Thi-Thiet Pham
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
Research Interests
Thi-Thiet Pham has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
Jiawei Luo
4 shared publicationsBay Vo
3 shared publications- Tzung-Pei Hong
Department of Computer Science and Information Engineering, National University of Kaohsiung, Kaohsiung 811726, Taiwan
3 shared publications - Quoc-Huy Le
Faculty of Information Technology, Industrial University of Ho Chi Minh City, Ho Chi Minh City, Vietnam
2 shared publications - Bay Vo
HUTECH University
1 shared publication - Huy Tran
Ho Chi Minh City University of Technology (HCMUT)
1 shared publication
Latest Publications
- Open Access
- 5 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 391-396
A Multi-Agent System for Equity Analysis
Financial markets produce large volumes of heterogeneous data that must be synthesized to inform investment decisions and risk evaluation. Conventional machine learning models offer strong predictive performance but demand programming expertise and lack natural-language interactivity, while large language models provide fluent communication yet cannot independently perform numerical simulations or scenario-based reasoning. This paper introduces a multi-agent architecture that unifies analytical AI and generative AI to support equity analysis. The system is orchestrated by a Manager Agent and includes six specialized agents: a Chat Agent for conversational interaction, a Text-to-SQL Agent enabling natural-language access to structured databases, an Unstructured Data Agent for document retrieval, an ML/SHAP Agent for explainable stock-movement prediction, a Visualization Agent for generating decision-support plots, and a Scenario Analysis Agent for simulating perturbed market conditions. Through a unified natural-language interface, the system integrates structured market data with unstructured financial documents. Experiments on Vietnam stock market data (January 2024–August 2025) show strong performance across components: the Text-to-SQL agent achieves 90% validity on complex queries with iterative correction, the Unstructured Data agent attains 91.9% accuracy on document-grounded questions, and ML outputs align 100% with manual benchmarks. The proposed architecture accelerates analytical workflows while ensuring transparency and auditability, offering a practical foundation for deploying multi-agent AI systems in emerging market equity research and providing pathways for future extensions in trading, risk management, and compliance applications.
- 5 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 391-396
Open AccessA Multi-Agent System for Equity Analysis
Financial markets produce large volumes of heterogeneous data that must be synthesized to inform investment decisions and risk evaluation. Conventional machine learning models offer strong predictive performance but demand programming expertise and lack natural-language interactivity, while large language models provide fluent communication yet cannot independently perform numerical simulations or scenario-based reasoning. This paper introduces a multi-agent architecture that unifies analytical AI and generative AI to support equity analysis. The system is orchestrated by a Manager Agent and includes six specialized agents: a Chat Agent for conversational interaction, a Text-to-SQL Agent enabling natural-language access to structured databases, an Unstructured Data Agent for document retrieval, an ML/SHAP Agent for explainable stock-movement prediction, a Visualization Agent for generating decision-support plots, and a Scenario Analysis Agent for simulating perturbed market conditions. Through a unified natural-language interface, the system integrates structured market data with unstructured financial documents. Experiments on Vietnam stock market data (January 2024–August 2025) show strong performance across components: the Text-to-SQL agent achieves 90% validity on complex queries with iterative correction, the Unstructured Data agent attains 91.9% accuracy on document-grounded questions, and ML outputs align 100% with manual benchmarks. The proposed architecture accelerates analytical workflows while ensuring transparency and auditability, offering a practical foundation for deploying multi-agent AI systems in emerging market equity research and providing pathways for future extensions in trading, risk management, and compliance applications.
- 2 December 2025
- Published by Springer Nature
- p. 279-289
Early Prediction Under Class Imbalance for a Programming Course: Feature Selection and Data Augmentation
- 2 December 2025
- Published by Springer Nature
- p. 279-289
Early Prediction Under Class Imbalance for a Programming Course: Feature Selection and Data Augmentation
- 1 July 2015
- Published by SAGE Publications in Fundamenta Informaticae
- Vol. 138 (3) , 373-386
Efficiently Mining Sequential Generator Patterns Using Prefix Trees
Mining long frequent sequences that contain a combinatorial number of frequent subsequences or using very low support thresholds to mine sequential patterns is usually both time- and memory-consuming. The mining of closed sequential patterns, sequential generator patterns, and maximum sequences has been proposed to overcome this problem. Sequential generator patterns, when used together with closed sequential patterns, can provide additional information that closed sequential patterns alone cannot provide. Mining sequential generator patterns is thus an important task in data mining as well. This paper proposes an algorithm called MSGP-PreTree for mining all sequential generator patterns based on the prefix-tree structure. The algorithm uses the characteristics of sequential generator patterns and sequence extensions to efficiently perform a depth-first search on a prefix tree. It also uses a vertical approach to list and count the supports of sequences based on the prime block encoding approach for representing candidate sequences and determining the frequencies of candidates. Besides, the search space of the MSGP-PreTree algorithm is much smaller than those of other algorithms because two pruning strategies are applied. Experimental results conducted on synthetic and real databases show that the proposed algorithm is effective than a previous one.
