Md. Saiful Islam

Md Kamrul Hasan, Md Saiful Islam, Sangwu Lee et al.

Pre-trained large language models have recently achieved ground-breaking performance in a wide variety of language understanding tasks. However, the same model can not be applied to multimodal behavior understanding tasks (e.g., video sentiment/humor detection) unless non-verbal features (e.g., acoustic and visual) can be integrated with language. Jointly modeling multiple modalities significantly increases the model complexity, and makes the training process data-hungry. While an enormous amount of text data is available via the web, collecting large-scale multimodal behavioral video datasets is extremely expensive, both in terms of time and money. In this paper, we investigate whether large language models alone can successfully incorporate non-verbal information when they are presented in textual form. We present a way to convert the acoustic and visual information into corresponding textual descriptions and concatenate them with the spoken text. We feed this augmented input to a pre-trained BERT model and fine-tune it on three downstream multimodal tasks: sentiment, humor, and sarcasm detection. Our approach, TextMI, significantly reduces model complexity, adds interpretability to the model's decision, and can be applied for a diverse set of tasks while achieving superior (multimodal sarcasm detection) or near SOTA (multimodal sentiment analysis and multimodal humor detection) performance. We propose TextMI as a general, competitive baseline for multimodal behavioral analysis tasks, particularly in a low-resource setting.

Abhik Bhattacharjee, Tahmid Hasan, Wasi Uddin Ahmad et al.

In this work, we introduce BanglaBERT, a BERT-based Natural Language Understanding (NLU) model pretrained in Bangla, a widely spoken yet low-resource language in the NLP literature. To pretrain BanglaBERT, we collect 27.5 GB of Bangla pretraining data (dubbed `Bangla2B+') by crawling 110 popular Bangla sites. We introduce two downstream task datasets on natural language inference and question answering and benchmark on four diverse NLU tasks covering text classification, sequence labeling, and span prediction. In the process, we bring them under the first-ever Bangla Language Understanding Benchmark (BLUB). BanglaBERT achieves state-of-the-art results outperforming multilingual and monolingual models. We are making the models, datasets, and a leaderboard publicly available at https://github.com/csebuetnlp/banglabert to advance Bangla NLP.

Khondoker Ittehadul Islam, Md. Saiful Islam, Md Ruhul Amin

Sentiment analysis (SA) in Bengali is challenging due to this Indo-Aryan language's highly inflected properties with more than 160 different inflected forms for verbs and 36 different forms for noun and 24 different forms for pronouns. The lack of standard labeled datasets in the Bengali domain makes the task of SA even harder. In this paper, we present manually tagged 2-class and 3-class SA datasets in Bengali. We also demonstrate that the multi-lingual BERT model with relevant extensions can be trained via the approach of transfer learning over those novel datasets to improve the state-of-the-art performance in sentiment classification tasks. This deep learning model achieves an accuracy of 71\% for 2-class sentiment classification compared to the current state-of-the-art accuracy of 68\%. We also present the very first Bengali SA classifier for the 3-class manually tagged dataset, and our proposed model achieves an accuracy of 60\%. We further use this model to analyze the sentiment of public comments in the online daily newspaper. Our analysis shows that people post negative comments for political or sports news more often, while the religious article comments represent positive sentiment. The dataset and code is publicly available at https://github.com/KhondokerIslam/Bengali\_Sentiment.

Raihan Kabir, Naznin Haque, Md Saiful Islam et al.

