Tianle Liang, Yifu Chen, Shengpeng Ji, Yijun Chen, Zhiyang Jia, Jingyu Lu, Fan Zhuo, Xueyi Pu, Yangzhuo Li, Zhou Zhao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Manufacturing industries play a pivotal role in global economies. It is a powerful driver of economic growth, creating a significant number of job opportunities and providing support for innovation and technological development. In the present world, where environmental concerns and resource conservation are paramount, reducing energy consumption while ensuring high processing quality is a significant challenge. During production, controlling the on/off of the machine reasonably can reduce energy consumption. This paper explores a method to analyze the performance of production lines containing buffers of finite capacity and machines that can be switched on and off when the specific conditions are met. We focus on finite production processes where the number of products is limited. In this paper, we establish a mathematical model for a three-machine production line and propose analytical methods for calculating production performance indicators. The analysis of the three-machine line employs the Markov method. Additionally, an aggregation method is introduced to extend the analysis to multi-machine systems, yielding promising results through numerical experiments. This research contributes to the advancement of production systems, improving a sustainable and competitive future for the manufacturing industry. Note to Practitioners—Small-batch production refers to the production of a small quantity of products, which can achieve customized order processing and improve resource utilization. Due to the dynamic nature of such processes, the traditional steady-state analysis method may not be applicable. This paper proposes an analytical method to predict the dynamic behavior of small batch production systems with Bernoulli machines that can be switch-on/off controlled and with finite capacity buffers. The algorithm developed in this paper can be used and support the production managers and engineers to predict the dynamic performance of the system with high accuracy. It can assist in decision-making in production control activities.

Multimodal sarcasm detection (MSD) requires predicting the sarcastic sentiment by understanding diverse modalities of data (e.g., text, image). Beyond the surface-level information conveyed in the post data, understanding the underlying deep-level knowledge-such as the background and intent behind the data-is crucial for understanding the sarcastic sentiment. However, previous works have often overlooked this aspect, limiting their potential to achieve superior performance. To tackle this challenge, we propose DeepMSD, a novel framework that generates supplemental deep-level knowledge to enhance the understanding of sarcastic content. Specifically, we first devise a Deep-level Knowledge Extraction Module that leverages large vision-language models to generate deep-level information behind the text-image pairs. Additionally, we devise a Cross-knowledge Graph Reasoning Module to model how humans use prior knowledge to identify sarcastic cues in multimodal posts. This module constructs cross-knowledge graphs that connect deep-level knowledge with surface-level knowledge. As such, it enables a more profound exploration of the cues underlying sarcasm. Experiments on the public MSD dataset demonstrate that our approach significantly surpasses previous state-of-the-art methods.

Customized production is becoming prevalent under smart manufacturing environment. With the increasing high standards from customers, products with satisfied quality are required. However, during some flexible machining processes, defective parts are produced occasionally. In the meanwhile, customized orders (commonly in small-lot) of parts having no defects are desired to be finished within lead times. Dynamic modeling and analysis are required since customized production cannot always reach steady state. Therefore, we investigate the problem in this article. Practical issues, such as unreliable machines and finite buffers are also taken into account. Specifically, a novel building block-based dynamic modeling and analysis method is proposed. The proposed method is validated to have a precise performance measures prediction effect, as well as an efficient computational performance.

A multiscale sequence information fusion (MSSIF) method is presented for dynamic facial expression recognition (DFER) in video sequences. It exploits multiscale information by integrating features from individual frames, subsequences, and entire sequences through a transformer-based architecture. This hierarchical feature fusion process includes deep feature extraction at the frame level to capture intricate visual details, intrasubsequence fusion using self-attention mechanisms for analyzing adjacent frames, and intersubsequence fusion to synthesize long-term emotional dynamics across time scales. The efficacy of MSSIF is demonstrated through extensive evaluation on three video datasets: eNTERFACE'05, BAUM-1s, and AFEW, where it achieves overall recognition accuracies of 60.1%, 60.7%, and 58.8%, respectively. These results substantiate MSSIF's superior performance in accurately recognizing facial expressions by managing short and long-term dependencies within video sequences, making it a potent tool for real-world applications requiring nuanced dynamic facial expression detection.

