Behnam Bahrak
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Behnam Bahrak has a total of 68 co-authors. Here are the top co-authors by number of shared publications:
- Ghazal Kalhor
School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran
4 shared publications - Zhivar Sourati
University of Southern California
4 shared publications - Amirhossein Abaskohi
Department of Electrical and Computer Engineering University of Tehran Tehran Iran
3 shared publications - Houmaan Chamani
Department of Electrical and Computer Engineering University of Tehran Tehran Iran
3 shared publications - Nazanin Sabri
University of California San Diego, La Jolla, California, USA
3 shared publications - Seyed Pooya Shariatpanahi
School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran
3 shared publications
Latest Publications
- Open Access
- 30 July 2026
- Published by MDPI AG in Journal of Theoretical and Applied Electronic Commerce Research
- Vol. 21 (8) , 241
Transformer-Based Price Personalization: Attention-Based Modeling of Willingness to Pay in Product Recommendation
The growing convergence of personalized recommendations and pricing strategies has created new opportunities for retailers to support more targeted promotional decisions and potentially improve revenue-related outcomes, provided that such systems are implemented with appropriate managerial and fairness safeguards. This study introduces a Transformer-driven modeling approach that jointly performs product recommendation and personalized price-level assignment by modeling observed discount-level acceptance as an operational proxy for consumers’ willingness to pay (WTP). Instead of handling price as an auxiliary or fixed attribute, the proposed model learns patterns of observed price responsiveness from sequential purchase histories that integrate user actions, item attributes, and discount exposure. The learning task is cast as a multi-class classification problem with five outcomes—non-purchase and purchases occurring at 0%, 10%, 20%, and above 30% discount levels—enabling the system to recommend not only relevant products but also economically appropriate pricing options. Empirical evaluation on the Dunnhumby retail dataset, which provides rich information on transactions and promotions, shows consistent performance across multiple evaluation criteria, including macro- and micro-level classification metrics, ranking-based measures such as NDCG@5, and Macro averaged AUC. Across the reported metrics, the Transformer-based framework performs better than our earlier deep reinforcement learning approach in this five-class setting, suggesting that sequence-based self-attention can better capture observed behavioral patterns related to discount-level acceptance. By producing integrated product–price recommendations, the proposed framework contributes toward closing the methodological gap between recommendation systems and dynamic pricing, and offers a scalable decision-support framework for data-driven, value-sensitive promotion design in retail and e-commerce contexts, subject to future validation in real-world and fairness-aware implementation settings.
- 30 July 2026
- Published by MDPI AG in Journal of Theoretical and Applied Electronic Commerce Research
- Vol. 21 (8) , 241
Open AccessTransformer-Based Price Personalization: Attention-Based Modeling of Willingness to Pay in Product Recommendation
The growing convergence of personalized recommendations and pricing strategies has created new opportunities for retailers to support more targeted promotional decisions and potentially improve revenue-related outcomes, provided that such systems are implemented with appropriate managerial and fairness safeguards. This study introduces a Transformer-driven modeling approach that jointly performs product recommendation and personalized price-level assignment by modeling observed discount-level acceptance as an operational proxy for consumers’ willingness to pay (WTP). Instead of handling price as an auxiliary or fixed attribute, the proposed model learns patterns of observed price responsiveness from sequential purchase histories that integrate user actions, item attributes, and discount exposure. The learning task is cast as a multi-class classification problem with five outcomes—non-purchase and purchases occurring at 0%, 10%, 20%, and above 30% discount levels—enabling the system to recommend not only relevant products but also economically appropriate pricing options. Empirical evaluation on the Dunnhumby retail dataset, which provides rich information on transactions and promotions, shows consistent performance across multiple evaluation criteria, including macro- and micro-level classification metrics, ranking-based measures such as NDCG@5, and Macro averaged AUC. Across the reported metrics, the Transformer-based framework performs better than our earlier deep reinforcement learning approach in this five-class setting, suggesting that sequence-based self-attention can better capture observed behavioral patterns related to discount-level acceptance. By producing integrated product–price recommendations, the proposed framework contributes toward closing the methodological gap between recommendation systems and dynamic pricing, and offers a scalable decision-support framework for data-driven, value-sensitive promotion design in retail and e-commerce contexts, subject to future validation in real-world and fairness-aware implementation settings.
