Sourav Mandal
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Sourav Mandal has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Arif Ahmed Sekh
UiT The Arctic University of Norway
8 shared publications - Rudra Mohan Tripathy
School of Computer Science and Engineering, XIM University, Bhubaneswar, India
6 shared publications - Barsha Pattanaik
XIM University
5 shared publications - Rohini Basak
Jadavpur University
4 shared publications - Dilip Kumar Prasad
UiT The Arctic University of Norway, Tromsø, Norway
3 shared publications - Sudip Kumar Naskar
Jadavpur University, Kolkata, India
3 shared publications
Latest Publications
- Open Access
- 21 September 2026
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 6 (1) , 1201
Dual stream adaptive sequential recommendation
Recommendation systems are AI-tools for assisting users to choose the items they may like. Based on what a user has clicked, rated, or purchased in the past, the system learns to predict what the user may want next. This paper proposes the Dual-Stream Adaptive Sequential Recommendation system (DSASR). DSASR employs two independent causal transformer encoders: one captures short-term preferences from recent interactions; the other captures stable long-term preferences from historical behavior. A Consistency Gate computes the preference drift to represent how much the recent preference has shifted from long-term history, and uses it to compute a personalized blend weight conditioned on user demographic attributes including age, gender, occupation, and a learned identity embedding. For training, combined Binary Cross-Entropy (BCE) and Bayesian Personalized Ranking (BPR) loss have been used, directly optimizing the ranking metrics used at evaluation. Experiments have been conducted on MovieLens 100K, MovieLens 1 M, and Amazon Beauty datasets. The model has been evaluated on metrics such as Hit Rate at N (HR@N), Normalized Discounted Cumulative Gain at N (NDCG@N), and Mean Reciprocal Rank at N (MRR@N). The results show that DSASR consistently outperforms existing methods, achieving HR@10 = 0.7253, NDCG@10 = 0.4650, MRR@10 = 0.3627 for MovieLens 100K, HR@10 = 0.8043, NDCG@10 = 0.5832, MRR@10 = 0.5130 for MovieLens 1 M, and HR@10 = 0.4353, NDCG@10 = 0.2686, MRR@10 = 0.2172 for Amazon Beauty. The improvement is due to the Consistency Gate computing a personalized blend weight for each user, which takes user demographic attributes as well as preference drift into account, making the recommendations more personalized than existing systems. The code is available in https://github.com/skarifahmed/DSASR .
- 21 September 2026
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 6 (1) , 1201
Open AccessDual stream adaptive sequential recommendation
Recommendation systems are AI-tools for assisting users to choose the items they may like. Based on what a user has clicked, rated, or purchased in the past, the system learns to predict what the user may want next. This paper proposes the Dual-Stream Adaptive Sequential Recommendation system (DSASR). DSASR employs two independent causal transformer encoders: one captures short-term preferences from recent interactions; the other captures stable long-term preferences from historical behavior. A Consistency Gate computes the preference drift to represent how much the recent preference has shifted from long-term history, and uses it to compute a personalized blend weight conditioned on user demographic attributes including age, gender, occupation, and a learned identity embedding. For training, combined Binary Cross-Entropy (BCE) and Bayesian Personalized Ranking (BPR) loss have been used, directly optimizing the ranking metrics used at evaluation. Experiments have been conducted on MovieLens 100K, MovieLens 1 M, and Amazon Beauty datasets. The model has been evaluated on metrics such as Hit Rate at N (HR@N), Normalized Discounted Cumulative Gain at N (NDCG@N), and Mean Reciprocal Rank at N (MRR@N). The results show that DSASR consistently outperforms existing methods, achieving HR@10 = 0.7253, NDCG@10 = 0.4650, MRR@10 = 0.3627 for MovieLens 100K, HR@10 = 0.8043, NDCG@10 = 0.5832, MRR@10 = 0.5130 for MovieLens 1 M, and HR@10 = 0.4353, NDCG@10 = 0.2686, MRR@10 = 0.2172 for Amazon Beauty. The improvement is due to the Consistency Gate computing a personalized blend weight for each user, which takes user demographic attributes as well as preference drift into account, making the recommendations more personalized than existing systems. The code is available in https://github.com/skarifahmed/DSASR .
