The integration of knowledge graphs (KGs) with large language models (LLMs) offers significant potential to enhance the retrieval stage in retrieval-augmented generation (RAG) systems.In this study, we propose KG-CQR 1 , a novel framework for Contextual Query Retrieval (CQR) that enhances the retrieval phase by enriching complex input queries with contextual representations derived from a corpuscentric KG.Unlike existing methods that primarily address corpus-level context loss, KG-CQR focuses on query enrichment through structured relation representations, extracting and completing relevant KG subgraphs to generate semantically rich query contexts.Comprising subgraph extraction, completion, and contextual generation modules, KG-CQR operates as a model-agnostic pipeline, ensuring scalability across LLMs of varying sizes without additional training.Experimental results on the RAGBench and MultiHop-RAG datasets demonstrate that KG-CQR outperforms strong baselines, achieving improvements of up to 4-6% in mAP and approximately 2-3% in [email protected], evaluations on challenging RAG tasks such as multi-hop question answering show that, by incorporating KG-CQR, the performance outperforms the existing baseline in terms of retrieval effectiveness.

Hoang Pham, Thanh-Do Nguyen, Khac-Hoai Nam Bui. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Task-oriented dialogue (TOD) systems facilitate goal-driven interactions between users and machines. While recent advances in deep learning have improved the performance, TOD systems often struggle in low-resource scenarios with limited labeled data. To address this challenge, we propose Spec-TOD, a novel framework designed to train an end-to-end TOD system with limited data. Spec-TOD introduces two main innovations: (i) a novel specialized end-to-end TOD framework that incorporates explicit task instructions for instruction-tuned large language models (LLMs), and (ii) an efficient training strategy that leverages lightweight, specialized LLMs to achieve strong performance with minimal supervision. Experiments on the MultiWOZ dataset, a widely used TOD benchmark, demonstrate that Spec-TOD achieves competitive results while significantly reducing the need for labeled data. These findings highlight the potential of the proposed framework in advancing efficient and effective TOD systems in low-resource settings.

Task-oriented dialogue (TOD) systems are introduced to solve specific tasks, which focus on training multiple tasks such as language understanding, tracking states, and generating appropriate responses to help users achieve their specific goals. Currently, one of the remaining challenges in this emergent research field is the capability to produce more robust architectures fine-tuned for end-to-end TOD systems. In this study, we consider this issue by exploiting the ability of pre-trained models to provide synthesis responses, which are then used as the input for the fine-tuned process. The main idea is to overcome the gap between the training process and inference process during fine-tuning end-to-end TOD systems. The experiment on Multiwoz datasets shows the effectiveness of our model compared with strong baselines in this research field. The source code is available for further exploitation.

Heterogeneous Graph Neural Networks (HeterGNN) have been recently introduced as an emergent approach for extracting document summarization (EDS) by exploiting the cross-relations between words and sentences. However, applying HeterGNN for long documents is still an open research issue. One of the main majors is the lacking of inter-sentence connections. In this regard, this paper exploits how to apply HeterGNN for long documents by building a graph on sentence-level nodes (homogeneous graph) and combine with HeterGNN for capturing the semantic information in terms of both inter and intra-sentence connections. Experiments on two benchmark datasets of long documents such as PubMed and ArXiv show that our method is able to achieve state-of-the-art results in this research field.

Graph Neural Network (GNN)-based models have proven effective in various Natural Language Processing (NLP) tasks in recent years. Specifically, in the case of the Extractive Document Summarization (EDS) task, modeling documents under graph structure is able to analyze the complex relations between semantic units (e.g., word-to-word, word-to-sentence, sentence-to-sentence) and enrich sentence representations via valuable information from their neighbors. However, long-form document summarization using graph-based methods is still an open research issue. The main challenge is to represent long documents in a graph structure in an effective way. In this regard, this paper proposes a new heterogeneous graph neural network (HeterGNN) model to improve the performance of long document summarization (HeterGraphLongSum). Specifically, the main idea is to add the passage nodes into the heterogeneous graph structure of word and sentence nodes for enriching the final representation of sentences. In this regard, HeterGraphLongSum is designed with three types of semantic units such as word, sentence, and passage. Experiments on two benchmark datasets for long documents such as Pubmed and Arxiv indicate promising results of the proposed model for the extractive long document summarization problem. Especially, HeterGraphLongSum is able to achieve state-of-the-art performance without relying on any pre-trained language models (e.g., BERT). The source code is available for further exploitation on the Github.

Summary Transfer markets in football have attracted the interest of researchers in economy and management. In this paper, we propose a high level analysis approach for classifying player valuation based on their performance during recent seasons. In particular, several data analysis techniques such as regression analysis, feature selection, and cluster analysis are presented for classifying players in term of performances and transfer fee. Specifically, by collecting and analyzing data from Wholescored, the largest detailed football statistics website, we have defined players into four groups, which include (1) Low performance and low transfer fee (LPLF), (2) Low performance and high transfer fee (LPHF), (3) high performance and high transfer fee (HPHF), and (4) high performance and low transfer fee (HPLF). The results in the implementation section show that, with the differences positions, there are different required skills that affect to the performance of players. We expect that this study can contribute to the management of Football Teams in terms of integrating these analyses into their management strategy.

