Nowadays, there is a plethora of computer data that needs to be compressed in order to save storage space, improve transmission, speed up sorting, etc. It is in this context that we propose to compare three different compression programs (winrar, winzip and a new prototype) in order to evaluate the characteristics of the prototype and classify it among these peers. This study could help us in the future to improve this prototype according to our needs. This ranking will therefore be based on the ease of use, operation, advantages and disadvantages of each candidate in order to facilitate the choice. For the sake of objectivity, this will be done without bias so as not to distort the final results.
Regarding the gas scheduling issue in the coal gas systems of steel enterprises, a gas scheduling optimization model is proposed based on the multi-input and multi-output characteristics of the gas system. This model combines knowledge-based and dynamic programming approaches, utilizing production experience to establish scheduling knowledge. The model calculates the scheduling results under two scenarios: non-optimized and optimized, across various time dimensions (1 minute, 15 minutes, 30 minutes) and different production conditions (with and without blast furnace maintenance). The results from simulation experiments with field data demonstrate the model's timeliness and practicality, providing a reasonable guide for developing gas system schemes. This model is particularly suitable for steel production enterprises with no gas holders and significant pipeline pressure fluctuations, offering a scientific and effective guidance method. This research can be applied to solve energy optimization problems in complex production systems and scenarios.
Quantum technology has become a strategic field that must be contested globally. On the basis of combing the current research situation at home and abroad, this paper analyzes the quantum science and technology plans of the United States, the European Union, Germany, the United Kingdom and other countries or regions from the aspects of “basic overview and implementation effect”, summarizes the main characteristics of these quantum science and technology plans, and proposes countermeasures and suggestions to promote China's quantum research on this basis, in order to provide empirical evidence for China's quantum planning and quantum technology breakthrough.
The learned index is a disruptive technique for big data management. The core idea of the learned index is to retrofit the data indexing in traditional index structures with machine learning (ML) models, providing fast data position prediction instead of slow index tree traversal. Yet, current learned indexes focus on fix-sized numeric data and all of them perform poorly for variable-length string data for two main reasons: 1) high computational overheads caused by lengthy string key prediction; 2) large ML model fitting errors due to variance in string prefixes. This paper introduces SLIN, a CPU-efficient, hybrid index for variable-length string data. SLIN features a hybrid tree and learned index structure that leverages the strength of each index to overcome the above-mentioned challenges. It proposes a string slice approach which converts variable-length strings into fix-sized integers, which facilitates using one-dimensional linear regression for key prediction to reduce computational overheads. Furthermore, it extracts the common prefix from string keys to decrease model fitting errors. Finally, we propose a strategy to adaptively choose the appropriate index to manage computational costs. Evaluations compare SLIN with state-of-the-art index structures. The experimental results show that SLIN achieves up to 2.47x higher throughputs and relatively small fitting errors.
The paper presents a comprehensive Natural Language Processing (NLP) evaluation framework for online public opinion analysis, addressing the limitations of singular NLP techniques. It highlights the significance of public opinion analysis in the era of big data and reviews commonly employed NLP methods. The proposed framework offers universality, comprehensiveness, and flexibility, aiding managers and users in better understanding product sentiments and purchasing behaviors for informed decision-making. The proposed NLP evaluation framework includes five stages, Data Collection, Qualitative Analysis, Sentiment Analysis, Semantic Relationship Mining, and Comprehensive Analysis. It involves techniques like Sentiment Analysis, Topic Modeling, and Co-Occurrence Networks to analyze sentiment and identify semantic relationships. Our case study on nursing care food market validates the framework's feasibility, proposing enhancements in taste, gifting nature, pricing, and product strategy. Future research will expand its applicability across domains and refine inter-stage connections while maintaining flexibility.
This paper proposes an adaptive multispectral pedestrian detection method that leverages multispectral data and optimizes both channel attention and spatial attention mechanisms. To address varying illumination intensities, a different fusion strategy is applied, leading to the development of an adaptive multispectral pedestrian detection network (ASP). In the preprocessing phase, the KAIST dataset is obtained, and training images are standardized in size and categorized based on lighting intensity. During the model construction and training phase, the categorized images are input into the ASP model for training. In the testing phase, the trained network is employed to perform real-time pedestrian detection on input images, generating accurate detection results. By effectively utilizing information from different modalities, the proposed network enhances feature richness and distinctiveness, facilitating complementary integration of multispectral data and achieving deep fusion. Experimental results demonstrate that the network exhibits significantly improved robustness in complex scenarios, achieving a pedestrian detection accuracy of up to 91%.
