To prevent viruses and zombie programs from affecting the secure communication of oil multi-channel industrial control networks, an intrusion detection model for oil multi-channel industrial control network communication based on AFSA-SVM is studied. Collect the communication status information of the oil multi-channel industrial control network. The communication status features of the oil multi-channel industrial control network are extracted from the collected information by using the foraging, clustering, and tail chasing behaviors of the artificial fish swarm algorithm (AFSA). The weighted information gain method is used to reduce the extracted features, take the reduced features as the input, and use the support vector machine (SVM) to realize the intrusion detection of the oil multi-channel industrial control network communication. The experimental results show that the communication state features extracted from the model are representative; after feature reduction, the weighted error measure of intrusion detection is significantly reduced; under different intrusion intensities, the single intrusion and mixed intrusion detection effects of the model are excellent, and the intrusion destination IP address can be obtained.

To address the urgent need for rapid source localization of hazardous chemical leaks, this study proposes a multi-strategy enhanced sparrow search algorithm (ESSA). Three key innovations are implemented. First, an improved logistic mapping initialization method is developed to optimize the initial sparrow population generation, thereby enhancing the diversity of initial solutions. Second, a nonlinear control factor is introduced to balance global exploration and local exploitation during position updates. This mechanism ensures effective exploration of the solution space in later iterations. Third, an adaptive step-size adjustment mechanism enables efficient escape from local optima. Simulation results demonstrate ESSA’s capability to rapidly and accurately identify both the intensity and location of chemical leaks. The proposed method achieves localization accuracy within 1% relative error for both leak intensity and location, significantly outperforming other swarm intelligence algorithms (such as glowworm swarm optimization, grey wolf optimizer, genetic algorithm, etc.). Statistical analysis based on 30 independent runs confirms the algorithm’s robustness, with standard deviation coefficients below 0.021. Compared with the original sparrow search algorithm (SSA), ESSA achieves significant error reduction while maintaining comparable computational efficiency. This study provides an effective method for improving the accuracy and efficiency of leak source localization, offering critical technical support for chemical accident emergency response.

This paper proposes a fuzzy clustering algorithm based on the linear Gaussian state space model (LGSSM), referred to as q-divergence-based fuzzy c-linear Gaussian state space models (QFCLGSSMs) for series data. QFCLGSSMs are constructed from mixtures of linear Gaussian state space models (MLGSSMs), which is a conventional probabilistic clustering algorithm based on LGSSM for series data. QFCLGSSMs are motivated by the relationship between the following clustering algorithms for vectorial data: Gaussian mixture model with identity covariances and q-divergence-based fuzzy c-means. In numerical experiments that use an artificial dataset, we revealed the effects of fuzzification parameters on the fuzziness of clustering results in the proposed algorithm demonstrating the close relationship between the proposed algorithm and the conventional algorithm, MLGSSMs. Moreover, through numerical experiments, using nine real datasets, we demonstrated that the proposed algorithm outperformed the conventional algorithm, MLGSSMs, in terms of clustering accuracy.

This study proposes a method for processing unordered datasets using deep learning techniques and introduces a model capable of simultaneously predicting three context effects and selections. “Context effects” refer to dramatic changes in the selection situation caused by the introduction of new options into a set of choices. Five experiments were conducted using different product-selection data. Each experiment focused on predicting context effects, selections, predictions under different context effects, predictions for different types of products, and predictions that considered context effects. The results demonstrated that the proposed model is suitable for classification prediction tasks in complex situations. The more complex the model and the larger the amount of data, the better the results. This study extends the application of neural networks to multi-attribute decision-making problems and contributes to the selection of decision-making models. It also improves the prediction accuracy and analyzes the impact of context effects on choices.

This study explores the cultural integration challenges Chinese state-owned enterprises (SOEs) face during the post-merger integration (PMI) process. Based on the NK fitness landscape, an active search model is developed to analyze the interaction among acculturation modes, management activities, and organizational preferences. This model captures the integration dynamics through four acculturation modes (integration, separation, assimilation, and reverse merger) and evaluates their performance under different interactive complexities. The results indicate that the assimilation mode favored by state-owned enterprises aligns well with their cultural preferences in the short term. However, this mode fails to maintain long-term adaptability and performance. Moreover, management activities significantly impact integration results, with experience dependence and coordination levels particularly significant in complex environments. These findings emphasize that SOEs must develop cultural strategies tailored to specific backgrounds, providing actionable insights for optimizing PMI processes. This study links theoretical modeling with practical applications, enhancing the understanding of cultural integration in organizational mergers.