- Open Access
- 7 July 2026
- Published by Springer Nature in Discover Computing
- Vol. 29 (1) , 425
A topic-driven local interpretable model for explainable rumor detection on social media
- 7 July 2026
- Published by Springer Nature in Discover Computing
- Vol. 29 (1) , 425
Open AccessA topic-driven local interpretable model for explainable rumor detection on social media
- 2 January 2026
- Published by Springer Nature
- p. 293-305
Explainable Rumor Detection Using Topic Modeling
- 2 January 2026
- Published by Springer Nature
- p. 293-305
Explainable Rumor Detection Using Topic Modeling
- Open Access
- 30 September 2025
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 5 (1) , 1-21
Convolutional block attention-based deep neural network for mushroom classification
- 30 September 2025
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 5 (1) , 1-21
Open AccessConvolutional block attention-based deep neural network for mushroom classification
- Open Access
- 17 September 2025
- Published by Springer Nature in Progress in Artificial Intelligence
- Vol. 14 (4) , 505-530
Deep learning-based kinship verification: a comprehensive survey
- 17 September 2025
- Published by Springer Nature in Progress in Artificial Intelligence
- Vol. 14 (4) , 505-530
Open AccessDeep learning-based kinship verification: a comprehensive survey
- Open Access
- 1 April 2025
- Published by Institute of Advanced Engineering and Science in International Journal of Informatics and Communication Technology (IJ-ICT)
- Vol. 14 (1) , 276-286
Ensemble approach to rumor detection with BERT, GPT, and POS features
As vast amounts of rumor content are transmitted on social media, it is very challenging to detect them. This study explores an ensemble approach to rumor detection in social media messages, leveraging the strengths of advanced natural language processing (NLP) models. Specifically, we implemented three distinct models: (i) generative pre-trained transformer (GPT) combined with a bidirectional long short-term memory (BiLSTM) network; (ii) a model integrating part-of-speech (POS) tagging with bidirectional encoder representations from transformers (BERT) and BiLSTM, and (iii) a model that merges POS tagging with GPT and BiLSTM. We included additional features from the dataset in all these models. Each model captures different linguistic, syntactical, and contextual features within the text, contributing uniquely to the classification task. To enhance the robustness and accuracy of our predictions, we employed an ensemble method using hard voting. This technique aggregates the predictions from each model, determining the final classification based on the majority vote. Our experimental results demonstrate that the ensemble approach significantly outperforms individual models, achieving superior accuracy in identifying rumors. To determine the performance of our model, we used PHEME and Weibo datasets available publicly. We found our model gave 97.6% and 98.4% accuracy, respectively, on the datasets and has outperformed the state-of-the-art models.
- 1 April 2025
- Published by Institute of Advanced Engineering and Science in International Journal of Informatics and Communication Technology (IJ-ICT)
- Vol. 14 (1) , 276-286
Open AccessEnsemble approach to rumor detection with BERT, GPT, and POS features
As vast amounts of rumor content are transmitted on social media, it is very challenging to detect them. This study explores an ensemble approach to rumor detection in social media messages, leveraging the strengths of advanced natural language processing (NLP) models. Specifically, we implemented three distinct models: (i) generative pre-trained transformer (GPT) combined with a bidirectional long short-term memory (BiLSTM) network; (ii) a model integrating part-of-speech (POS) tagging with bidirectional encoder representations from transformers (BERT) and BiLSTM, and (iii) a model that merges POS tagging with GPT and BiLSTM. We included additional features from the dataset in all these models. Each model captures different linguistic, syntactical, and contextual features within the text, contributing uniquely to the classification task. To enhance the robustness and accuracy of our predictions, we employed an ensemble method using hard voting. This technique aggregates the predictions from each model, determining the final classification based on the majority vote. Our experimental results demonstrate that the ensemble approach significantly outperforms individual models, achieving superior accuracy in identifying rumors. To determine the performance of our model, we used PHEME and Weibo datasets available publicly. We found our model gave 97.6% and 98.4% accuracy, respectively, on the datasets and has outperformed the state-of-the-art models.
