BRAC University Institutional Repository
Preserving Knowledge, Advancing Research, Sharing Scholarship
A digital platform for collecting, preserving, and sharing BRAC University’s scholarly, academic, and institutional outputs.
Contribute to the Repository
Faculty members and students are invited to submit their research publications, theses, dissertations, and scholarly works to increase visibility, access, and long-term preservation.
Recent Submissions
Feature-based mobile phone rating using sentiment analysis and machine learning approaches
(Institute of Electrical and Electronics Engineers Inc., 2019-05-01) Kafi, Abdullahil; Ashikul Alam, M. Shaikh; Bin Hossain, Sayeed; Awal, Siam Bin; Arif, Hossain
This paper proposes a model of sentiment analysis of various features of different companies' mobile phones and their overall rating. Before buying a phone, customers usually look for reviews to decide which phone to buy. The model proposed in this paper provides an optimal solution for the customer for making this decision more efficiently. In this model, each feature of a mobile phone is rated based on public opinion and an overall rating for each phone is provided. Amazon is one of the largest Internet retailers, which makes way for most public reviews on their products. These reviews are collected as a form of an open source platform and used as the dataset in this model. The gathered data is preprocessed and then separated into two different sets - Training Set and Testing Set which are used to train and test the supervised machine learning algorithms for classification. 15 most common features of the mobile phones based on public reviews are selected from the training data set and used as the feature set in this model. Different algorithms which include Naïve Bayes, Support Vector Machine, Logistic Regression, and Stochastic Gradient Descent algorithms are used in this model and the comparison of their performance is shown. This model provides a rating of each feature and an average rating of the mobile phone based on sentiment polarity. Thus, this research work can assist potential customers to choose the best product based on the opinion of the other users.
CSTRL: Context-driven sequential transfer learning for abstractive radiology report summarization
(Association for Computational Linguistics (ACL), 2025-01-01) Naznin, Fahmida Sultana; Faruq, Adnan Ibney; Tazwar, Mostafa Rifat; Jobayer, Md; Shawon, Md Mehedi Hasan; Hasan, Md Rakibul
A radiology report comprises several sections, including the Findings and Impression of the diagnosis. Automatically generating the Impression from the Findings is crucial for reducing radiologists' workload and improving diagnostic accuracy. Pretrained models that excel in common abstractive summarization problems encounter challenges when applied to specialized medical domains, largely due to the complex terminology and the necessity for accurate clinical context. Such tasks in medical domains demand extracting core information, avoiding context shifts, and maintaining proper flow. Misuse of medical terms can lead to drastic clinical errors. To address these issues, we introduce a sequential transfer learning that ensures key content extraction and coherent summarization. Sequential transfer learning often faces challenges like initial parameter decay and knowledge loss, which we resolve with the Fisher matrix regularization. Using MIMIC-CXR and Open-I datasets, our model, CSTRL-Context-driven Sequential TRansfer Learning-achieved state-of-the-art performance: 56.2% improvement in BLEU-1, 40.5% in BLEU-2, 84.3% in BLEU-3, 28.9% in ROUGE-1, 41.0% in ROUGE-2 and 26.5% in ROGUE-3 score over benchmark methods. We further analyze factual consistency scores while preserving the medical context. Our code is publicly available at https://github.com/fahmidahossain/Report_Summarization. © 2025 Association for Computational Linguistics.
Inceptive transformers: Enhancing contextual representations through multi-scale feature learning across domains and languages
(Association for Computational Linguistics (ACL), 2025-01-01) Shahriar, Asif; Shahriyar, Rifat; Saifur Rahman M.
Encoder transformer models compress information from all tokens in a sequence into a single [CLS] token to represent global context. This approach risks diluting fine-grained or hierarchical features, leading to information loss in downstream tasks where local patterns are important. To remedy this, we propose a lightweight architectural enhancement: an inception-style 1-D convolution module that sits on top of the transformer layer and augments token representations with multi-scale local features. This enriched feature space is then processed by a self-attention layer that dynamically weights tokens based on their task relevance. Experiments on five diverse tasks show that our framework consistently improves general-purpose, domain-specific, and multilingual models, outperforming baselines by 1% to 14% while maintaining efficiency. Ablation studies show that multi-scale convolution performs better than any single kernel and that the self-attention layer is critical for performance. © 2025 Association for Computational Linguistics.
Heart failure risk prediction and medicine recommendation using exploratory data analysis
(Institute of Electrical and Electronics Engineers Inc., 2019-05-01) Habib, Sumaya; Moin, Maisha Binte; Aziz, Sujana; Banik K.; Arif, Hossain; Department of Computer Science and Engineering
With the ever increasing population of the world, diseases and their possibilities are also increasing at an alarming rate. As time passes by, diagnosing diseases and providing appropriate treatment at the right time has become quite a challenge. Heart diseases have been a major cause of death worldwide. Therefore, this research has been focused on finding an efficient way to predict the chances of a heart failure and accordingly, recommend appropriate medicines to aid cardiologists in quicker decision making. The research includes finding the correlations or associations between the various medical profiles of the patients by utilizing the standard techniques of exploratory analysis and hence using the attributes suitably to predict the chances of a heart failure, as well as the medicine recommendations. A comparative study has also been included which shows the various attained accuracy rates of different machine learning algorithms including - Logistic Regression, Naïve Bayes, Decision Tree, Linear Support Vector Classifier, Random Forest, and Gradient Boosting Classifier. The Apache Spark framework has been used in order to make the system capable of handling big data.
An IoT based smart energy meter for the idle use of electricity: An industry perspective
(Institute of Electrical and Electronics Engineers Inc., 2019-05-01) Muhammad Masum A.K.; Manzur S.; Amzad Chy M.K.; Islam Khan S.; Alam, Md. Golam Rabiul; Taslim Reza S.M.; Department of Computer Science and Engineering
In the perspective of the Bangladesh garment industry, a vast amount of power is wastage due to forgetting in turning off the machines at break time or other reason. Also, the existing prepaid meter doesn't include any system to understand the wastage of the power and control the machines remotely. This research demonstrates a smart energy meter to understand the idle use of the loads and control the loads remotely. This efficient and cost-effective system offers an accurate estimation of the consumed power and cost, reducing the labor cost. A microcontroller is involved in calculating the voltage, current, energy and cost. After deploying in the garment industry, the first few daysthe systemmeasures the consumed amount of power by different loads and store it. From these values, an average value is estimated as a threshold value for that load. In case of exceeding the threshold value of this load, an auto-notification is sent to the authority via a GSM module. Being notified about the excessive usage via that load, the authority confirms about the machine is in working state or idle state through announcing in a microphone. According to the response, the authority takes the decision and sent a specific command through SMS to the system. The GSM module converts the SMS into a signal that is supplied to the microcontroller to turn off the load. In consequence of this feature, the system will assist the authority of the garment industry to be economical in power consumption, and the amount of power wastage will be reduced drastically.