With the rapid advancement of large-language models in natural language processing (NLP), many studies have explored their role in brain encoding and decoding. In this study, we developed BrainLM, a pre-trained multimodal model that incorporates paired brain activity data from text stimuli. We demonstrated its accuracy in brain encoding and decoding across multiple NLP tasks. Our research produced several notable findings: we successfully developed a model for brain encoding and decoding, validated its reliability through bidirectional experiments, and outperformed 20 state-of-the-art models in brain encoding tasks. Additionally, we designed an autoencoder module to extract brain features. We extended the capabilities of BrainLM to new datasets and explored multilingual tasks using transfer learning, which enhanced the generalization ability of the model. Notably, BrainLM achieved 51.75% accuracy in binary classification tasks and increased the correlation coefficient by 3%–15% in brain prediction tasks. This study expands the applications of BrainLM and uncovers the complex interactions between brain regions and language models across different linguistic environments.

This study aims to develop a monitoring system for elderly individuals living solitarily using time-series data generated via simulation as training data. In particular, we focus on classifying three types of motion: falling, static standing, and walking. First, we create a system that calculates body velocity and acceleration using a depth camera. Based on actual measurements of each motion, we identify their distinct characteristics. Subsequently, we implement an inverted-pendulum model, which is commonly used for human-motion analysis, in a dynamics simulator. Simulations of falling, static standing, and walking are conducted, which successfully generated time-series data closely resembling the actual measured motions. Finally, using the simulation-derived time-series data as training data, we perform a machine-learning-based classification of falling, static standing, and walking motions measured using Azure Kinect. Although some misclassifications occurred, the system accurately classified most of the motions.

This study aims to develop a cable laying robot capable of operating in both ceiling spaces and on cable racks. The robot needs to possess the ability to pull lead cables through these environments. To achieve this, a reconfigurable autonomous robot mechanism is proposed to move across cable racks and ceiling spaces using a traveling wave generated by a single motor. The performance and operation of the robot developed in this research are verified through experiments and comparison with previously developed robots. As a result, the advantages have been successfully demonstrated by reducing the number of motors, weight, and driving voltage by half while maintaining comparable environmental adaptability and propulsion speed.

Since financial reports usually contain a large amount of data and complex information, traditional methods for quality inspection are not only slow but also difficult, which greatly affects the efficiency of quality inspection. This paper adopts knowledge graph and artificial intelligence methods to convert unstructured data in financial reports into structured data that can be quickly processed, thereby improving the efficiency and performance of financial report quality inspection. Therefore, this paper proposes an ALBERT-BiGRU-CRF model algorithm to perform named entity recognition on financial reports, which can effectively identify complex entities in financial reports; in addition, a RoBERTa-BiGRU model algorithm is proposed to extract the relationship between entities and finally construct the relevant knowledge graph. By analyzing the knowledge graph, relevant data inside the financial report can be obtained. The F1 score of the ALBERT-BiGRU-CRF model proposed in this paper is 6.1% higher than that of the BERT-BiGRU-CRF model, and the F1 score of the RoBERTa-BiGRU model proposed in this paper is 4.1% higher than that of BiGRU. The model proposed in this paper is of great significance for the knowledge graph modeling and quality inspection of financial reports.

Recent advancements in mechanization and automation have significantly transformed households and retail environments, with automated services becoming increasingly prevalent. In general, smart appliances utilizing the IoT technology have gained widespread adoption, and computerized systems, such as self-checkout machines, are now commonplace in retail settings. However, these services require users to follow specific procedures and operate the systems according to predefined capabilities, which may exclude users who are unfamiliar with the systems or who require additional support. Although robots deliver essential services efficiently, their rigid designs limits their adaptability. By contrast, human service providers can flexibly tailor services by observing a customer’s condition through visual and auditory cues. For robots to offer more inclusive and user-friendly services, they must be capable of assessing user conditions and adapting their behaviors accordingly. Therefore, this paper proposes a control support system that analyzes user gaze behavior during interactions with smart appliances to provide context-aware support. Gaze data were collected using HoloLens 2, a mixed reality device, allowing the system to deliver information tailored to the user’s gaze direction. By providing an information support service through a robot based on an analysis of the user’s gaze, the user’s level of interest in the targeted environmental objects could be confirmed. Accordingly, a service that improves convenience and is tailored to the user could be provided. Finally, we discuss the effectiveness of the proposed human-centric robotic system through experiments.