People's life is closely related to the quality and safety of agricultural products. Traditional agricultural product traceability technology uses centralized management, which has many drawbacks, particularly easy manipulation, and unreliable centralization. This paper proposes a traceability model based on blockchain that consists of the agricultural product blockchain model, distributed storage model DSB, agricultural product traceability model, practical Byzantine consensus algorithm PBFT, and smart contract to realize the safe and distributed traceability of agricultural products. The results showed that the blockchain-based method of agricultural product traceability is accurate, practical, and efficient. The model has a success rate of 100%, a minimum average time delay of 20326 ms, a maximum throughput of traceability of 4702.03 b/s, and distinct curvilinear characteristics in the performance parameters as the number of nodes is raised. The study's conclusions provide a useful design framework and implementation method for blockchain-based agricultural product traceability technology, Serve as a guide for research and development on related technologies, and have reference value.
Social media platforms have become significant channels for the public to express their opinions, and consumers' shared shopping experiences are crucial for consumer purchase decisions as well as for merchants' product improvements. This study focuses on capturing e-commerce product information from popular social media platforms, converting video data into image and text forms, using pre trained models such as CNN to extract image and text features, achieving modal transformation, and constructing entity relationships. Ultimately, the integrated multi-modal data and cross-platform data are stored in the Neo4j database to build a knowledge graph. Taking footwear products as an example, this article further practices the application of pre trained models in building knowledge graphs, including directly using pre trained models, fine-tuning pre trained models using existing datasets, and fine-tuning pre trained models using manually annotated datasets. Ultimately, a multimodal footwear product knowledge graph is constructed to help businesses improve their products and assist consumers in making purchasing decisions.
Currently, there are many algorithms that can be used for text multicategorization, but each algorithm has its own specific assumptions, advantages, and disadvantages. Aiming at the multiclassification problem of short text, this paper crawls the title and abstract text of blogs on 51CTO blog website and its domain classification as a dataset, proposes a feature representation method of fusing Word2Vec word vectors with Bert characterised text vectors after weighting them with TF-IDF, based on the three commonly used machine learning base classifiers and their deep learning models, namely, Decision Tree, Random Forest and KNN BERT, and introduce the integration algorithm for the calculation of the F1 value of the model, which is used as an evaluation index of the merits of text classification algorithms, and compared with the experimental results of the single text feature representation method, with a view to improving the performance of the classification model through the fusion of word vectors. The experiment shows that compared with the single text feature representation method, the model performance is improved under the fusion feature vector representation, and its F1 value reaches 86.6% on the integration algorithm in traditional machine learning, which is improved by 0.6%∼1% compared with the original one; the splicing of fused features with BERT makes its F1 value also improved by about 2%, which indicates that the fusion features proposed in this experiment can be more comprehensive, accurately describe the semantic information of text, and can more effectively characterize short texts.
In recent years, with the rise of artificial intelligence technologies such as deep learning, natural language processing (NLP) has become a hot topic. However, research on contractual texts, particularly employment contracts, remains scarce due to challenges such as the limited availability of contract data. This paper pioneers a study on the classification and risk assessment of employment contract texts. Considering the unique characteristics of employment contract texts, we propose a fusion model that leverages the strengths of the BERT, CNN, and LSTM models, termed the BCBR model. This model integrates BERT's self-attention mechanism to capture semantic relationships between words, CNN's ability to extract local spatial features, and LSTM's capacity to learn time-series dependencies. Initially, the self-attention mechanism in BERT is utilized to grasp the semantic relations among words. Subsequently, CNN is employed to extract textual features, while LSTM captures long-term dependencies. The classification results from BERT-CNN and BERT-LSTM are then linearly weighted. For comparative analysis, we benchmark the BCBR model against several others, including BERT-CNN, BERT-LSTM, TEXTCNN, TEXTRNN, and BERT.The results demonstrate that the BCBR model achieves the highest F1 scores, which is the harmonic mean of precision and recall, at 0.9168, 0.9143, and 0.9136, respectively. Based on these findings, we conduct a risk analysis of the contracts and present our conclusions along with future prospects.
