Master's Thesis

1. Docusage: Harnessing Hierarchical Clustering in Salience-Driven Narrative Synthesis

   Abstract: Text summarization remains a crucial yet challenging task in natural language processing, especially as the volume of text data grows exponentially. This thesis introduces Sumsage, a new optimization-based text summarization method that synthesizes concise yet informative summaries. Our work presents several notable contributions to the field. We developed the Syn-D-sum dataset from the CNN/DailyMail dataset, creating a robust resource for training and evaluating summarization models. We also propose the Sumsage algorithm, which leverages hierarchical clustering to extract key sentences and construct coherent summaries, closely emulating human summarizers. Additionally, we designed two new evaluation methods: the Symphony penalty and the Captured Importance Quantification scores, which assess the quality of generated summaries by considering both narrative structure and sentence order. Sumsage’s dynamic tree structure and hierarchical clustering approach enable efficient and scalable summarization while maintaining contextual relevance and minimizing hallucination. Additionally, our experiments show that Sumsage yields superior performance over GPT-3.5-turbo, generating summaries similar to those written by humans and capturing more essential information. Sumsage represents a novel advancement in text summarization, offering a robust and interpretable method for generating high-quality summaries. This approach not only addresses current challenges but also lays the foundation for future innovations in narrative synthesis and evaluation.

   Read more

Peer Reviewed Publications

As of 15 April 2025 - Citations: 23 | h-index: 3

1. Harnessing Hierarchical Clustering in Salience-Driven Text Summarization

   Authors: Akib Sadmanee, Sukhwa Hong, Jason Leigh, Mahdi Belcaid

   Conference: The 31st Americas Conference on Information Systems 2025 (Impact Factor: 1.00)

   Contribution: I took a lead role in coordinating the work, helping to design and set up each experiment and making sure we had the right data, tools, and settings. I ran the tests, collected the results, and organized the data for analysis. Using straightforward statistical methods and clear charts, I helped interpret what the numbers showed. I also drafted the introduction, methods, results, and discussion sections of the paper, then worked closely with my co-authors to refine and polish the final manuscript.

   In press - Accepted for presentation August 2025 in Montréal, Canada.

2. Intrinsic Evaluation of Bangla Word Embeddings

   Authors: Akib Sadmanee ^*, Nafiz Sadman^*, Md Iftekhar Tanveer, Md Ashraful Amin, Amin Ahsan Ali (* co-first author)

   Conference: International Conference on Bangla Speech and Language Processing, 2019 (Acceptance Rate: 28%)

   Contribution: Led the project by strategically designing experiments and implementing a robust evaluation framework to comprehensively assess Bangla word embeddings. Developed tailored intrinsic evaluation metrics and executed extensive comparative experiments across various embedding models, providing insightful analysis of their semantic and syntactic capabilities. Also drafted the manuscript for the systematic approach that significantly enhanced understanding of embedding performance and effectiveness in the context of Bangla language.

   Read more

3. Variational stacked local attention networks for diverse video captioning

   Authors: Tonmoay Deb, Akib Sadmanee, Kishor Kumar Bhaumik, Amin Ahsan Ali, M Ashraful Amin, AKM Rahman

   Conference: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2022 (Acceptance Rate: 27.4%)

   Contribution: Worked on the language modeling and evaluation part of the project. My task was to generate, verify and evaluate the Captions generated from videos. I designed the caption generation and evaluation system by fine tuning existing language models.

   Read more

4. A Safer Approach to build Recommendation Systems on Unidentifiable Data

   Authors: Kishor Datta Gupta, Akib Sadmanee, Nafiz Sadman

   Conference: Proceedings of the 14th International Conference on Agents and Artificial Intelligence, 2022 (Acceptance Rate: 26%)

   Contribution: Led the project by completing experiment design and implemention. Designed the UI where the users interracted with the system and watched videos for data collection purposes.

   Read more

5. HeteroGenius: An Improvised ‘Intelligence’ in Heterogeneous Graph Transformers

   Authors: Nafiz Sadman, Akib Sadmanee, Kishor Datta Gupta, Roy George

   Conference: International Conference on Machine Learning and Applications, 2022 (Acceptance Rate: 31%)

   Contribution: Worked mainly as a programmer and proofreader on the project. In the research part, my task was to maintain the therotical integrity of the transformer model design.

   Read more

6. Behavioral recommendation engine driven by only non-identifiable user data

   Authors: Kishor Datta Gupta, Nafiz Sadman, Akib Sadmanee, Md Kamruzzaman Sarker, Roy George

   Journal: Machine Learning with Applications, 2023 (Impact Factor: 6.0)

   Contribution: Worked mostly as a programmer running the experiments and as a proofreader. The work was built on top of my other work where we built a recommendation system using only unidentifiable data

   Read more