This paper proposes a concrete ticket-sales strategy in the live entertainment market by verifying the effectiveness of hybrid sales, which combine auction sales and uniform pricing methods, based on actual performances held in Japan. To conduct the social simulation of hybrid ticket-sales, we calculated the seat utility values of a particular theater and estimated the reserve price distribution of customers using transaction price data from ticket resale platforms. The results showed that while customer utility and revenue increased when auction sales seats were added, the introduction of more than a certain number of auction sales seats beyond the acceptable level of auction sales decreased customer utility. Furthermore, by allocating additional revenue from auction sales to adjust the uniform ticket price appropriately, we observed that setting approximately 40% of the total seats (700 seats) for auction is optimal for maximizing customer utility. This sales method ensures accessibility for a wide range of customers by providing premium seating to high-paying customers while maintaining affordable pricing for uniform pricing sales. Additionally, it was observed to be an effective sales system that can control illicit profits in the resale market and redistribute the profits from high resale prices to consumers and organizers.

With the development of social economy and technological progress, the integration of higher education and industrial demand has become an important way to enhance the entrepreneurial ability of college students. Currently, universities are manually selecting industries and entrepreneurship majors that integrate industry and education. The application of neural networks has brought new opportunities for the integration of industry and education. We propose a method of integrating industry and education based on neural networks, which optimizes the allocation of educational resources through neural networks and improves the entrepreneurial abilities of college students. By utilizing the deep learning capabilities of neural networks, the corresponding entrepreneurial needs can be intelligently matched with educational resources to achieve precise and personalized educational guidance. The superiority of the algorithm proposed in this study was verified through experiments. The classification accuracy has been improved to 95.5%, with an F1 score of 94.2%, which is 4.6% higher than traditional methods such as multi-layer perceptron. The coverage rate of the recommendation system is 87%, and the novelty index is 0.76, both of which are better than existing models. The success rate of student entrepreneurship has increased from 82.1% to 98.9%, and user satisfaction has increased to 99.1%.

Predicting the rate of penetration (ROP) is essential for improving drilling efficiency by optimizing the operational parameters. Accurate ROP prediction facilitates better decision-making, reduces drilling costs, and helps obtain optimal operational parameters. This paper proposes a new prediction model that combines Gaussian process regression and Bayesian optimization methods. First, the interquartile range and Savitzky-Golay filtering methods are used to denoise the data. To reduce model redundancy, appropriate input variables are identified based on Spearman correlation analysis. Second, a Gaussian process regression model tuned using Bayesian optimization is established to predict the ROP. Finally, public data sourced from the UTAH FORGE Well 58-32 dataset are used to validate the proposed model. The results indicate that the proposed model offers reliable prediction accuracy and enhances the ROP during drilling.

In modern society, the internet plays a crucial role in collecting tourism-related information. Traditionally, travelers collected tourism information from travel magazines, television, and travel agencies. However, with the widespread use of the internet, these methods of information collection have quickly shifted online. Travelers can easily access real-time, detailed information from official tourist destination websites, travel review sites, and social networking services (SNSs) through the internet. In addition, experiences, photos, and reviews posted by visitors on SNS serve as valuable reference information for other travelers. Such information frequently serves as a key factor in travelers’ decision-making when selecting destinations and planning their trips. Personal experience-based information is conveyed with reliability and familiarity on SNS, making it influential for many people. Therefore, collecting and analyzing tourism-related information from SNS is considered extremely beneficial for promoting tourism. This study uses text mining to analyze data collected from SNS. Furthermore, it is hypothesized that the most critical source of information on SNS is personal experience-based content. Therefore, this research also explores methods for extracting personal review information from the collected data.

An overlap window-based transformer is proposed for infrared and visible image fusion. A multi-head self-attention mechanism based on overlapping windows is designed. By introducing overlapping regions between windows, local features can interact across different windows, avoiding the discontinuity and information isolation issues caused by non-overlapping partitions. The proposed model is trained using an unsupervised loss function composed of three terms: pixel, gradient, and structural loss. With the end-to-end model and the unsupervised loss function, our method eliminates the need to manually design complex activity-level measurements and fusion strategies. Extensive experiments on the public TNO (grayscale) and RoadScene (RGB) datasets demonstrate that the proposed method achieves the expected long-distance dependency modeling capabilities when fusing infrared and visible images, as well as the positive results in both qualitative and quantitative evaluations.