- 10 April 2026
- Published by Springer Nature in Transportation
- p. 1-34
GraphSAGE-based travel demand prediction for expanding networks
- 10 April 2026
- Published by Springer Nature in Transportation
- p. 1-34
GraphSAGE-based travel demand prediction for expanding networks
- Open Access
The Sufficiency-Conciseness Trade-off in LLM Self-Explanation from an Information Bottleneck Perspective
Large Language Models increasingly rely on self-explanations, such as chain of thought reasoning, to improve performance on multi step question answering. While these explanations enhance accuracy, they are often verbose and costly to generate, raising the question of how much explanation is truly necessary. In this paper, we examine the trade-off between sufficiency, defined as the ability of an explanation to justify the correct answer, and conciseness, defined as the reduction in explanation length. Building on the information bottleneck principle, we conceptualize explanations as compressed representations that retain only the information essential for producing correct answers.To operationalize this view, we introduce an evaluation pipeline that constrains explanation length and assesses sufficiency using multiple language models on the ARC Challenge dataset. To broaden the scope, we conduct experiments in both English, using the original dataset, and Persian, as a resource-limited language through translation. Our experiments show that more concise explanations often remain sufficient, preserving accuracy while substantially reducing explanation length, whereas excessive compression leads to performance degradation.
Open AccessThe Sufficiency-Conciseness Trade-off in LLM Self-Explanation from an Information Bottleneck Perspective
Large Language Models increasingly rely on self-explanations, such as chain of thought reasoning, to improve performance on multi step question answering. While these explanations enhance accuracy, they are often verbose and costly to generate, raising the question of how much explanation is truly necessary. In this paper, we examine the trade-off between sufficiency, defined as the ability of an explanation to justify the correct answer, and conciseness, defined as the reduction in explanation length. Building on the information bottleneck principle, we conceptualize explanations as compressed representations that retain only the information essential for producing correct answers.To operationalize this view, we introduce an evaluation pipeline that constrains explanation length and assesses sufficiency using multiple language models on the ARC Challenge dataset. To broaden the scope, we conduct experiments in both English, using the original dataset, and Persian, as a resource-limited language through translation. Our experiments show that more concise explanations often remain sufficient, preserving accuracy while substantially reducing explanation length, whereas excessive compression leads to performance degradation.
- Open Access
A Hybrid TGN-SEAL Model for Dynamic Graph Link Prediction
Predicting links in sparse, continuously evolving networks is a central challenge in network science. Conventional heuristic methods and deep learning models, including Graph Neural Networks (GNNs), are typically designed for static graphs and thus struggle to capture temporal dependencies. Snapshot-based techniques partially address this issue but often encounter data sparsity and class imbalance, particularly in networks with transient interactions such as telecommunication call detail records (CDRs). Temporal Graph Networks (TGNs) model dynamic graphs by updating node embeddings over time; however, their predictive accuracy under sparse conditions remains limited. In this study, we improve the TGN framework by extracting enclosing subgraphs around candidate links, enabling the model to jointly learn structural and temporal information. Experiments on a sparse CDR dataset show that our approach increases average precision by 2.6% over standard TGNs, demonstrating the advantages of integrating local topology for robust link prediction in dynamic networks.
Open AccessA Hybrid TGN-SEAL Model for Dynamic Graph Link Prediction
Predicting links in sparse, continuously evolving networks is a central challenge in network science. Conventional heuristic methods and deep learning models, including Graph Neural Networks (GNNs), are typically designed for static graphs and thus struggle to capture temporal dependencies. Snapshot-based techniques partially address this issue but often encounter data sparsity and class imbalance, particularly in networks with transient interactions such as telecommunication call detail records (CDRs). Temporal Graph Networks (TGNs) model dynamic graphs by updating node embeddings over time; however, their predictive accuracy under sparse conditions remains limited. In this study, we improve the TGN framework by extracting enclosing subgraphs around candidate links, enabling the model to jointly learn structural and temporal information. Experiments on a sparse CDR dataset show that our approach increases average precision by 2.6% over standard TGNs, demonstrating the advantages of integrating local topology for robust link prediction in dynamic networks.
