modifier algorithms there is an increase with a value of 92.8% F1 and 92.4% accuracy. in conclusion, the use of informal word modifier algorithms can increase the evaluation value in detecting hate speech.

  • Open Access

    Sentiment Analysis For The 2024 Presidential Election (Pilpres) Using BERT CNN

    by Daffa Fadhilah Putra, Yuliant Sibaroni

    The 2024 presidential election in Indonesia has generated tremendous enthusiasm on social media, particularly on the X platform. This research aims to analyze public sentiment regarding the 2024 presidential election by utilizing BERT and CNN methods. Sentiment analysis in the digital era is key to understanding the diverse social perspectives within society. The use of BERT, which has proven effective in understanding natural language context, and CNN, initially used for image analysis, will help in understanding public sentiment on X leading up to the 2024 presidential election. The research results show that the BERT model provides the best performance with an average accuracy of 90.02%, while CNN achieved 88.19%. The sentiment-based predictions using BERT for the three presidential candidates indicate that Prabowo Subianto is predicted to receive the highest support at 43.82%, followed by Ganjar Pranowo with 33.83%, and Anies Baswedan with 22.35%. A comparison of the prediction results with the actual election results shows that Prabowo Subianto was predicted to receive 43.82% of the vote, while the actual election results reached 58.58%, a difference of 14.76%. Ganjar Pranowo was predicted to receive 33.83% of the vote, while the actual results were 16.47%, with a difference of 17.36%. Anies Baswedan was predicted to receive 22.35% of the vote, with the actual result being 24.95%, a difference of 2.60%. This study indicates that the BERT model is effective in providing an accurate depiction of the 2024 Indonesian presidential election results.


  • Open Access

    Multi-Aspect Sentiment Analysis Using Elman Recurrent Neural Network (ERNN) Method for TripAdvisor App User Reviews

    by Fahrul Raykhan Ridho, Yuliant Sibaroni, Dyas Puspandari

    TripAdvisor is the world's largest travel platform that assists 463 million travelers each month in making their trips the best they can be. Users of TripAdvisor can provide reviews, comments, and ratings of travel destinations. However, reviews on TripAdvisor are considered insufficient in helping prospective travelers understand the strengths and weaknesses of a hotel. Therefore, a multiaspect sentiment analysis of TripAdvisor reviews on hotels was conducted to identify commonly discussed rating aspects among visitors and to determine specific evaluations. In this study, the Elman Recurrent Neural Network (ERNN) method was employed to build a classification system for multiaspect sentiment analysis of user reviews on the TripAdvisor application. The aspects examined in this research include Service, Cleanliness, Location, Value, Rooms, and Overall Experience, aiming to provide insights into the hotels under consideration. The results indicate that the ERNN method can deliver superior outcomes in multiaspect sentiment analysis of TripAdvisor hotel reviews. The ERNN model's performance in multiaspect sentiment analysis shows optimal accuracies: 81.35% for Service, 98.71% for Cleanliness, 74.87% for Location, 93.84% for Value and 71.52% for Rooms. These findings can assist travelers in better understanding the strengths and weaknesses of accommodations.

  • Open Access

    Multi-aspect Sentiment Analysis of Shopee Application Reviews using RNN Method and Query Expansion Ranking

    by Ariqoh Novitasari, Yuliant Sibaroni, Diyas Puspandari

    Online shopping using e-commerce is a common activity society does in this digital era. Shopee is one of the well-known e-commerce in Indonesia. There are a lot of e-commerce platforms that can easily be accessed through mobile applications like Google Play Store. Users are allowed to review and rate the application they have downloaded. The reviews from the users become an opportunity for e-commerce companies to advance their performances and services. To enhance the understandability of user reviews, a system that can efficiently analyze the sentiment is needed. This study aims to design and establish a system that can perform sentiment analysis on the selected aspects. Sentiment classification is implemented by using the Recurrent Neural Network (RNN) algorithm and Query Expansion Ranking feature selection to classify Shopee application reviews into two classes, which are positive and negative. Feature selection is used to reduce less useful features so that the classification model conducts the classification process optimally and more efficiently. In conclusion, the evaluation results based on an 80:20 data split ratio indicate that the RNN achieves the highest accuracy of 95% in the delivery cost aspect, 93% in the delivery speed aspect, and 86% in the application access aspect.