Yueheng Mao, Min Yu, Gengwang Li, Jianguo Jiang, Gang Li, Meng Zhang, Zhen Xu, Weiqing Huang, Ming Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Pre-trained language models are increasingly being used in multi-document summarization tasks. However, these models need large-scale corpora for pre-training and are domain-dependent. Other non-neural unsupervised summarization approaches mostly rely on key sentence extraction, which can lead to information loss. To address these challenges, we propose a lightweight yet effective unsupervised approach called GLIMMER: a Graph and LexIcal features based unsupervised Multi-docuMEnt summaRization approach. It first constructs a sentence graph from the source documents, then automatically identifies semantic clusters by mining low-level features from raw texts, thereby improving intra-cluster correlation and the fluency of generated sentences. Finally, it summarizes clusters into natural sentences. Experiments conducted on Multi-News, Multi-XScience and DUC-2004 demonstrate that our approach outperforms existing unsupervised approaches. Furthermore, it surpasses state-of-the-art pre-trained multi-document summarization models (e.g. PEGASUS and PRIMERA) under zero-shot settings in terms of ROUGE scores. Additionally, human evaluations indicate that summaries generated by GLIMMER achieve high readability and informativeness scores. Our code is available at https://github.com/Oswald1997/GLIMMER.

Since the current cyberspace is becoming changeable and complex over times, the situation of cyber security is becoming increasingly severe, and one of the important issues is that there is still lack of a general applicability defense model for open and dynamic networks. Recent research suggests that the evolutionary game methods have the advantage of improving the defensive capabilities based on the internal decision and learning mechanisms. In this paper, by exploiting the advantage of collaborative decision-making in multi-agent system, we constructed the dynamic cyber defense problem into a decentralized multi-agent cooperative decision framework, whose core idea is the initiative decentralized interactions among defense agents. Then, we contributed a heuristic imprecise probabilistic based interaction decision algorithm, HIDS, that is, which utilizes the multidimensional semantic relevance among observation, tasks and agents, so that agents can continuously improve cognition and optimize decision-making by learning interactive records. In addition, we analyzed the equivalence and the transformation conditions between the proposed model and the existing decision models, and combined the evolutionary game with the nonlinear stochastic theory, then the evolution process of the defense policies are analyzed. Finally, the performance comparison of the proposed algorithm and the influence of different intensity random disturbances on the evolution process are analyzed.

The traffic explosion and the rising of diverse requirements lead to many challenges for the mobile networks on flexibility, scalability, and deployability. To meet the challenges, end-to-end network slicing and mobile edge computing (MEC) are introduced into beyond fifth-generation (B5G) mobile communication networks. However, the edge servers are limited by computing and data processing capabilities, how to implement efficient load balancing and resource management in data center are still challenges for B5G network slices. In this paper, we first propose a model of load balancing for B5G network slices and formulate a total cost minimization problem, considering the energy cost, processing delay, backhaul bandwidth cost, and revenue loss associated with the backhaul delay. Then, a distributed algorithm based on Proximal Jacobian Alternating Direction Method of Multipliers (PJ-ADMM) is proposed to solve the formulated problem. Simulations are conducted to show that the proposed algorithm are effective in terms of reducing network operating cost.

Recent advances in the Internet-of-Things has given rise to the possibility of connecting different kinds of devices to the internet. With low computing resources constraining most IoT devices and their networks, managing the capabilities and services of the various constrained devices over the internet has been an issue for IoT implementations. Existing approaches to this problem lack end-to-end resource-oriented Device Management architecture for resource constrained devices and their networks as a result of protocol fragmentation in the domain. This study first proposes a Device Management architecture which is able to leverage more than one protocol in addressing this problem. We then base on the architecture to put forward an implementation approach to manifesting this solution through our usage of a simulated Smart Building case study. The analysis of the case study shows that the proposed resource-oriented architecture achieves device management and service enablement operations with good performance.