- Open Access
MasalBench: A Benchmark for Contextual and Cross-Cultural Understanding of Persian Proverbs in LLMs
In recent years, multilingual Large Language Models (LLMs) have become an inseparable part of daily life, making it crucial for them to master the rules of conversational language in order to communicate effectively with users. While previous work has evaluated LLMs' understanding of figurative language in high-resource languages, their performance in low-resource languages remains underexplored. In this paper, we introduce MasalBench, a comprehensive benchmark for assessing LLMs' contextual and cross-cultural understanding of Persian proverbs, which are a key component of conversation in this low-resource language. We evaluate eight state-of-the-art LLMs on MasalBench and find that they perform well in identifying Persian proverbs in context, achieving accuracies above 0.90. However, their performance drops considerably when tasked with identifying equivalent English proverbs, with the best model achieving 0.79 accuracy. Our findings highlight the limitations of current LLMs in cultural knowledge and analogical reasoning, and they provide a framework for assessing cross-cultural understanding in other low-resource languages. MasalBench is available at https://github.com/kalhorghazal/MasalBench.
Open AccessMasalBench: A Benchmark for Contextual and Cross-Cultural Understanding of Persian Proverbs in LLMs
In recent years, multilingual Large Language Models (LLMs) have become an inseparable part of daily life, making it crucial for them to master the rules of conversational language in order to communicate effectively with users. While previous work has evaluated LLMs' understanding of figurative language in high-resource languages, their performance in low-resource languages remains underexplored. In this paper, we introduce MasalBench, a comprehensive benchmark for assessing LLMs' contextual and cross-cultural understanding of Persian proverbs, which are a key component of conversation in this low-resource language. We evaluate eight state-of-the-art LLMs on MasalBench and find that they perform well in identifying Persian proverbs in context, achieving accuracies above 0.90. However, their performance drops considerably when tasked with identifying equivalent English proverbs, with the best model achieving 0.79 accuracy. Our findings highlight the limitations of current LLMs in cultural knowledge and analogical reasoning, and they provide a framework for assessing cross-cultural understanding in other low-resource languages. MasalBench is available at https://github.com/kalhorghazal/MasalBench.
- Open Access
Can LLMs Faithfully Explain Themselves in Low-Resource Languages? A Case Study on Emotion Detection in Persian
Large language models (LLMs) are increasingly used to generate self-explanations alongside their predictions, a practice that raises concerns about the faithfulness of these explanations, especially in low-resource languages. This study evaluates the faithfulness of LLM-generated explanations in the context of emotion classification in Persian, a low-resource language, by comparing the influential words identified by the model against those identified by human annotators. We assess faithfulness using confidence scores derived from token-level log-probabilities. Two prompting strategies, differing in the order of explanation and prediction (Predict-then-Explain and Explain-then-Predict), are tested for their impact on explanation faithfulness. Our results reveal that while LLMs achieve strong classification performance, their generated explanations often diverge from faithful reasoning, showing greater agreement with each other than with human judgments. These results highlight the limitations of current explanation methods and metrics, emphasizing the need for more robust approaches to ensure LLM reliability in multilingual and low-resource contexts.
Open AccessCan LLMs Faithfully Explain Themselves in Low-Resource Languages? A Case Study on Emotion Detection in Persian
Large language models (LLMs) are increasingly used to generate self-explanations alongside their predictions, a practice that raises concerns about the faithfulness of these explanations, especially in low-resource languages. This study evaluates the faithfulness of LLM-generated explanations in the context of emotion classification in Persian, a low-resource language, by comparing the influential words identified by the model against those identified by human annotators. We assess faithfulness using confidence scores derived from token-level log-probabilities. Two prompting strategies, differing in the order of explanation and prediction (Predict-then-Explain and Explain-then-Predict), are tested for their impact on explanation faithfulness. Our results reveal that while LLMs achieve strong classification performance, their generated explanations often diverge from faithful reasoning, showing greater agreement with each other than with human judgments. These results highlight the limitations of current explanation methods and metrics, emphasizing the need for more robust approaches to ensure LLM reliability in multilingual and low-resource contexts.
