ABSTRACT Customer service primarily involves interaction with clients through phone calls. Precise keyword extraction from customer complaint texts facilitates the implementation of intelligent task assignment and efficient response systems. However, existing keyword extraction technologies perform sub‐optimally in the customer service domain of telecommunications operators and require substantial manual word segmentation. Given the pronounced clustering of customer service data, this research introduces a synonym matching approach and a few‐shot learning‐based method tailored for extracting keywords in this sector. This enables model training with minimal labelled data and computational resources. Using a dataset generated from the transcription of customer service calls, the proposed model demonstrates a 24.94% improvement in accuracy compared to popular existing methods.

Accurate flood prediction is essential for disaster mitigation, life protection, and minimizing community and infrastructure impact. However, current flood prediction models often struggle with challenges such as capturing intricate spatiotemporal dynamics, adapting to changing environmental conditions, and integrating diverse variables effectively. In this paper, we propose a novel approach for flood forecasting, say a Heterogeneous Dynamic Temporal Graph Convolutional Network (HD-TGCN). The Dynamic Temporal Graph Convolution Module (D-TGCM) adapts to evolving temporal graph structures, utilizing a Multi-Head Self-Attention mechanism to generate adjacency matrices dynamically. This enhances the adaptability of the model to changing temporal graph structures and captures intricate dependencies among features at different time steps. Additionally, HD-TGCN incorporates parallel DTGCMs to address the heterogeneity in flood spatiotemporal data. This module effectively captures intricate relationships among diverse variable types, improving the model's ability to handle complex multivariable interactions and enabling the model to process heterogeneous graphs effectively. Experiments on real-world datasets demonstrate that our model outperforms state-of-the-art models. At a prediction horizon of 30 minutes, the model exhibits improvements of 78.02%, 66.30%, and 0.15% in terms of MAE, RMSE, and NSE, respectively.

Federated learning (FL) is an emerging paradigm for privacy-preserving machine learning, in which multiple clients collaborate to generate a global model through training individual models with local data. However, FL is vulnerable to model poisoning attacks (MPAs) as malicious clients are able to destroy the global model by modifying local models. Although numerous model poisoning defense methods are extensively studied, they remain vulnerable to newly proposed optimized MPAs and are constrained by the necessity to presume a certain proportion of malicious clients. To this end, in this paper, we propose MODEL, a model poisoning defense framework for FL through truth discovery (TD). A distinctive aspect of MODEL is its ability to effectively prevent both optimized and byzantine MPAs. Furthermore, it requires no presupposed threshold for different settings of malicious clients (e.g., less than 33% or no more than 50%). Specifically, a TD-based metric and a clustering-based filtering mechanism are proposed to evaluate local models and avoid presupposing a threshold. Furthermore, MODEL is effective for non-independent and identically distributed (non-IID) training data. In addition, inspired by game theory, we incorporate a truthful and fair incentive mechanism in MODEL to encourage active client participation while mitigating the potential desire for attacks from malicious clients. Extensively comparative experiments demonstrate that MODEL effectively safeguards against optimized MPAs and outperforms the state-of-the-art.

Abstract Automatic speech recognition (ASR) has been widely used in the field of customer service, but the performance of general ASR in dialect transcription is not satisfactory, especially in Guangdong Province. Targeted training of ASR transcription engine will produce effect, but the training cost is high, and it is not suitable for small‐scale training with multiple dialects and frequencies. The complaint problems in the customer service field have obvious clustering and are suitable for few‐shot and multi‐frequency training. In view of this, in the actual engineering application, the method of ASR transcribed into the dialect error correction thesaurus is tried to be used to replace the wrong words, and have achieved good results. The optimization technology after automatic speech transcription proposed in this study can improve the recognition accuracy of general ASR by 13.75% for dialect words.

Graph-based Aspect-based Sentiment Classification (ABSC) approaches have yielded stateof-the-art results, expecially when equipped with contextual word embedding from pretraining language models (PLMs).However, they ignore sequential features of the context and have not yet made the best of PLMs.In this paper, we propose a novel model, BERT4GCN, which integrates the grammatical sequential features from the PLM of BERT, and the syntactic knowledge from dependency graphs.BERT4GCN utilizes outputs from intermediate layers of BERT and positional information between words to augment GCN (Graph Convolutional Network) to better encode the dependency graphs for the downstream classification.Experimental results demonstrate that the proposed BERT4GCN outperforms all state-of-the-art baselines, justifying that augmenting GCN with the grammatical features from intermediate layers of BERT can significantly empower ABSC models.