This paper focuses on severe variable missingness and variable sparsity in device-edge collaboration. Inference-based methods yield unacceptable reconstruction errors under high missing rates, thus necessitating retransmission. However, this approach can lead to network congestion and even break-down in emergency response scenarios. To overcome these limitations, we propose a Similarity-based Multivariate Data Completion Framework (SMDCF). Unlike traditional retransmission approaches, SMDCF adopts an innovative similarity-based learning paradigm. For handling variable missingness, the framework retrieves similar samples from offline datasets by analyzing sparse observations and then effectively transfers relevant variables from these reference samples to complement the incomplete data; for handling variable sparsity, SMDCF employs graph neural networks to systematically model three key relationships: inter-sample correlations, inter-variable dependencies, and intra-variable patterns, thereby achieving robust and reliable missing-value inference. Comprehensive experimental results demonstrate that SMDCF consistently delivers superior performance under various data missing scenarios. Notably, even under an extreme 90% missing data rate, our framework reduces the reconstruction error by 8.37%.

Recent studies highlight the significant potential of edge computing and federated learning (FL) in advancing artificial intelligence. However, challenges, such as unstable device performance and the heterogeneously distributed feature of local data, pose threats to the efficiency of global model training. To address these issues, we propose an efficient federated edge learning protocol with key innovations: 1) propose a timing query mechanism, which controls the impact of slow clients on interaction time and changes the parameter server from a passive receiver to an active querier to guarantee the aggregated subset size; 2) propose a screening supplementary strategy from the perspective of optimizing the quality of the data combination; and 3) integrate the timing query mechanism and screening supplementary strategy in a flexible manner. This combination improves the interaction efficiency of global model training, focusing on both interaction time and aggregation update quality. Experiments show that while ensuring the convergence efficiency of each round and the global model performance, the interaction efficiency has been improved by 25% to 75%, thus ensuring the efficiency of FL.

Based on the LDA model, this paper builds a three-layer semantic model of Web English educational resources “document-topic-keyword”, models the semantic topics of resource documents, and obtains the semantic topics and keywords of document resources as the semantic labels of resources. The experimental results show that document LDA topic modeling is beneficial to the macroscopic classification of Web English educational resources. The experimental results show that LDA topic modeling of documents is useful for macroscopic cataloging of Web English educational resources, highlighting teaching priorities, difficulties, and interrelationships, while LDA modeling of teaching topics with the same teaching content expands the metadata generation method of resource description based on the basic education metadata standard and provides more information about the inherent characteristics of resources. The semantic information can be used to mine the semantic thematic features and detailed differences inherent in the resources, and the final performance analysis verifies the parallel computing advantages of the LDA model in a big data environment.