Automatic Service Composition Agent Based on Large Model

With the rapid development of information technology, there is an escalating demand for services that are both comprehensive and personalized. Obviously, a single service can no longer satisfy with the needs of life. Scholars have attempted to compose a large number of services using artificial intelligence methods such as deep learning and reinforcement learning to achieve better results. Despite the inherent adaptability of these methodologies, they are still beset with challenges in accurately gauging the quality of services, leading to suboptimal composition in environments characterized by high dynamism. To address this issue, this paper creatively proposes an Intelligent Agent for Automatic Service Composition based on Large Language Models (ASCALM). Firstly, it can automatically generate a list of the most suitable APIs needed for service composition according to the specific needs of the user, next, we use ChatGPT-4 to select and confirm the candidate API, perform the execution operations on the candidate APIs through the decision tree algorithm, and finally, the large language model combines and analyzes all the results to generate the final answer. The experiments are carried out based on thousands of diverse API's from different domains such as finance, food, economy, travel, science, etc. as a dataset. Under the action of our proposed method, the effect of using the big model to compose different API executions can reach the level of composition by experts in relevant specific fields. In the future, this kind of automatic composition model can be applied to all aspects of human life.