Auto machine learning recently has been introduced as a trending technique for learning applications, including smart transportation. In this study, we focus on applying auto-machine learning for hyperparameter tuning to learn traffic datasets at the main regions of highway systems. Particularly, deep learning models have been recently introduced as emergent methods for traffic prediction. However, training deep learning models requires expensive works (e.g., time-consuming and human expertise), especially in terms of determining the configurations of hyperparameters in the models. In this regard, this paper introduces an automated framework for hyperparameter tuning to learn traffic datasets at an ecosystem in terms of reducing time-consuming tasks. Specifically, we first propose the HyperNet framework, using advanced data science techniques (e.g., Bayesian optimization and meta-learning) for the automated hyperparameter search process. Then, a deep learning model with the long short term memory network based on the HyperNet framework has presented for learning the temporal variation of traffic datasets at main regions of highway traffic systems. Regarding the experiment, we take data from the Korean highway system into account as a case study to evaluate the proposed approach. The evaluation indicates promising results of the proposed framework for learning multiple datasets of the traffic highway systems.

Abstract In this study, we propose a computational intelligence model for the Internet of Things applications by applying the concept of swarm intelligence (SI) into connected devices. Particularly, decentralized management of smart home energy management system (HEMS) is taken into account in which connected appliances, by sharing information with each other, make the individual decisions for optimizing electricity prices of smart HEMS. Specifically, the study includes two main issues: (a) We propose a framework for decentralized management in smart HEMS; and (b) artificial bee colony (ABC) algorithm, a typical algorithm of SI techniques, has been applied for connected appliances in terms of communication and collaboration with each other to optimize the performance of the energy management system. Moreover, regarding the implementation, we develop and simulate a connected environment of smart home systems to evaluate the proposed approach. The simulation indicates the promising results in terms of optimizing the load balancing problem comparing with the conventional approach of the decentralized management system in smart home applications.

In this paper, we proposed an approach to clustering based on bio-inspired behaviour and distributed energy efficient model. The motivation to propose this clustering approach is due to the challenge of performance in terms of finding an efficient way to send data packets to base stations and to maintain the lifetime performance of wireless sensor networks. The bioinspired approach adopted the behaviour of a bird called Kestrel. This behaviour is expressed using mathematical formulation and then translated into an algorithm. The bio-inspired algorithm is combined with the distributed energy efficient model for clustering to ensure efficient energy optimization. The proposed clustering approach, referred to as DEEC-KSA, is evaluated through simulation and compared with benchmarked clustering algorithms. The result of simulation showed that the performance of DEEC-KSA is efficient among the comparative clustering algorithms for energy optimization in terms of stability period, network lifetime and network throughput. Additionally, the proposed DEEC-KSA has the optimal time (in seconds) to send packets to base station successfully.

With the rapid development of the internet of things (IoT), connected vehicles are set to become a huge industry over the next few years. In this study, we take an investigation of the distributed intelligent traffic system by pushing intelligence into connected vehicles in terms of dynamic decision making for traversing a certain area (e.g., roundabout and intersection). In particular, we propose a model for the next generation of intelligent transportation system, which focuses on dynamic decision making of connected vehicles based on Ant Colony Optimization, a typical Swarm Intelligence (SI)-based algorithm. Specifically, we first present a communication framework among connected vehicles for sharing information of traffic flow. Then, by applying the concept of SI, connected vehicles are regarded as artificial ants which are able to self-calculate to make an adaptive decision following the dynamics of traffic flow. Furthermore, for evaluating the effectiveness of the proposed approach, we have constructed a framework to model and simulate the traffic system in IoT environment. Simulations with different scenarios of transportation systems indicate promising results comparing with previous works.

Small cells are low-powered cellular radio access nodes which make best use of available spectrum by reusing the same frequencies many times within a geographical area. However, the deployment of small cells (e.g., femtocells) may introduce extra interferences such as cross-tier(macrocell-femtocell) and co-tier (femtocell-femtocell) interferences. In this regard, an effective interference management mechanism is required for improving the performance of the network. In this study, a distributed resource allocation is introduced to deal with the interference problem in two-tier network. Specifically, we first apply FFR scheme to mitigate interference for cross-tier problem. To reduce co-tier interference, a cooperative game model is proposed where each femtocells are regard as players of the game. Simulation results reveal that the proposed approach significantly improves capacity of the networks.

Summary Smart traffic light control at intersections is 1 of the major issues in Intelligent Transportation System. In this paper, on the basis of the new emerging technologies of Internet of Things, we introduce a new approach for smart traffic light control at intersection. In particular, we firstly propose a connected intersection system where every objects such as vehicles, sensors, and traffic lights will be connected and sharing information to one another. By this way, the controller is able to collect effectively and mobility traffic flow at intersection in real‐time. Secondly, we propose the optimization algorithms for traffic lights by applying algorithmic game theory. Specially, 2 game models (which are Cournot Model and Stackelberg Model) are proposed to deal with difference scenarios of traffic flow. In this regard, based on the density of vehicles, controller will make real‐time decisions for the time durations of traffic lights to optimize traffic flow. To evaluate our approach, we have used Netlogo simulator, an agent‐based modeling environment for designing and implementing a simple working traffic. The simulation results shows that our approach achieves potential performance with various situations of traffic flow.