- Open Access
- 5 November 2025
- Published by MDPI AG in Algorithms
- Vol. 18 (11) , 706
Product Recommendation with Price Personalization According to Customer’s Willingness to Pay Using Deep Reinforcement Learning
Integrating recommendation systems with dynamic pricing strategies is essential for enhancing product sales and optimizing revenue in modern business. This study proposes a novel product recommendation model that uses Reinforcement Learning to tailor pricing strategies to customer purchase intentions. While traditional recommendation systems focus on identifying products customers prefer, they often neglect the critical factor of pricing. To improve effectiveness and increase conversion, it is crucial to personalize product prices according to the customer’s willingness to pay (WTP). Businesses often use fixed-budget promotions to boost sales, emphasizing the importance of strategic pricing. Designing intelligent promotions requires recommending products aligned with customer preferences and setting prices reflecting their WTP, thus increasing the likelihood of purchase. This research advances existing recommendation systems by integrating dynamic pricing into the system’s output, offering a significant innovation in business practice. However, this integration introduces technical complexities, which are addressed through a Markov Decision Process (MDP) framework and solved using Reinforcement Learning. Empirical evaluation using the Dunnhumby dataset shows promising results. Due to the lack of direct comparisons between combined product recommendation and pricing models, the outputs were simplified into two categories: purchase and non-purchase. This approach revealed significant improvements over comparable methods, demonstrating the model’s efficacy.
- 5 November 2025
- Published by MDPI AG in Algorithms
- Vol. 18 (11) , 706
Open AccessProduct Recommendation with Price Personalization According to Customer’s Willingness to Pay Using Deep Reinforcement Learning
Integrating recommendation systems with dynamic pricing strategies is essential for enhancing product sales and optimizing revenue in modern business. This study proposes a novel product recommendation model that uses Reinforcement Learning to tailor pricing strategies to customer purchase intentions. While traditional recommendation systems focus on identifying products customers prefer, they often neglect the critical factor of pricing. To improve effectiveness and increase conversion, it is crucial to personalize product prices according to the customer’s willingness to pay (WTP). Businesses often use fixed-budget promotions to boost sales, emphasizing the importance of strategic pricing. Designing intelligent promotions requires recommending products aligned with customer preferences and setting prices reflecting their WTP, thus increasing the likelihood of purchase. This research advances existing recommendation systems by integrating dynamic pricing into the system’s output, offering a significant innovation in business practice. However, this integration introduces technical complexities, which are addressed through a Markov Decision Process (MDP) framework and solved using Reinforcement Learning. Empirical evaluation using the Dunnhumby dataset shows promising results. Due to the lack of direct comparisons between combined product recommendation and pricing models, the outputs were simplified into two categories: purchase and non-purchase. This approach revealed significant improvements over comparable methods, demonstrating the model’s efficacy.
- Open Access
Probing Gender Bias in Multilingual LLMs: A Case Study of Stereotypes in Persian
Multilingual Large Language Models (LLMs) are increasingly used worldwide, making it essential to ensure they are free from gender bias to prevent representational harm. While prior studies have examined such biases in high-resource languages, low-resource languages remain understudied. In this paper, we propose a template-based probing methodology, validated against real-world data, to uncover gender stereotypes in LLMs. As part of this framework, we introduce the Domain-Specific Gender Skew Index (DS-GSI), a metric that quantifies deviations from gender parity. We evaluate four prominent models, GPT-4o mini, DeepSeek R1, Gemini 2.0 Flash, and Qwen QwQ 32B, across four semantic domains, focusing on Persian, a low-resource language with distinct linguistic features. Our results show that all models exhibit gender stereotypes, with greater disparities in Persian than in English across all domains. Among these, sports reflect the most rigid gender biases. This study underscores the need for inclusive NLP practices and provides a framework for assessing bias in other low-resource languages.