Visual question answering (VQA) refers to the problem where, given an image and a natural language question about the image, a correct natural language answer has to be generated. A VQA model has to demonstrate both the visual understanding of the image and the semantic understanding of the question, demonstrating reasoning capability. Since the inception of this field, a plethora of VQA datasets and models have been published. In this article, we meticulously analyze the current state of VQA datasets and models, while cleanly dividing them into distinct categories and then summarizing the methodologies and characteristics of each category. We divide VQA datasets into four categories: (1) available datasets that contain a rich collection of authentic images, (2) synthetic datasets that contain only synthetic images produced through artificial means, (3) diagnostic datasets that are specially designed to test model performance in a particular area, e.g., understanding the scene text, and (4) KB (Knowledge-Based) datasets that are designed to measure a model's ability to utilize outside knowledge. Concurrently, we explore six main paradigms of VQA models: fusion, where we discuss different methods of fusing information between visual and textual modalities; attention, the technique of using information from one modality to filter information from another; external knowledge base, where we discuss different models utilizing outside information; composition or reasoning, where we analyze techniques to answer advanced questions that require complex reasoning steps; explanation, which is the process of generating visual and textual descriptions to verify sound reasoning; and graph models, which encode and manipulate relationships through nodes in a graph. We also discuss some miscellaneous topics, such as scene text understanding, counting, and bias reduction.

Aisha Khatun, Anisur Rahman, Md Saiful Islam et al.

Authorship Attribution is the task of creating an appropriate characterization of text that captures the authors' writing style to identify the original author of a given piece of text. With increased anonymity on the internet, this task has become increasingly crucial in various security and plagiarism detection fields. Despite significant advancements in other languages such as English, Spanish, and Chinese, Bangla lacks comprehensive research in this field due to its complex linguistic feature and sentence structure. Moreover, existing systems are not scalable when the number of author increases, and the performance drops for small number of samples per author. In this paper, we propose the use of Average-Stochastic Gradient Descent Weight-Dropped Long Short-Term Memory (AWD-LSTM) architecture and an effective transfer learning approach that addresses the problem of complex linguistic features extraction and scalability for authorship attribution in Bangla Literature (AABL). We analyze the effect of different tokenization, such as word, sub-word, and character level tokenization, and demonstrate the effectiveness of these tokenizations in the proposed model. Moreover, we introduce the publicly available Bangla Authorship Attribution Dataset of 16 authors (BAAD16) containing 17,966 sample texts and 13.4+ million words to solve the standard dataset scarcity problem and release six variations of pre-trained language models for use in any Bangla NLP downstream task. For evaluation, we used our developed BAAD16 dataset as well as other publicly available datasets. Empirically, our proposed model outperformed state-of-the-art models and achieved 99.8% accuracy in the BAAD16 dataset. Furthermore, we showed that the proposed system scales much better even with an increasing number of authors, and performance remains steady despite few training samples.

Jumman Hossain, Rajib Chandra Das, Md. Ruhul Amin et al.

There have thousands of crimes are happening daily all around. But people keep statistics only few of them, therefore crime rates are increasing day by day. The reason behind can be less concern or less statistics of previous crimes. It is much more important to observe the previous crime statistics for general people to make their outing decision and police for catching the criminals are taking steps to restrain the crimes and tourists to make their travelling decision. National institute of justice releases crime survey data for the country, but does not offer crime statistics up to Union or Thana level. Considering all of these cases we have come up with an approach which can give an approximation to people about the safety of a specific location with crime ranking of different areas locating the crimes on a map including a future crime occurrence prediction mechanism. Our approach relies on different online Bangla newspapers for crawling the crime data, stemming and keyword extraction, location finding algorithm, cosine similarity, naive Bayes classifier, and a custom crime prediction model

Nobel Dhar, Gaurob Saha, Prithwiraj Bhattacharjee et al.

