In the field of personalized learning path planning for higher education, traditional methods lack in-depth analysis of students' dynamically changing learning status and interests, resulting in insufficient personalization, and focus on the analysis of a single data type, making it difficult to integrate data of different modalities. This paper proposes a multimodal learning analysis method for personalized learning path planning in higher education, addressing the limitations of traditional methods that do not account for dynamic changes in students' learning behaviors and interests. The approach integrates Transformer models, adversarial training, and quantum state classification to analyze multimodal data, including text, audio, and video, to capture learning patterns. The Transformer model uses a self-attention mechanism to generate personalized learning paths based on integrated data. Adversarial training is applied to simulate abnormal data and enhance the model's robustness to various learning scenarios. Quantum state classification improves data processing efficiency, addressing challenges in handling high-dimensional multimodal data. Experimental results show that the Transformer model achieves stable accuracy of 0.95 and recall of 0.91 for personalized learning path generation. The adversarial training method reduces the loss value to around 0.05, while the introduction of quantum state classification reduces processing time to 56 s in the 18th round, a 31-s improvement. These results confirm the effectiveness of the proposed method in enhancing the accuracy, robustness, and computational efficiency of personalized learning path generation in higher education.

This study introduces an optimized image encryption algorithm that integrates Henon mappin g and Arnold transformation to enhance the security and randomness of digital image encryption. The algor ithm is designed to address the vulnerabilities of open network environments where data theft or corruption can compromise image quality during encryption. Initially, images are processed through grayscale convers ion to reduce dimensionality, followed by Henon mapping to induce a chaotic sequence that scrambles the p ixel matrix. Subsequently, Arnold transformation is applied, iterating 100 times to further disrupt the image structure, ensuring a high level of diffusion and complexity. The proposed method demonstrates superior pe rformance with an average pixel change rate (NPCR) of 0.9982 and a normalized average change intensity (NACI) of 0.3654, significantly increasing resistance to differential attacks. The encrypted images exhibit h igher information entropy and effectively mask the original data, although at the cost of extended encryptio n time due to the dual scrambling process. The study concludes that the combined use of Henon mapping an d Arnold transformation not only strengthens encryption against various attacks but also introduces a novel approach to image encryption, with potential for further optimization to enhance efficiency in handling larg er images. This advancement is crucial for protecting privacy and ensuring data integrity in the transmission and storage of images, particularly in the face of evolving cyber threats.

This study proposes a domestic edge computing terminal pest identification system based on large model compression and lightweight technology, integrating new algorithms based on Transformer and ConvLSTM, and optimizing its performance in resource-constrained environments through adaptive deployment strategies. By evaluating the performance of four different models (Transformer ConvLSTM, CNN, LSTM and Transformer) in terms of recognition accuracy, inference speed, computing resource consumption and adaptive deployment effect, this study reveals the advantages and disadvantages of each model in edge computing tasks. The model based on Transformer ConvLSTM performed best in recognition accuracy, reaching 94.5%, significantly better than other models; while the CNN model was the most efficient in inference speed, reaching 52 frames per second (FPS). The adaptive deployment strategy of the Transformer ConvLSTM model, the Transformer ConvLSTM model can dynamically adjust the model complexity to effectively cope with higher computing resource requirements, thereby improving inference speed and reducing resource consumption. On resource-constrained devices, the optimized Transformer ConvLSTM model can maintain high recognition accuracy and significantly reduce latency. Although the compression strategy slightly affects the accuracy, it greatly improves the inference speed, especially in the CNN model.

This article aims to design and apply a quality evaluation index system based on Bloom's cognitive domain classification method to evaluate the training quality of big data professionals in universities. With the rapid development of big data technology, the importance of cultivating big data professionals in universities is becoming increasingly prominent. However, the existing training quality evaluation system often lacks systematicity and scientificity, making it difficult to fully reflect the comprehensive abilities of students. This article combines key knowledge of big data in the field of finance and economics with Bloom’s cognitive domain classification method to divide the cognitive process of talent cultivation into six levels: memory, understanding, application, analysis, evaluation, and creation. Through expert interviews and the Delphi method, specific evaluation indicators for each cognitive level were determined, and 18 indicators were extracted. Through data collection and empirical analysis, the performance of students at various cognitive levels was evaluated. The results showed that third year students performed the best on average at all cognitive levels, reflecting their rich learning experience and practical skills. The analysis also shows that there are significant differences in the performance of students at higher-order cognitive levels, with a coefficient of variation of 0.23 for the A1 indicator and 0.84 for the F3 indicator. This indicates that higher-order cognitive activities require higher demands from students, with significant individual differences. Through correlation analysis between various indicators, it was found that the proficiency level of tool usage in the project is strongly correlated with other indicators. Bloom’s cognitive domain classification system can reflect students’ comprehensive abilities and provide a scientific evaluation method for universities. Through continuous improvement and in-depth research, we hope to provide more scientific and comprehensive support for the cultivation of big data talents in universities.

This study presents an innovative image encryption algorithm anchored on the two-dimensional Logistic chaotic system, designed to address the limitations of traditional methods in terms of speed and security. The algorithm employs pixel shifting and scrambling, coupled with a segmented grayscale value substitution, to achieve a high level of security with strong key sensitivity and robust resistance to both differential and statistical attacks. Experimental validation demonstrates its superiority with fast encryption speeds, as evidenced by an encryption time of 0.2136 seconds for a 256×256 color image, and an entropy close to its ideal value, indicating excellent randomness. This method also exhibits high performance in resisting differential attacks, with Number of Pixels Change Rate (NPCR) and the Unified Average Changing Intensity (UACI) values nearing their optimal expected values. The integration of Quantum Key Distribution (QKD) is suggested for future work to enhance security against quantum computing threats, solidifying the algorithm’s potential as a leading solution for the image encryption.

Modern enterprises increasingly depend on data‐driven decision‐making, yet traditional SQL queries require technical expertise, limiting accessibility for nonspecialists. Advances in natural language processing, particularly deep learning generative models, have enabled text‐to‐SQL (text2SQL) conversion, making database interaction more intuitive. Retrieval‐Augmented Generation (RAG) enhances this by integrating retrieval and generation for greater accuracy and relevance. This article proposes a text2SQL business intelligence system based on RAG, allowing enterprise users to extract insights from complex databases using natural language queries. By streamlining data retrieval and lowering technical barriers, the system achieves state‐of‐the‐art performance in generating SQL queries for complex tasks. It leverages the BERT (Bidirectional Encoder Representations from Transformers) model for vectorized retrieval, Generative Pretrained Transformer 4 (GPT‐4) for query generation, and Graph Neural Networks (GNNs) for modeling database structures. User interaction and feedback mechanisms further refine semantic understanding and query accuracy. Experimental results demonstrate the system's effectiveness. For multitable joins, query matching accuracy using BERT + GPT‐4 + GNN reaches 52.3% and 55.1% with beam widths of 1 and 10, respectively. For nested queries involving multitable joins, accuracy increases to 60.2% and 61.9% under the same conditions. Additionally, the system achieves the highest user satisfaction scores, validating its practical utility. By enhancing the ability to handle complex queries and reducing data access barriers, the proposed RAG‐based text2SQL system provides enterprise users with an efficient, user‐friendly tool for database interaction, significantly improving decision‐making processes.