Sixth-Generation (6G) networks necessitate intelligent and energy-efficient operations. However, the direct deployment of Large Language Models (LLMs) for real-time 6G control is hindered by significant latency and energy constraints, conflicting with green 6G imperatives. This paper pioneers a collaborative architecture of large-small AI models to address these challenges. In this architecture, resource-intensive LLMs, i.e., large models, are strategically employed offline for comprehensive Wireless Data Knowledge Graph (WDKG) construction, effectively distilling domain knowledge. This WDKG, in turn, enables the development and deployment of lightweight, efficient small models for real-time network tasks by facilitating the generation of optimized feature datasets. To operationalize this, we first introduce a novel multi-agent collaborative LLM framework, bolstered by an enhanced semantic representation method incorporating domain-adaptive embedding fine-tuning and mutual information (MI)-based feature encoding, for automated high-fidelity WDKG construction. Subsequently, we propose the Semantic-Data and Spatio-Temporal (SD-ST) model, which uniquely fuses LLM-extracted semantic information with the spatio-temporal characteristics of wireless network data and WDKG structure. Insights from the SD-ST model guide a WDKG-driven method for generating optimized feature datasets by evaluating node influence and redundancy. Experimental validation demonstrates that these distilled feature datasets lead to substantial reductions in training and inference overhead for the lightweight downstream AI models, offering a tangible pathway towards greener, more efficient, and intelligent 6G networks.

In this paper, a hybrid model composed of the correlation ratio analysis and the Support Vector Machine (SVM) is proposed for trip mode recognition. The correlation ratio analysis algorithm is applied to determine the optimal time window of the input attributes so as to optimize the input parameter s. The SVM is applied to carry out the trip mode recognition for the whole trip. On this basis, the influence of data sampling frequency on trip mode recognition is further evaluated using large-scale field test data. The results show that: (1) the correlation ratio analysis and the SVM hybrid model attains the best performance for trip mode recognition, the average mode recognition precision reaches 89.8% at the sampling frequency of 1 s; (2) the data sampling frequency significantly affects the trip mode recognition effect; when the sampling frequency is high (less than 5 s), the mode recognition precision is above 70%, however, when the sampling frequency is relatively low (more than 30 s), the bus and car mode recognition precision falls rapidly below 37%. These results provide a reference for using the smartphone sensor datasets to supplement or even replace household travel surveys in transportation planning in the future.