Despite the success of the neural sequence-to-sequence model for abstractive text summarization, it has a few shortcomings, such as repeating inaccurate factual details and tending to repeat themselves. We propose a hybrid pointer generator network to solve the shortcomings of reproducing factual details inadequately and phrase repetition. We augment the attention-based sequence-to-sequence using a hybrid pointer generator network that can generate Out-of-Vocabulary words and enhance accuracy in reproducing authentic details and a coverage mechanism that discourages repetition. It produces a reasonable-sized output text that preserves the conceptual integrity and factual information of the input article. For evaluation, we primarily employed "BANSData" - a highly adopted publicly available Bengali dataset. Additionally, we prepared a large-scale dataset called "BANS-133" which consists of 133k Bangla news articles associated with human-generated summaries. Experimenting with the proposed model, we achieved ROUGE-1 and ROUGE-2 scores of 0.66, 0.41 for the "BANSData" dataset and 0.67, 0.42 for the BANS-133k" dataset, respectively. We demonstrated that the proposed system surpasses previous state-of-the-art Bengali abstractive summarization techniques and its stability on a larger dataset. "BANS-133" datasets and code-base will be publicly available for research.

Oishee Bintey Hoque, Mohammad Imrul Jubair, Al-Farabi Akash et al.

Bangladeshi Sign Language (BdSL) is a commonly used medium of communication for the hearing-impaired people in Bangladesh. A real-time BdSL interpreter with no controlled lab environment has a broad social impact and an interesting avenue of research as well. Also, it is a challenging task due to the variation in different subjects (age, gender, color, etc.), complex features, and similarities of signs and clustered backgrounds. However, the existing dataset for BdSL classification task is mainly built in a lab friendly setup which limits the application of powerful deep learning technology. In this paper, we introduce a dataset named BdSL36 which incorporates background augmentation to make the dataset versatile and contains over four million images belonging to 36 categories. Besides, we annotate about 40,000 images with bounding boxes to utilize the potentiality of object detection algorithms. Furthermore, several intensive experiments are performed to establish the baseline performance of our BdSL36. Moreover, we employ beta testing of our classifiers at the user level to justify the possibilities of real-world application with this dataset. We believe our BdSL36 will expedite future research on practical sign letter classification. We make the datasets and all the pre-trained models available for further researcher.

Nauros Romim, Mosahed Ahmed, Hriteshwar Talukder et al.

Social media sites such as YouTube and Facebook have become an integral part of everyone's life and in the last few years, hate speech in the social media comment section has increased rapidly. Detection of hate speech on social media websites faces a variety of challenges including small imbalanced data sets, the findings of an appropriate model and also the choice of feature analysis method. further more, this problem is more severe for the Bengali speaking community due to the lack of gold standard labelled datasets. This paper presents a new dataset of 30,000 user comments tagged by crowd sourcing and varified by experts. All the comments are collected from YouTube and Facebook comment section and classified into seven categories: sports, entertainment, religion, politics, crime, celebrity and TikTok & meme. A total of 50 annotators annotated each comment three times and the majority vote was taken as the final annotation. Nevertheless, we have conducted base line experiments and several deep learning models along with extensive pre-trained Bengali word embedding such as Word2Vec, FastText and BengFastText on this dataset to facilitate future research opportunities. The experiment illustrated that although all deep learning models performed well, SVM achieved the best result with 87.5% accuracy. Our core contribution is to make this benchmark dataset available and accessible to facilitate further research in the field of in the field of Bengali hate speech detection.

Ali Hasan Md. Linkon, Md. Mahir Labib, Faisal Haque Bappy et al.

Automatic detection and recognition of banknotes can be a very useful technology for people with visual difficulties and also for the banks itself by providing efficient management for handling different paper currencies. Lightweight models can easily be integrated into any handy IoT based gadgets/devices. This article presents our experiments on several state-of-the-art deep learning methods based on Lightweight Convolutional Neural Network architectures combining with transfer learning. ResNet152v2, MobileNet, and NASNetMobile were used as the base models with two different datasets containing Bangladeshi banknote images. The Bangla Currency dataset has 8000 Bangladeshi banknote images where the Bangla Money dataset consists of 1970 images. The performances of the models were measured using both the datasets and the combination of the two datasets. In order to achieve maximum efficiency, we used various augmentations, hyperparameter tuning, and optimizations techniques. We have achieved maximum test accuracy of 98.88\% on 8000 images dataset using MobileNet, 100\% on the 1970 images dataset using NASNetMobile, and 97.77\% on the combined dataset (9970 images) using MobileNet.

