Tidal mobile traffic patterns offer opportunities for efficient Cloud Radio Access Network (CRAN) scheduling by leveraging its disaggregated and virtualized baseband processing, where baseband functions form a virtualized network function service chain (VNF-SC, or RAN slice) deployed across metro access, aggregation, and core networks. By dynamically reconfiguring and migrating RAN slices, processing pools can be powered down during low-demand periods to save energy. However, RAN slice migration causes service disruptions and degrades Quality of Service (QoS), making the tradeoff between energy efficiency and QoS a key challenge in CRAN scheduling. Existing approaches, such as heuristic and Deep Reinforcement Learning (DRL)-based methods, have achieved certain optimizations but rely on fixed scheduling intervals, which require provisioning for peak demand within the interval, leading to resource overprovisioning and inefficiency. To enable flexible and adaptive scheduling, we propose threshold-triggered heuristic-assisted DRL (TT-HA-DRL), which employs a threshold-triggered mechanism based on varying service demands and a heuristic-assisted DRL framework for adaptive RAN slice migration. Heuristic algorithms are used for action pruning to optimize the action space, enhancing scheduling performance. Baseline heuristics are incorporated to construct the Normalized Performance Loss as the reward function, enabling a tradeoff among the multiple optimization objectives. Extensive simulations validate the effectiveness and scalability of the proposed TT-HA-DRL. Compared to fixed-interval HA-DRL, our approach achieves reductions of up to 10.8% in power consumption, 12.3% in migration time, and 23.9% in Maximum Frequency Slot Index (MFSI) in a 30-node network. These results confirm TT-HA-DRL's ability for elastic and QoS-guaranteed RAN slice scheduling.

Document-level relation extraction aims to extract entity relations that span across multiple sentences. This task faces two critical issues: long dependency and mention selection. Prior works address the above problems from the textual perspective, however, it is hard to handle these problems solely based on text information. In this paper, we leverage video information to provide additional evidence for understanding long dependencies and offer a wider perspective for identifying relevant mentions, thus giving rise to a new task named Multimodal Document-level Relation Extraction (MDocRE). To tackle this new task, we construct a human-annotated dataset including documents and relevant videos, which, to the best of our knowledge, is the first document-level relation extraction dataset equipped with video clips. We also propose a hierarchical framework to learn interactions between different dependency levels and a textual-guided transformer architecture that incorporates both textual and video modalities. In addition, we utilize a mention gate module to address the mention-selection problem in both modalities. Experiments on our proposed dataset show that 1) incorporating video information greatly improves model performance; 2) our hierarchical framework has state-of-the-art results compared with both unimodal and multimodal baselines; 3) through collaborating with video information, our model better solves the long-dependency and mention-selection problems.

Relation extraction is a critical task in the field of natural language processing with numerous real-world applications. Existing research primarily focuses on monolingual relation extraction or cross-lingual enhancement for relation extraction. Yet, there remains a significant gap in understanding relation extraction in the mix-lingual (or code-switching) scenario, where individuals intermix contents from different languages within sentences, generating mix-lingual content. Due to the lack of a dedicated dataset, the effectiveness of existing relation extraction models in such a scenario is largely unexplored. To address this issue, we introduce a novel task of considering relation extraction in the mix-lingual scenario called MixRE and constructing the human-annotated dataset MixRED to support this task. In addition to constructing the MixRED dataset, we evaluate both state-of-the-art supervised models and large language models (LLMs) on MixRED, revealing their respective advantages and limitations in the mix-lingual scenario. Furthermore, we delve into factors influencing model performance within the MixRE task and uncover promising directions for enhancing the performance of both supervised models and LLMs in this novel task.

With the advancement of network function virtualization, 5G RAN slice’s baseband processing functions, such as distributed unit and centralized unit, can be implemented via virtual machines in processing pools (PPs). When traffic demand decreases, we can sleep the low-utilized PPs and migrate the slice requests they serve to other PPs for energy savings. However, migrations of RAN slice can cause service interruptions, leading to degraded Quality of Service (QoS). Existing works have addressed the energy-efficient and QoS-guaranteed RAN slice migrations problems effectively. However, their scheduling schemes are based on fixed-time intervals, which inevitably leads to the over-provisioning issue. To ensure service quality, fixed-time interval scheduling requires resource allocation based on the maximum demand within a time interval, leading to resource wastage. To address this issue, we propose an Adaptive Threshold-Triggered Heuristic-assisted Deep Reinforcement Learning (ATT-HADRL) algorithm that schedules RAN slice migrations in response to tidal traffic demands. Simulation results confirm that the proposed ATT-HA-DRL algorithm not only reduces power consumption and minimizes resource wastage but also decreases the number of scheduling events and shortens the total migration time, thereby maintaining high service quality and outperforming fixed-interval scheduling approaches.

Relation extraction (RE) is a vital task within natural language processing. Previous works predominantly focus on extracting relations from plain text. However, with the evolution of communication habits, many individuals employ symbolic representations, e.g. emoticons, to convey nuanced information. This shift in communication prompts a pertinent question: How do emoticons impact the performance of RE models? In response, we introduce a novel Emoticon-infused Relation Extraction (EmoRE) task and present the EmoRED dataset, the first human-annotated dataset specifically designed for relation extraction in text containing emoticons. EmoRED samples encompass various emoticon types, each serving a unique role—some aid in entity identification, while others facilitate relation comprehension. Alongside dataset construction, we conduct extensive experiments to scrutinize behaviors of both supervised models and large language models within emoticon-rich text. Experimental results reveal significant variations in the behavior of existing models in the EmoRE task, highlighting the need for future models that can consistently harness emoticon information.

To ensure reliable network services, the link protection method is widely employed for light-path provision. However, it inevitably increases propagation delay due to different transmission distances between active and backup light-paths, leading to a longer transport delay. Consequently, a crucial challenge is how to coordinate link protection and transport delay to maximize service availability while satisfying the delay requirements of each service. In this paper, we investigate the availability-aware and delay-sensitive (AADS) radio access network (RAN) slicing mapping problem with link protection in metro-access/aggregation elastic optical networks (EONs). We initially provide the mathematical model of availability and propagation delay for both unprotected and protected RAN slicing requests. Subsequently, we propose a mixed-integer linear programming (MILP) model and a deep reinforcement learning (DRL)-based algorithm to maximize the availability of RAN requests while satisfying the specified delay requirements of each slice. Finally, we analyze the availability under various 5G services (i.e., enhanced Mobile Broadband, ultra-Reliable Low-Latency Communication, and massive Machine Type Communication) from a delay perspective in both small-scale and large-scale networks. Simulation results demonstrate that our proposed DRL-based method can achieve up to a 14.1% increase in availability compared to the benchmarks.