- 1 July 2015
- Published by SAGE Publications in Fundamenta Informaticae
- Vol. 138 (3) , 373-386
Efficiently Mining Sequential Generator Patterns Using Prefix Trees
Mining long frequent sequences that contain a combinatorial number of frequent subsequences or using very low support thresholds to mine sequential patterns is usually both time- and memory-consuming. The mining of closed sequential patterns, sequential generator patterns, and maximum sequences has been proposed to overcome this problem. Sequential generator patterns, when used together with closed sequential patterns, can provide additional information that closed sequential patterns alone cannot provide. Mining sequential generator patterns is thus an important task in data mining as well. This paper proposes an algorithm called MSGP-PreTree for mining all sequential generator patterns based on the prefix-tree structure. The algorithm uses the characteristics of sequential generator patterns and sequence extensions to efficiently perform a depth-first search on a prefix tree. It also uses a vertical approach to list and count the supports of sequences based on the prime block encoding approach for representing candidate sequences and determining the frequencies of candidates. Besides, the search space of the MSGP-PreTree algorithm is much smaller than those of other algorithms because two pruning strategies are applied. Experimental results conducted on synthetic and real databases show that the proposed algorithm is effective than a previous one.
- 1 June 2014
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 32, 88-99
An efficient method for mining non-redundant sequential rules using attributed prefix-trees
- 1 June 2014
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 32, 88-99
An efficient method for mining non-redundant sequential rules using attributed prefix-trees
- 1 January 2014
- Published by Inderscience Publishers in International Journal of Intelligent Systems Technologies and Applications
- Vol. 13 (3) , 157
An efficient algorithm for mining sequential generator pattern using prefix trees and hash tables
Mining long frequent sequences that contain a combinatorial number of frequent subsequences or using very low support thresholds to mine sequential patterns is both time- and memory-consuming. The mining of closed sequential patterns, sequential generator patterns, and maximum sequences has been proposed to overcome this problem. This paper proposes an algorithm for generating all sequential generator patterns. This algorithm uses a vertical approach to listing and counting the support of sequence based on the prime block encoding approach to represent candidate sequences and determine the frequency for each candidate. The search space of the proposed algorithm is much smaller than those of other algorithms because super sequence frequency-based pruning and non-generator-based pruning are applied. Besides, hash tables are also used for fast checking the existed sequential generator patterns. Experimental results conducted on synthetic and real databases show that the proposed algorithm is effective.
- 1 January 2014
- Published by Inderscience Publishers in International Journal of Intelligent Systems Technologies and Applications
- Vol. 13 (3) , 157
An efficient algorithm for mining sequential generator pattern using prefix trees and hash tables
Mining long frequent sequences that contain a combinatorial number of frequent subsequences or using very low support thresholds to mine sequential patterns is both time- and memory-consuming. The mining of closed sequential patterns, sequential generator patterns, and maximum sequences has been proposed to overcome this problem. This paper proposes an algorithm for generating all sequential generator patterns. This algorithm uses a vertical approach to listing and counting the support of sequence based on the prime block encoding approach to represent candidate sequences and determine the frequency for each candidate. The search space of the proposed algorithm is much smaller than those of other algorithms because super sequence frequency-based pruning and non-generator-based pruning are applied. Besides, hash tables are also used for fast checking the existed sequential generator patterns. Experimental results conducted on synthetic and real databases show that the proposed algorithm is effective.
- 1 January 2013
- Published by Inderscience Publishers in International Journal of Intelligent Information and Database Systems
- Vol. 7 (4) , 324
An effective algorithm for mining closed sequential patterns and their minimal generators based on prefix trees
Sequential generator patterns and closed sequential patterns play an important role in data mining tasks. They are proposed to address difficult problems in mining sequential pattern and have often been used together to generate non-redundant rules. Based on their important role, this paper proposes an efficient algorithm called CloGen for mining closed sequential patterns and their minimal sequential generator patterns. The CloGen algorithm uses the parent-child relationship on prefix tree structure and inserts fields into each node on prefix tree to determine whether that is a minimal sequential generator pattern or closed sequential pattern. Experimental results show that the performance runtime of CloGen algorithm is much faster than that of other algorithms by more than one order of magnitude.
- 1 January 2013
- Published by Inderscience Publishers in International Journal of Intelligent Information and Database Systems
- Vol. 7 (4) , 324
An effective algorithm for mining closed sequential patterns and their minimal generators based on prefix trees
Sequential generator patterns and closed sequential patterns play an important role in data mining tasks. They are proposed to address difficult problems in mining sequential pattern and have often been used together to generate non-redundant rules. Based on their important role, this paper proposes an efficient algorithm called CloGen for mining closed sequential patterns and their minimal sequential generator patterns. The CloGen algorithm uses the parent-child relationship on prefix tree structure and inserts fields into each node on prefix tree to determine whether that is a minimal sequential generator pattern or closed sequential pattern. Experimental results show that the performance runtime of CloGen algorithm is much faster than that of other algorithms by more than one order of magnitude.
- 1 January 2012
- Published by Springer Nature
- p. 393-401
MSGPs: A Novel Algorithm for Mining Sequential Generator Patterns
- 1 January 2012
- Published by Springer Nature
- p. 393-401
MSGPs: A Novel Algorithm for Mining Sequential Generator Patterns