Open AccessProbing Gender Bias in Multilingual LLMs: A Case Study of Stereotypes in Persian
Multilingual Large Language Models (LLMs) are increasingly used worldwide, making it essential to ensure they are free from gender bias to prevent representational harm. While prior studies have examined such biases in high-resource languages, low-resource languages remain understudied. In this paper, we propose a template-based probing methodology, validated against real-world data, to uncover gender stereotypes in LLMs. As part of this framework, we introduce the Domain-Specific Gender Skew Index (DS-GSI), a metric that quantifies deviations from gender parity. We evaluate four prominent models, GPT-4o mini, DeepSeek R1, Gemini 2.0 Flash, and Qwen QwQ 32B, across four semantic domains, focusing on Persian, a low-resource language with distinct linguistic features. Our results show that all models exhibit gender stereotypes, with greater disparities in Persian than in English across all domains. Among these, sports reflect the most rigid gender biases. This study underscores the need for inclusive NLP practices and provides a framework for assessing bias in other low-resource languages.
- 18 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 1863-1873
A GNN-Based Autopilot Recommendation Strategy to Mitigate Payment Channel Imbalance Problem in Bitcoin Lightning Network
The Bitcoin Lightning Network, as a second-layer solution for enhancing the scalability of Bitcoin transactions, facilitates transactions through payment channels between nodes. However, the rapid growth of the network and rising transaction volumes have exacerbated the challenge of managing payment channel imbalances. Payment channel imbalance, characterized by the concentration of liquidity in one direction, leads to a decrease in payment success rates, a reduction in the effective lifespan of payment channels, and a decline in the network’s overall efficiency and throughput. This study introduces a graph neural network-based recommendation strategy designed to enhance the Lightning Network’s autopilot system. The proposed approach proactively mitigates channel imbalances by optimizing channel recommendations, enabling dynamic and scalable liquidity management for network users. Simulations conducted using the CLoTH tool demonstrate a 45% increase in payment success rates, a 46% reduction in imbalanced channels, and a 14% increase in the lifespan of payment channels across the network compared to the existing autopilot recommendation strategies, and when compared with the commonly adopted circular rebalancing method, the proposed strategy achieves a 27% improvement in payment success rates. Additionally, we offer a comparative topological analysis between two snapshots of the LN, taken in November 2021 and August 2023, to facilitate unsupervised learning tasks. The results highlight an increase in network centralization alongside a decrease in the number of network size, emphasizing the growing need for decentralization strategies in the LN, such as the approach proposed in this study.
- 18 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 1863-1873
A GNN-Based Autopilot Recommendation Strategy to Mitigate Payment Channel Imbalance Problem in Bitcoin Lightning Network
The Bitcoin Lightning Network, as a second-layer solution for enhancing the scalability of Bitcoin transactions, facilitates transactions through payment channels between nodes. However, the rapid growth of the network and rising transaction volumes have exacerbated the challenge of managing payment channel imbalances. Payment channel imbalance, characterized by the concentration of liquidity in one direction, leads to a decrease in payment success rates, a reduction in the effective lifespan of payment channels, and a decline in the network’s overall efficiency and throughput. This study introduces a graph neural network-based recommendation strategy designed to enhance the Lightning Network’s autopilot system. The proposed approach proactively mitigates channel imbalances by optimizing channel recommendations, enabling dynamic and scalable liquidity management for network users. Simulations conducted using the CLoTH tool demonstrate a 45% increase in payment success rates, a 46% reduction in imbalanced channels, and a 14% increase in the lifespan of payment channels across the network compared to the existing autopilot recommendation strategies, and when compared with the commonly adopted circular rebalancing method, the proposed strategy achieves a 27% improvement in payment success rates. Additionally, we offer a comparative topological analysis between two snapshots of the LN, taken in November 2021 and August 2023, to facilitate unsupervised learning tasks. The results highlight an increase in network centralization alongside a decrease in the number of network size, emphasizing the growing need for decentralization strategies in the LN, such as the approach proposed in this study.
- Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 19-27
Probing Gender Bias in Multilingual LLMs: A Case Study of Stereotypes in Persian
Ghazal Kalhor, Behnam Bahrak. Proceedings of the 9th Widening NLP Workshop. 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 19-27
Open AccessProbing Gender Bias in Multilingual LLMs: A Case Study of Stereotypes in Persian
Ghazal Kalhor, Behnam Bahrak. Proceedings of the 9th Widening NLP Workshop. 2025.