This study systematically investigated the intricate relationship between viewers’ emotional responses and their viewing intentions toward animated promotional videos via a visual analog scale and fuzzy c-means clustering (FCM). Survey data collected from students in Japan (n=71) and Singapore (n=27) were analyzed via FCM, revealing four distinct viewer clusters: “high evaluation group,” “medium evaluation group,” “mixed group,” and “low evaluation group,” each exhibiting characteristic emotional response patterns. Multiple regression analysis revealed that joy (β=0.503) and excitement (β=0.276) had significant positive effects on viewing intention, accounting for 54% of the variance in viewing intention (adjusted R²=0.524). Statistically significant differences (p<0.05) were observed across cultural backgrounds, particularly in emotional responses to joy, with Singaporean students exhibiting greater appreciation. These findings contribute to optimizing promotional strategies for international video distribution platforms, emphasizing the importance of eliciting positive emotional responses and considering cultural variations in audience segmentation and targeting. A limitation of this study is its relatively small sample size, which may not fully represent the broader populations of Japan and Singapore. Future research should validate our findings using larger and more diverse samples to enhance their generalizability.

Machine learning, particularly deep learning, is a powerful tool for assisting radiologists in analyzing large volumes of chest X-ray images, significantly accelerating disease diagnosis. However, privacy regulations and data ownership challenges often hinder the centralization of sensitive patient data required for training. Federated learning (FL) addresses these issues by enabling decentralized model training while preserving data confidentiality. This paper introduces a novel FL framework for secure COVID-19 screening using chest X-rays. Our approach incorporates asynchronous communication to overcome delays caused by device heterogeneity and minimizes server-client interactions to reduce network traffic, enhancing scalability. Furthermore, the framework supports heterogeneous client models, ensuring optimized local training. These innovations preserve privacy while achieving performance levels comparable to centralized systems, setting a benchmark for privacy-preserving AI in healthcare.

Deep learning has achieved significant advancements in natural language processing. However, applying these methods to languages with complex morphological and syntactic structures—such as Russian—remains challenging. To address these challenges, this paper presents an optimized sentiment analysis model, GNN–BERT–AE, specifically designed for the Russian language. The model integrates graph neural networks (GNNs) with the contextualized embeddings of bidirectional encoder representations from transformers (BERT), enabling it to capture both syntactic dependencies and nuanced semantic information inherent in the Russian language. Whereas GNN excels in modeling the intricate word dependencies within the language, the contextualized representations of BERT provide a deep understanding of the text, improving the ability of the model to accurately interpret sentiments. The model further incorporates traditional feature extraction techniques—bag of words and term frequency–inverse document frequency—to preprocess text and emphasize critical features for sentiment analysis. To further enhance these features, a self-encoder clustering algorithm is employed, enabling the identification of latent patterns and improving the sensitivity of the model to subtle sentiment variations. The final phase of the model involves sentiment classification, categorizing emotions based on the enriched feature set. Experimental results showed that the GNN–BERT–AE model outperformed existing models—CNN–Transformer, RNN–LSTM–GRU, and Text–BiLSTM–CNN—on Russian social media datasets, achieving 1.25% to 3.1% accuracy improvements. These results highlight the robustness of the model and its significant potential for advancing sentiment analysis in the Russian language, particularly in handling complex linguistic features.

Accelerating the emergence of new quality productivity is a pivotal strategy for securing a proactive stance in the developmental trajectory of the contemporary era and forthcoming journey. Intelligent transformation plays a key role in this process. This study uses advanced text processing techniques and analyzes corporate annual reports to quantify the extent of intelligent transformation in listed Chinese firms between 2011 and 2022. It explores the holistic influence, underlying mechanisms, and diverse attributes of intelligent transformation in the cultivation of new quality productivity in enterprises. The results provide several key insights. (1) Intelligent transformation exerts a substantial positive effect on enterprises’ new quality productivity enhancement, which persists after considering endogeneity concerns and performing rigorous robustness checks. (2) Regarding the underlying mechanisms, intelligent transformation catalyzes an increase in enterprises’ new quality productivity by amplifying their technological innovation capabilities. (3) The magnitude of intelligent transformation’s impact on new quality productivity varies markedly across enterprises, influenced by factors such as ownership structure, geographical locale, and marketization degree.

