Scholarlink

<p>Service-driven compositions or mashups have successfully alleviated the invocation constraints associated with specific domains, leading to an increased frequency of service evolution. Current approaches struggle to recommend services absent in the training data due to the dynamics of evolving services and cannot automatically compose services according to the recommendations. To tackle these problems, this paper proposes a <b>La</b>rge Language Models (LLMs)-based <b>S</b>ervice <b>P</b>rocess <b>A</b>utomation framework named LaSPA. Firstly, LaSPA includes a system that can handle the evolution of software services, involving adaptively adding and deleting services. When user requirements arise, LaSPA retrieves the Top-<i>k</i> services by assessing the similarities between the requirements and available services in the repository. Then, LaSPA transforms the service recommendations into multiple-choice questions for accurate recommendations that rely on the semantic understanding of services with the help of LLMs. Finally, LaSPA utilizes the reasoning ability of LLMs to orchestrate the recommended services for the service process automation. Our observations suggest that LaSPA outperforms baseline methods. Compared to the recommendations approach, LaSPA can more easily cope with the evolution of services. The evolution experiment results indicate a 35.9% improvement in recall@5 and a 76.6% increment in precision@5, on average. Furthermore, a case evaluation shows the possibility of automating service processes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LLMs-based decision making for service recommendations and process automation under evolving ecosystem

  • 期刊论文

  • 发表日期：

    2026年2月27日

查看全文
  • Guodong Fan,
  • Shizhan Chen,
  • Hongyue Wu,
  • Cuiyun Gao,
  • Jian Wang,
  • Zhiyong Feng

摘要

Service-driven compositions or mashups have successfully alleviated the invocation constraints associated with specific domains, leading to an increased frequency of service evolution. Current approaches struggle to recommend services absent in the training data due to the dynamics of evolving services and cannot automatically compose services according to the recommendations. To tackle these problems, this paper proposes a Large Language Models (LLMs)-based Service Process Automation framework named LaSPA. Firstly, LaSPA includes a system that can handle the evolution of software services, involving adaptively adding and deleting services. When user requirements arise, LaSPA retrieves the Top-k services by assessing the similarities between the requirements and available services in the repository. Then, LaSPA transforms the service recommendations into multiple-choice questions for accurate recommendations that rely on the semantic understanding of services with the help of LLMs. Finally, LaSPA utilizes the reasoning ability of LLMs to orchestrate the recommended services for the service process automation. Our observations suggest that LaSPA outperforms baseline methods. Compared to the recommendations approach, LaSPA can more easily cope with the evolution of services. The evolution experiment results indicate a 35.9% improvement in recall@5 and a 76.6% increment in precision@5, on average. Furthermore, a case evaluation shows the possibility of automating service processes.