Mobile crowdsensing is a promising scheme for performing large-scale urban monitoring, but it always faces the issue of unstable spatiotemporal coverage, which results in the incompletion of data collection. The common solutions for tackling this issue, are to use the existing subset of measurements for inferring the remaining unsensed data by leveraging the latent data correlations. However, existing data inference techniques, both for matrix/tensor factorization based methods and deep learning based methods, cannot well capture the high-order and dynamic data correlations simultaneously under the mobile crowdsensing scheme. In this paper, we propose a novel 3Dimensional (3D) channel attention aided neural tensor completion method, called "3DA-NTC", for more accurate crowdsensing data inference, through leveraging both the multi-dimensional data structure mining ability of tensor factorization as well as the high-order, dynamic correlation learning ability of deep neural network. Specifically, to capture the spatiotemporal and multitype data correlations, we first use a 3D tensor to model the 3Order interaction among crowdsensing data. Then, we combine the traditional inner product based tensor factorization with outer product computing to enhance the modeling of nonlinear data correlations and form an interaction tensor, based on which, we apply a 3D channel attention aided convolutional neural network to further extract the features of high-order and dynamic data interactions for missing value inference. Extensive experiments on two real-world urban sensing datasets, including U-Air and SensorScope, are conducted to evaluate the performance of 3DANTC, and the results demonstrate the superiority of our method compared with the state-of-the-art (SOTA) baselines in missing data recovery.

To meet the growing need for customized products, flexible production systems are gaining widespread application in modern factories. The inherent complexity of these flexible systems poses new challenges for production management. In this paper, we intend to contribute an operation-level analytical model for such systems, thus providing an effective evaluation tool for productivity and energy consumption analysis. Specifically, a multi-type serial production line with machines obeying the geometric reliability model and intermediate buffers having limited capacity is considered, in which K batches of parts are processed sequentially in each shift. Analytical solutions for key operation metrics are first developed for small systems using a Markov model. For multi-machine lines, a semi-analytical method based on aggregation is proposed. The effectiveness of the proposed analysis method is verified through simulation experiments. Besides, system properties are analyzed to provide insights for continuous improvement and optimization.

Serial lines with multi-type products, finite buffers, and stochastic machine breakdowns are commonly seen in flexible production systems, where multiple small-size production batches with various part types are performed according to the customer orders. Such production systems rarely reach their steady states. Therefore, analyzing the transient behavior of these systems has an important research significance, especially for real-time operations and management. In this paper, we analyze the performance of the multi-type serial line under a given production schedule. Mathematical models and analytical solutions are developed for one- and two-machine lines. For longer lines, an efficient aggregation method is proposed to approximate the performance measures with high accuracy. In addition, based on performance analysis, the optimal scheduling of batches is discussed. Numerical experiments show that our proposed analysis method can be successfully applied to production scheduling problems. Note to Practitioners—Flexible serial lines with multi-type products are widely seen in practice, e.g., in automobile paint shops for vehicles with multiple types, and furniture assembly plants with various styles. In order to optimize such systems, achieving a real-time transient performance measure with high accuracy is of great importance. In this paper, mathematical models, as well as analytical solutions, are developed for transient performance analysis. A computationally efficient method is proposed to approximate the transient performance for longer lines. In addition, scheduling algorithms to lower the maximum batch completion time and total flow time are also discussed. These analysis methods and scheduling strategies can guide the operators and managers to improve production efficiency.