Manal, Ali Hasan Md. Linkon, Md. Mahir Labib et al.

In today's world, Neural Style Transfer (NST) has become a trendsetting term. NST combines two pictures, a content picture and a reference image in style (such as the work of a renowned painter) in a way that makes the output image look like an image of the material, but rendered with the form of a reference picture. However, there is no study using the artwork or painting of Bangladeshi painters. Bangladeshi painting has a long history of more than two thousand years and is still being practiced by Bangladeshi painters. This study generates NST stylized image on Bangladeshi paintings and analyzes the human point of view regarding the aesthetic preference of NST on Bangladeshi paintings. To assure our study's acceptance, we performed qualitative human evaluations on generated stylized images by 60 individual humans of different age and gender groups. We have explained how NST works for Bangladeshi paintings and assess NST algorithms, both qualitatively \& quantitatively. Our study acts as a pre-requisite for the impact of NST stylized image using Bangladeshi paintings on mobile UI/GUI and material translation from the human perspective. We hope that this study will encourage new collaborations to create more NST related studies and expand the use of Bangladeshi artworks.

Prithwiraj Bhattacharjee, Avi Mallick, Md Saiful Islam et al.

Abstractive summarization is the process of generating novel sentences based on the information extracted from the original text document while retaining the context. Due to abstractive summarization's underlying complexities, most of the past research work has been done on the extractive summarization approach. Nevertheless, with the triumph of the sequence-to-sequence (seq2seq) model, abstractive summarization becomes more viable. Although a significant number of notable research has been done in the English language based on abstractive summarization, only a couple of works have been done on Bengali abstractive news summarization (BANS). In this article, we presented a seq2seq based Long Short-Term Memory (LSTM) network model with attention at encoder-decoder. Our proposed system deploys a local attention-based model that produces a long sequence of words with lucid and human-like generated sentences with noteworthy information of the original document. We also prepared a dataset of more than 19k articles and corresponding human-written summaries collected from bangla.bdnews24.com1 which is till now the most extensive dataset for Bengali news document summarization and publicly published in Kaggle2. We evaluated our model qualitatively and quantitatively and compared it with other published results. It showed significant improvement in terms of human evaluation scores with state-of-the-art approaches for BANS.

Mahfuz Ahmed Masum, Sheikh Junayed Ahmed, Ayesha Tasnim et al.

Due to the breathtaking growth of social media or newspaper user comments, online product reviews comments, sentiment analysis (SA) has captured substantial interest from the researchers. With the fast increase of domain, SA work aims not only to predict the sentiment of a sentence or document but also to give the necessary detail on different aspects of the sentence or document (i.e. aspect-based sentiment analysis). A considerable number of datasets for SA and aspect-based sentiment analysis (ABSA) have been made available for English and other well-known European languages. In this paper, we present a manually annotated Bengali dataset of high quality, BAN-ABSA, which is annotated with aspect and its associated sentiment by 3 native Bengali speakers. The dataset consists of 2,619 positive, 4,721 negative and 1,669 neutral data samples from 9,009 unique comments gathered from some famous Bengali news portals. In addition, we conducted a baseline evaluation with a focus on deep learning model, achieved an accuracy of 78.75% for aspect term extraction and accuracy of 71.08% for sentiment classification. Experiments on the BAN-ABSA dataset show that the CNN model is better in terms of accuracy though Bi-LSTM significantly outperforms CNN model in terms of average F1-score.

Sarkar Snigdha Sarathi Das, Subangkar Karmaker Shanto, Masum Rahman et al.