Intelligent transformation is one of the primary strategies driving industrial upgrading, enhancing quality, and increasing efficiency in China. This study quantifies the extent of intelligent transformation among Chinese listed companies from 2009 to 2023, employing text processing techniques and analyzing annual reports. It subsequently investigates the comprehensive impact of intelligent transformation on these enterprises’ total factor productivity (TFP) and clarifies the dynamic mechanism enterprise environmental, social, and governance (ESG) performance plays in this process. The findings reveal that: (1) the introduction of intelligent capital leads to improved factor market competition, thereby reducing the dispersion of nominal TFP among firms and ultimately driving TFP growth; (2) intelligent transformation significantly enhances firms’ TFP, a conclusion that remains valid after considering endogeneity issues and conducting a series of robustness checks, thereby disproving the “productivity paradox;” (3) in terms of impact mechanisms, it promotes the improvement of TFP by enhancing corporate ESG performance; however, (4) the enabling effect of intelligent transformation on TFP varies significantly across firms based on the nature of their ownership, factor intensity, and geographical location.

This study aimed to examine the relationship between interns’ job satisfaction, job performance, including professional competence and teamwork performances, retention intention, and managers’ willingness to reemploy interns after graduating from internship programs and universities. A quantitative questionnaire survey method was adapted in this study. A total of 317 participants were recruited for a 24-item pilot questionnaire, and 329 valid formal questionnaires were confirmed for analysis using descriptive statistics, difference testing, and regression analysis. Then, there were five research major findings, and some implications and practical suggestions of the research were discussed.

This study investigates the impact of abacus learning on math scores among 1,226 second-year junior high school students in Taichung County and City. Controlling for the co-variable “intelligence test scores,” the study derives four conclusions. First, students whose abacus skills reached the “duan level” do not perform better in math scores. Second, the interactive effects of “gender,” “age at which one began to learn abacus,” and “starting mathematical foundation for abacus learning” do not affect math exam scores. Third, abacus ability has a minimal predictive function for math scores, with “intelligence test scores” having the highest explanatory power at 41% and “abacus ability” at a minuscule 1%. Fourth, it is suggested that kindergarten students should not receive abacus instruction but should rather develop their insight and thinking abilities.

This study introduces a specialized pre-marking robotic system that boasts a high degree of autonomy in response to low efficiency and inaccuracy in pre-marking operations for road delineations on newly constructed roads. The system is designed for autonomous navigation and precise spray-painting of road markings. It employs dynamic point sampling technology, enabling continuous and real-time acquisition of road coordinate information, thereby significantly improving pre-marking efficiency. A three-point circle correction method is implemented to generate the robot’s target path that includes curvature information. A curvature-adaptive pure pursuit control strategy is executed to ensure high-precision tracking of the pre-marking robot along the target path. Simulation experiments have confirmed the effectiveness and reliability of the robotic system. Practical applications reveal a marking error of less than 1.5 cm in long curved road scenario and 2 cm in right-angle curve road scenario. This result achieves efficient and accurate pre-marking operations and provides substantial technical support for road construction and maintenance.

During offshore exploration operations, vessels experience six degrees of freedom motion due to wind, waves, and currents. Among these motions, heave is particularly challenging to compensate for and significantly impacts the drilling process. This paper presents an active heave compensation control method for drill string systems in offshore exploration. To address parameter uncertainties, a single-neuron PID control approach based on a quadratic performance index is proposed. A simulation analysis was conducted using MATLAB software. The results demonstrate that the proposed controller provides smoother outputs than alternative controllers, highlighting its effectiveness in heave compensation.