Information sharing is critical to multi-agent team for cooperative decision making in dynamic and partially observable environments. Other than building a full information coverage, if agents in a team can be self-directed to valuate a potential receiver and where to communicate, the coordination efficiency will be greatly enhanced. Although intensive studies of information valuation approaches have been developed in ontology graph matching and natural language processing, these models fail to perform fast reasoning for large-scale decentralized agents dynamic coordination. In this paper, we propose a fast information valuation approach based on a complex network graph model, which helps to indicate the information importance. Similar to vague information valuation by human, the key is that important information always significantly changes their complex information graph with its incorporation. Therefore, we calculate the semantic based value of this new information in a graph model and build a local graph evaluation algorithm to estimate information graph evolution, instead of performing expensive complete graph search. The advantage is that the local valuation algorithm can be easily transformed into efficient queries in agents' information base so that they can make fast decisions. Although the decision may not be precise, similar to human communication, the information sharing performance is good enough to disseminate valuable information in the multi-agent team. We demonstrate the feasibility of the proposed information valuation approach in a multi-agent cooperation case study.

Process management is practical in the state-of-the-art of Internet of things research. However, this has become a bottleneck in recent years since an extreme amount of heterogeneous items have to be recorded and traced with radio frequency identification (RFID) tags. In a typical application of process synthesis management, each item will be involved with multiple processes but when those processes are interconnected, an extremely complex network emerges and has to be managed. Existing works on processing management systems, however, are always case-based and only focus on specific application domains. Thus, the general applied processing management model is rather limited. In this paper, we summarize the characteristics of the RFID application domains, and by abstraction, we propose an innovation of the RFID processing model. In this model, each RFID data were deemed as an operation record and all their processing services are logically interconnected and organized as a procedure graph. In addition, we abstract and summarize the basic RFID data processes into a few types. The advantage of this design is that the basic RFID data processing logic can be preprogrammed and in a real domain, the system can be dynamically programmed by automatically constructing the procedure graph nodes with those basic processes and mapping the interconnection logic according to the topology of the graph. As the last part of this paper, we designed a prototype system for medical instruments infection control to demonstrate our approach.

As intelligence in smart homes increasingly get sophisticated, it has become necessary adopting ontological reasoning in these domains. However, ontologies presently lack a standardised representation for uncertainty in knowledge. A key challenge therefore lies in developing ontology-based decision-making models that can integrate domain uncertainty. In this paper, a decision network extension to OWL ontology using only a subset of domain concepts relevant for probabilistic modelling is proposed. This relevant set of concepts in ontology can be generated on the fly using an algorithm, OWLMB, introduced in this paper. Given an ontology and a class as inputs to OWLMB, Markov boundary of the class is returned as this minimum relevant set. Also, representations for decision and utility nodes are proposed as an extension of valid decision situation to ontology. Validation of this approach in a smart home scenario shows its feasibility in a real application domain.

With the expansion of decentralized multi-robot system's applications, limited communication resources has become a critical factor toward the efficiency of multi-robot coordination when both system sizes scale up and their tasks become more complex. In current researches, it is only possible for every robot in a decentralized setting either to choose a part of communication channels to communicate or interact with the other robots. However, due to the constraint on robots' limited communication capability as well as their local observations of the whole network, each robot's cooperative communication channel choice, based on the partial observation of the states of the system, and communication network is under uncertainty. It is also a major challenge to work out multi-robots' cooperative decision by uncertainty reasoning with only a partial observation to the environment. In this paper, we propose a decentralized approach that allows robots to cooperatively search channels and independently choose channels. The key of this approach is to build an up-to-date observation for each robot's view so that a local decision model is achievable. Based on this model, we simplify the Dec-POMDP problem model and each robot can jointly work out its communication policy in order to improve its local decision utilities for the choice of the communication channel using its current observation of the limited system's states as well as its neighbors' choices. In this way, an approximate balance between timeliness of decision making and cooperatively enhancing the efficiency of using these channels can be reached.