Aiming at the problem that traditional short term memory networks (LSTM) and convolutional neural networks (CNN) fail to effectively reflect the importance of each word in the text in the process of feature extraction, this study proposes a hybrid model based on BiLSTM, Attention and CNN (BiLSACNN). In order to further improve the performance of the text classification model, we can fully mine the temporal, global and local information in the text. By comparing SVM, CNN, BiLSTM, BILSTM-attention and fusion model on cnews data set and privately constructed csdns data set, it is found that the hybrid model has excellent performance in different fields, compared with traditional LSTM and CNN. It can better reflect the importance of each word in the text, which can effectively improve the effect of text classification.
Designed experiments can be a powerful tool for gaining fundamental understanding of systems and processes. Many experiments on a system or process have multiple responses of interest, each with potentially different data types and hence underlying distributions. In development tests and operational tests, tests with multiple responses are common. Many tests have continuous responses, binary responses, and counting responses. However, due to limited conditions, it is impossible to conduct tests for different responses separately. Instead, a relatively optimal experimental design must be used to simultaneously test whether these responses meet specific index requirements. However, even based on the same optimization criterion, the optimal design for a normal response may be very different from the optimal design for a non-normal response. This work presents a new approach to experimental designs for multi-response systems. The normal, binomial and Poisson distributions are considered as potential responses. This method further adds the statistical power optimization criterion on the basis of the weighted D-optimal design criterion, which can take into account the D-optimal design efficiency and statistical power of the experimental design and achieve the optimal design of multi-response experimental problems. The statistical power of normal and non-normal responses is evaluated by analytical method and Monte Carlo method respectively, and is implemented in the point exchange algorithm. The method is demonstrated and illustrated through a case study of experimental design for testing the detection performance of an unmanned surface vessel(USV) against an aerial target.
This study uses Chinese listed companies as a sample and employs an entropy weight-TOPSIS model combined with a panel data analysis method that integrates machine learning to explore how media attention affects corporate asset efficiency. The findings reveal that media attention significantly enhances corporate asset efficiency, with particularly pronounced effects in non-state-owned enterprises, online media, and enterprises in the eastern regions. Heterogeneity tests further reveal the complexity of the relationship between media attention and corporate asset efficiency. The research outcomes provide strategic guidance for companies on how to utilize media attention to optimize asset management and offer valuable insights for policymakers.
In recent years, the rapid development of the new energy vehicle (NEV) industry has been driven by the global shift in energy structures and increased environmental awareness. Accurate sales forecasting has become crucial in the automotive industry to enhance production efficiency and market responsiveness. This paper focuses on China FAW's NEVs, utilizing data from 2019 to 2023 to predict sales using an ensemble of neural network algorithms. We constructed three models—Convolutional Neural Network (CNN), Feedforward Neural Network (FNN), and Multilayer Perceptron (MLP)—and analyzed multiple factors, including automotive indices, new energy indices, and loan interest rates, to forecast sales for 2024. The experimental results indicate that CNN outperforms traditional statistical methods and other neural network models across various metrics. Additionally, integrating the three neural network models further enhances predictive performance, demonstrating high accuracy and stability. In the future, optimizing and combining multiple models may provide even more precise forecasts in complex market environments.