Smartwatches or fitness trackers have garnered a lot of popularity as potential health tracking devices due to their affordable and longitudinal monitoring capabilities. To further widen their health tracking capabilities, in recent years researchers have started to look into the possibility of Atrial Fibrillation (AF) detection in real-time leveraging photoplethysmography (PPG) data, an inexpensive sensor widely available in almost all smartwatches. A significant challenge in AF detection from PPG signals comes from the inherent noise in the smartwatch PPG signals. In this paper, we propose a novel deep learning based approach, BayesBeat that leverages the power of Bayesian deep learning to accurately infer AF risks from noisy PPG signals, and at the same time provides an uncertainty estimate of the prediction. Extensive experiments on two publicly available dataset reveal that our proposed method BayesBeat outperforms the existing state-of-the-art methods. Moreover, BayesBeat is substantially more efficient having 40-200X fewer parameters than state-of-the-art baseline approaches making it suitable for deployment in resource constrained wearable devices.

Aisha Khatun, Anisur Rahman, Md. Saiful Islam et al.

Characters are the smallest unit of text that can extract stylometric signals to determine the author of a text. In this paper, we investigate the effectiveness of character-level signals in Authorship Attribution of Bangla Literature and show that the results are promising but improvable. The time and memory efficiency of the proposed model is much higher than the word level counterparts but accuracy is 2-5% less than the best performing word-level models. Comparison of various word-based models is performed and shown that the proposed model performs increasingly better with larger datasets. We also analyze the effect of pre-training character embedding of diverse Bangla character set in authorship attribution. It is seen that the performance is improved by up to 10% on pre-training. We used 2 datasets from 6 to 14 authors, balancing them before training and compare the results.

Aisha Khatun, Anisur Rahman, Hemayet Ahmed Chowdhury et al.

Language models are at the core of natural language processing. The ability to represent natural language gives rise to its applications in numerous NLP tasks including text classification, summarization, and translation. Research in this area is very limited in Bangla due to the scarcity of resources, except for some count-based models and very recent neural language models being proposed, which are all based on words and limited in practical tasks due to their high perplexity. This paper attempts to approach this issue of perplexity and proposes a subword level neural language model with the AWD-LSTM architecture and various other techniques suitable for training in Bangla language. The model is trained on a corpus of Bangla newspaper articles of an appreciable size consisting of more than 28.5 million word tokens. The performance comparison with various other models depicts the significant reduction in perplexity the proposed model provides, reaching as low as 39.84, in just 20 epochs.

Md Mahadi Hasan Nahid, Md. Ashraful Islam, Bishwajit Purkaystha et al.

Speech recognition has received a less attention in Bengali literature due to the lack of a comprehensive dataset. In this paper, we describe the development process of the first comprehensive Bengali speech dataset on real numbers. It comprehends all the possible words that may arise in uttering any Bengali real number. The corpus has ten speakers from the different regions of Bengali native people. It comprises of more than two thousands of speech samples in a total duration of closed to four hours. We also provide a deep analysis of our corpus, highlight some of the notable features of it, and finally evaluate the performances of two of the notable Bengali speech recognizers on it.

Md. Saiful Islam, Fazla Elahi Md Jubayer, Syed Ikhtiar Ahmed

Document categorization is a technique where the category of a document is determined. In this paper three well-known supervised learning techniques which are Support Vector Machine(SVM), Naïve Bayes(NB) and Stochastic Gradient Descent(SGD) compared for Bengali document categorization. Besides classifier, classification also depends on how feature is selected from dataset. For analyzing those classifier performances on predicting a document against twelve categories several feature selection techniques are also applied in this article namely Chi square distribution, normalized TFIDF (term frequency-inverse document frequency) with word analyzer. So, we attempt to explore the efficiency of those three-classification algorithms by using two different feature selection techniques in this article.

Sadia Tasnim Swarna, Shamim Ehsan, Md. Saiful Islam et al.

Hidden Markov model based various phoneme recognition methods for Bengali language is reviewed. Automatic phoneme recognition for Bengali language using multilayer neural network is reviewed. Usefulness of multilayer neural network over single layer neural network is discussed. Bangla phonetic feature table construction and enhancement for Bengali speech recognition is also discussed. Comparison among these methods is discussed.