The quality of method names is crucial for the readability and maintainability of programs. However, constructing high-quality method names often presents a significant challenge. To address this issue, numerous method name recommendation approaches have been proposed in existing research. These approaches typically require large-scale, highquality software projects to build benchmark function libraries to recommend or generate method names for given method bodies. They generally rely on extracted function features (including function parameters, return values, etc.) as feature vectors. However, these approaches have limited success rates and accuracy in practical applications. To overcome these limitations, we propose a method name recommendation approach, called LMMName, based on large-language models. This approach considers not only the method body but also its interacting functions (i.e., code context), including called and calling functions, and uses this as input to the large language model to generate method names. Our versatile approach implements LMMName using the state-of-the-art large-language model ChatGPT (4.0). Experimental validation in two medium-sized open-source Java projects demonstrates that method name recommendations using this approach have a success rate of 100%. This represents an improvement of 9.86% and 12.54% compared to the recommendation of the method name based on the source code repository and feature matching, regardless of whether the code context is considered. Taking into account the code context relationships, this approach recommends method names that are completely identical to the original project method names at a rate of 55.54 % and 53.99%, respectively. This marks an increase of 27.76% and 29.60% compared to when code context information is not considered. Furthermore, after factoring in the code context, the F1 metrics of this approach are 74.90% and 67.69%, respectively. These figures represent improvements of 37.17% and 21.08% compared to the machine learning-based code2vec approach, and 50.61% and 33.26% compared to the HeMa heuristic search-based approach. This demonstrates the effectiveness of our proposed approach.