- Open Access
- 13 March 2025
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 5 (1) , 25
BMWP: the first Bengali math word problems dataset for operation prediction and solving
Solving math word problems of varying complexities is one of the most challenging and exciting research questions in artificial intelligence (AI), particularly in natural language processing (NLP) and machine learning (ML). Foundational language models such as GPT must be evaluated for intelligence, and solving word problems is a key method for this assessment. These problems become especially difficult when presented in low-resource regional languages such as Bengali. Word problem solving integrates the cognitive domains of language processing, comprehension, and transformation into real-world solutions. During the past decade, advances in AI and machine learning have significantly progressed in addressing this complex issue. Although researchers worldwide have primarily utilized datasets in English and some in Chinese, there has been a lack of standard datasets for low-resource languages such as Bengali. In this pioneering study, we introduce the first Bengali Math Word Problem Benchmark Data Set (BMWP), comprising 8653 word problems. We detail the creation of this dataset and the benchmarking methods employed. Furthermore, we investigate operation prediction from Bengali word problems using state-of-the-art deep learning (DL) techniques. We implemented and compared various standard DL-based neural network architectures, achieving an accuracy of $$92 \pm 2\%$$ . The data set and the code will be available at https://github.com/SanchitaMondal/BMWP.- 13 March 2025
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 5 (1) , 25
Open AccessBMWP: the first Bengali math word problems dataset for operation prediction and solving
Solving math word problems of varying complexities is one of the most challenging and exciting research questions in artificial intelligence (AI), particularly in natural language processing (NLP) and machine learning (ML). Foundational language models such as GPT must be evaluated for intelligence, and solving word problems is a key method for this assessment. These problems become especially difficult when presented in low-resource regional languages such as Bengali. Word problem solving integrates the cognitive domains of language processing, comprehension, and transformation into real-world solutions. During the past decade, advances in AI and machine learning have significantly progressed in addressing this complex issue. Although researchers worldwide have primarily utilized datasets in English and some in Chinese, there has been a lack of standard datasets for low-resource languages such as Bengali. In this pioneering study, we introduce the first Bengali Math Word Problem Benchmark Data Set (BMWP), comprising 8653 word problems. We detail the creation of this dataset and the benchmarking methods employed. Furthermore, we investigate operation prediction from Bengali word problems using state-of-the-art deep learning (DL) techniques. We implemented and compared various standard DL-based neural network architectures, achieving an accuracy of $$92 \pm 2\%$$ . The data set and the code will be available at https://github.com/SanchitaMondal/BMWP. - 1 March 2025
- Published by Springer Nature
- p. 184-195
MythBuster: A Comparative Analysis of Few Machine Learning and Deep Learning Models for Fake News Detection
- 1 March 2025
- Published by Springer Nature
- p. 184-195
MythBuster: A Comparative Analysis of Few Machine Learning and Deep Learning Models for Fake News Detection
- Open Access
- 19 November 2024
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 4 (1) , 1-15
Rumor detection using dual embeddings and text-based graph convolutional network
Social media platforms like Twitter and Facebook have gradually become vital for communication and information exchange. However, this often leads to the spread of unreliable or false information, such as harmful rumors. Currently, graph convolutional networks (GCNs), particularly TextGCN, have shown promise in text classification tasks, including rumor detection. Their success is due to their ability to identify structural patterns in rumors and effectively use neighborhood information. We present a novel rumor detection model using TextGCN, which utilizes a word-document graph to represent rumor texts. This model uses dual embedding from two pre-trained transformer models: generative pre-trained transformers (GPT) and bidirectional encoder representations from transformers (BERT). These embeddings serve as node representations within the graph, enhancing rumor detection. Combining these deep neural networks effectively extracts significant contextual features from rumors. This graph undergoes convolution, and through graph-based learning, the model detects a rumor. We evaluated our model using publicly available rumor datasets, such as PHEME, Twitter15, and Twitter16. It achieved 88.64% accuracy on the PHEME dataset, surpassing similar models, and performed well on Twitter15 and Twitter16 with accuracies of 81.98% and 83.41%, respectively.
- 19 November 2024
- Published by Springer Nature in Discover Artificial Intelligence
- Vol. 4 (1) , 1-15
Open AccessRumor detection using dual embeddings and text-based graph convolutional network
Social media platforms like Twitter and Facebook have gradually become vital for communication and information exchange. However, this often leads to the spread of unreliable or false information, such as harmful rumors. Currently, graph convolutional networks (GCNs), particularly TextGCN, have shown promise in text classification tasks, including rumor detection. Their success is due to their ability to identify structural patterns in rumors and effectively use neighborhood information. We present a novel rumor detection model using TextGCN, which utilizes a word-document graph to represent rumor texts. This model uses dual embedding from two pre-trained transformer models: generative pre-trained transformers (GPT) and bidirectional encoder representations from transformers (BERT). These embeddings serve as node representations within the graph, enhancing rumor detection. Combining these deep neural networks effectively extracts significant contextual features from rumors. This graph undergoes convolution, and through graph-based learning, the model detects a rumor. We evaluated our model using publicly available rumor datasets, such as PHEME, Twitter15, and Twitter16. It achieved 88.64% accuracy on the PHEME dataset, surpassing similar models, and performed well on Twitter15 and Twitter16 with accuracies of 81.98% and 83.41%, respectively.