Vehicle routing problem is a NP problem. With the continuous in-depth study of VRP model and its solving methods, many branch problems with more constraints have been developed, and their solving methods have become more and more diverse. In order to further review the research status of VRP at home and abroad. This paper employs bibliometric methods and knowledge graph analysis to examine publishing institutions, authors, and research hotspots in vehicle routing optimization studies both domestically and internationally from 2008 to 2023, aiming to provide insights for the field of vehicle routing optimization in China. The main research content includes the use of CiteSpace visualization software to generate the keyword co-occurrence and clustering knowledge map of both domestic and international literature, as well as keyword burst detection maps, so as to interpret the research status in the field. The results show that in the field of vehicle routing optimization, the cooperation relationship between foreign institutions is closer than that of domestic institutions, and research hotspots differ significantly between domestic and foreign studies. Foreign scholars pay more attention to the combination of theory and practice, while domestic scholars focus more on deepening algorithm research. Meanwhile, the research processes at home and abroad show different characteristics. Cold chain logistics and urban transportation have become areas of focus for domestic scholars, who have begun to shift from traditional vehicle route optimization problems to in-depth studies of green transportation. Integrating big data analysis and artificial intelligence technologies into vehicle route optimization is a future research direction.
Next POI recommendation aims to forecast their subsequent locations, which base been an important applications for Location-Based Services (LBS). Recently, there has been a growing interest in the investigation of POI recommendation using various deep learning techniques, and achieved great success. However, most of them only model the users preferences from a single view, such as the graph-structured features in the trajectory that indicate the preferences of individuals, or the dynamic change of personal preferences from trajectory sequences. Moreover, they ignore leverage the unique self-supervised signals to guide those two views of user preferences leaning perform better. To fill these critical voids, this paper proposes a Multi-View Contrastive Learning for next POI recommendation named MVCL-POI. Specifically, MVCL-POI first exploits dynamic-aware GCN from trajectory graph to automatically extract graph-structured features with considerations of rich node connections. After that, MVCL-POI utilizes a temporal-aware transformer to capture users’ dynamic interest from their historical trajectory sequences. Finally, a multi-view contrastive learning is designed to maximize the agreement between two views of users’ representations. Experiments results on two real-life LBSN datasets show that MVCL-POI yields significant performance advantages over the competitive next POI recommendation methods.
The high-dimensional data in market price prediction is a great challenge that was not effectively addressed by the traditional data-driven feature selection approaches. This paper introduces a novel Domain-Knowledge based Feature Selection Framework (DKFS), specifically designed for product-oriented applications. By adopting a two-step approach, the traditional statistical method (i.e., filter method) is integrated with domain-specific knowledge, offering an enhanced layer of selection that ensures a rigorous and efficient exclusion of irrelevant or redundant features. The framework was applied to a real-world application of sailboat price prediction, with three modelling techniques (Multiple Linear Regression, Random Forest, and Gradient Boosting) evaluated based on a comprehensive dataset of over 2,500 sailboat transactions. The adopted approach demonstrated an exceptional performance capturing 90.8% variability with a small set of 26 features, including economic indicators and geographical factors. The proposed framework illustrates significant effectiveness in dimensionality reduction and offers broad applicability across various domains. It also presents a promising direction for further research into the use of expert systems and adaptive feature selection design.
Tea picking has always been mainly manual, but with the development of technology, it has been possible to use computer image recognition technology to assist robots in identifying tea leaves and picking them. However, the special characteristics of tea leaves make it difficult for the machine to recognize them with high accuracy and efficiency. In this paper, we compare the currently available algorithms and perform tea image recognition based on the YOLOv5 algorithm model. Based on the original YOLOv5 model, the CBAM attention mechanism is added to improve the recognition ability of tea. The mAP of our model can reach 89.1%, which is 13.6 percentage points higher than the original YOLOv5 model.
Aiming at the problems of high risk and low efficiency in the process of transportation of chemical materials, which are corrosive, toxic and harmful, flammable, and explosive in the modern intelligent warehousing logistics and supply chain system, a universal dual-side butt automatic flip machine was designed, and the main structure and working principle of the flip machine were introduced. Through the theoretical calculation and virtual simulation, the structural design of the flip machine was completed, and it was concluded that the deformation and stress of the key parts of the flip machine, such as the clamping mechanism, main rotating shaft, and support frame, meet the design requirements under 1.35t rated external load. On this basis, the prototype of the flip machine was made, and 2880 times of flipping tests were carried out in three days. The results show that the flip machine has a remarkable level of safety and a high degree of automation, and the flipping efficiency of materials under 1.35t can reach 40 pallets/h, which is greatly improved compared with the traditional mode, and the feasibility and reliability were preliminarily verified.