Addressing the challenges in small object detection, particularly the issue that small objects often suffer from a lack of sufficient semantic information and are highly susceptible to background noise, this paper proposes an innovative algorithm, namely YOLOv8-FE. Firstly, to enhance the network’s sensitivity to small object detection, a P2-scale detection layer specifically designed for small objects is integrated into the model. Secondly, addressing the potential information loss during downsampling in traditional convolutional layers, an innovative downsampling module named RFAC-SPD is designed, aiming to more effectively capture and utilize features of small objects, thereby assisting the model in improving performance. Additionally, to mitigate the interference from background noise and strengthen the network’s ability to focus on object information, the study builds the C2f-CBAM module based on the convolutional block attention module (CBAM). Moreover, to fully integrate low-level feature information, minimize the loss of underlying detail information, and further enhance the network’s representational capability, an enhanced path aggregation network is proposed, significantly improving the effectiveness of network feature fusion. Experiments on the dataset VisDrone2019 show that the YOLOv8-FE algorithm exhibits superior performance and detection efficiency. Compared to the baseline algorithm YOLOv8n, its mAP50 and mAP50-95 have increased by 8.3% and 5.3%, respectively. Furthermore, with an inference speed of 77 frames per second, YOLOv8-FE meets real-time requirements, thereby validating the advancement and effectiveness of the proposed improvement algorithm. Furthermore, generalization experiments conducted on the DOTA and Caltech Pedestrian datasets demonstrate that the improved model achieves an increase of 2.7% and 6.8% in mAP50, respectively, fully validating the generality of the proposed model.

Image–text retrieval, as a fundamental task in the cross-modal domain, centers on exploring semantic consistency and achieving precise alignment between related image–text pairs. Existing approaches primarily depend on co-occurrence frequency to construct coherent representations of commonsense knowledge introduction patterns, thereby facilitating high-quality semantic alignment across the two modalities. However, these methods often overlook the conceptual and syntactic correspondences between cross-modal fragments. To overcome these limitations, this work proposes a consensus knowledge-guided semantic enhanced interaction method, referred to as CSEI, for image–text retrieval. This method correlates both intra-modal and inter-modal semantics between image regions or objects and sentence words, aiming to minimize cross-modal discrepancies. Specifically, the initial step involves constructing visual and textual corpus sets that encapsulate rich concepts and relationships derived from commonsense knowledge. Subsequently, to enhance intra-modal relationships, a semantic relation-aware graph convolutional network is employed to capture more comprehensive feature representations. For inter-modal similarity reasoning, local and global similarity features are extracted through two cross-modal semantic enhancement mechanisms. In the final stage, the approach integrates commonsense knowledge with internal semantic correlations to enrich concept representation and further optimize semantic consistency by regularizing the importance disparities among association-enhanced concepts. Experiments conducted on MS-COCO and Flickr30K validate the effectiveness of the proposed method.

This study applies the data analysis method of fuzzy clustering to conduct profile analysis and retrospective measurement of STEM university students in project-based learning (PBL) in a special project course. This study analyzes the changes in non-cognitive latent traits of STEM university students after they participate in PBL. The variables used for fuzzy clustering are the changes in non-cognitive latent traits of STEM university students after they participate in PBL. The profile analysis explores the differences in non-cognitive latent traits among the clusters of STEM students. The sample consists of 230 STEM students from a public university in Taiwan. These non-cognitive latent traits include learning satisfaction, grit, mindset (growth mindset/fixed mindset) and self-efficacy. The STEM students come from four departments, namely science (S), technology (T), engineering (E), and mathematics (M). The results of the study indicate that after one semester of PBL in a special project course, the students’ non-cognitive latent traits significantly improve. Students majoring in science and engineering have significantly improvement in learning satisfaction, grit, growth mindset, and self-efficacy, but have slightly declined in fixed mindset, not to the significant level. Students majoring in technology and mathematics have significantly improved their learning satisfaction, grit, growth mindset, and self-efficacy, while their fixed mindset has significantly decreased. For students of different genders, both of them have significant improvement in learning satisfaction, grit, growth mindset, and self-efficacy. On the contrary, fixed mindset has significantly decreased. Based on the changes in non-cognitive latent traits, fuzzy clustering identifies three clusters of STEM students. Additionally, profile analysis reveals that each cluster exhibits unique characteristics and there are significant differences in the changes in their non-cognitive latent trait among clusters. This study provides valuable methodological insights by integrating fuzzy clustering and profile analysis. Moreover, the findings of this study also provide insights and implications for STEM education.

This study examines the perspectives of 431 junior high school mathematics teachers on the correlation between early childhood abacus education and the enhancement of mathematical abilities. The key findings are as follows. (1) The regression model indicates a moderate explanatory power (54%) concerning teachers’ educational background and abacus learning experience in predicting their views on abacus education’s role in developing mathematical abilities. (2) Approximately 60% of the respondents suggested that abacus education should complement formal mathematics instruction. (3) About 70% of teachers advocate for innovative teaching methods. (4) Around 73% of participants believe that kindergartens should prioritize fostering children’s holistic development.