As one of the most commonly used clustering methods, the fuzzy C-means (FCM) clustering strategy extends the notion of hard clustering to associate each pattern with every cluster using a membership function. Although a lot of efforts have been made by the clustering community, it is still unclear how to evaluate the fuzziness of different versions of FCM. To fill this theoretical blank, by observing a family of objective functions, a definition of fuzzy degree is provided to quantify the fuzziness of different versions of FCM according to their dedicated objective functions. Then, a general fuzzy C-means (GFCM) clustering algorithm is proposed to solve the clustering problem of using FCM under different distance metrics and fuzzy degrees. From the perspective of fuzzy attribute, the using of this fuzzy degree can be used to reveal the essential difference among a group of FCM-based clustering algorithms, and the proposed GFCM achieves the aim of using objective functions to control the fuzziness of clustering results. Additionally, some properties and relations of these FCM-based clustering algorithms under different fuzzy degrees are discussed and the convergence and stability of the proposed GFCM are proved. Finally, extensive experiments have been performed to demonstrate that by comparison with the effect caused by distance metric, the choices of fuzzy degree have a more significant effect and improvement on the performances of a FCM-based clustering algorithm.

Steady-state analysis on production systems, especially serial lines with rigid production mode, have been carried out in a large amount of research literature. With the development of the advanced manufacturing technologies, as well as the national strategies, for example, Made in China 2025, real-time performance evaluation, prediction, production management, and efficient control of systems with flexible production mode are gradually becoming important issues. Under the assumption of assembly systems with small batch-based production tasks, we develop a mathematical model for the real-time performance analysis of such systems with the geometric machine reliability model, which is applicable in many manufacturing environments. The formulas for real-time performance analysis are derived first using the Markovian approach for three-machine assembly systems. Then, based on the idea of decomposition and aggregation, an algorithm with high accuracy and efficiency for system real-time performance approximation is proposed. Finally, the decomposition and aggregation-based real-time analysis approach is extended to the generalized assembly systems.Note to Practitioners—As the concept of manufacturing intelligence being proposed and promoted, system flexibility appears frequently in production in recent years. A flexible production refers to the implementation of small batch-based production with multitype of products on the same production line, by quick setup or machine adjustments. The steady-state analysis method that is traditionally used might be not applicable in this case since most of the production process is in transient state. An approximation-based analytical method, for real-time performance evaluation of assembly systems with small batch-based production tasks and with machines having the geometric reliability model, is proposed. The algorithm developed in this article can be utilized by production managers and engineers, with which the real-time performance of the systems can be predicted with high accuracy. It can also provide assistance for decision-making on activities of production control.

In manufacturing systems with large variety of products, small- to medium-size production runs are typically carried out according to the customers' orders. When the volume of a production run is small, the production system operates partially (or entirely) in the transient regime and the traditional steady-state analysis may become inapplicable. In this paper, under the framework of serial production lines with finite buffers and with machines having the Bernoulli reliability model, we study the problems of performance evaluation, system-theoretic properties, and bottleneck of such systems. Specifically, we study the mathematical model for the one- and two-machine cases and derive analytical formulas to evaluate the performance measures of such systems. Then, for longer lines, a computationally efficient algorithm based on aggregation is developed to approximate the system performance measures with high accuracy. In addition, the monotonicity and reversibility properties and completion time bottleneck are discussed. Finally, a case study in a lighting equipment assembly plant is described to illustrate the theoretical results obtained.

Serial lines and assembly systems are two of the most fundamental production system structures used in practical manufacturing environments. While serial lines have been studied extensively in production systems literature and practice, assembly systems have received much less attention. Among existing results on assembly systems, most are focused on steady state performance evaluation, system properties, and improvement. On the other hand, transient behavior of such systems remains largely unexplored. In the framework of assembly systems with Bernoulli machines and finite buffers, this paper develops a mathematical model for transient analysis and derives closed-form formulas to calculate the real-time production rate, consumption rates, work-in-process, and probabilities of machine starvation and blockage, during transients. Then, an improved computationally efficient algorithm for transient performance evaluation in Bernoulli serial lines is proposed. Based on the improved algorithm, an effective and efficient method for transient performance evaluation in Bernoulli assembly systems is derived. Finally, the method is extended to complex assembly system structures.