Dataset and Sources for the Paper: Are the Tides Turning Against Microservices? Tracking the Sentiments Towards Microservice Architectures in Developer Communities

This repository contains the data set and source codes for the Paper

Are the Tides Turning Against Microservices? Tracking the Sentiments Towards Microservice Architectures in Developer Communities

Summary:This study investigates whether the decline in microservice adoption reported by individual larger companies is also reflected in developer communities, with a particular focus on their associated challenges. We analyzed 20,000 HackerNews posts from the past decade and applied a selection of language and transformer models to classify the sentiments toward microservices in those discussions. To ensure reliable results, we evaluated several state-of-the-art models using our dataset and selected the model with the best performance for sentiment classification. While older approaches such as RoBERTa and EASTER struggled with three-class sentiment analysis, GPT-5.4 achieved the best results, reaching an unweighted F1 score of 73%, slightly outperforming its predecessor GPT-4.1, which nevertheless yielded a better result than GPT-5-mini. Our analysis revealed a steady rise in negative sentiments, peaking between 2021 and 2023 and continuing to outweigh positive emotions. We attribute this trend partly to the rapid growth of developer teams during the pandemic, which introduced more developers to complex distributed microservice architectures. Additionally, developers faced numerous challenges specific to microservices, such as deploying and orchestrating services with Kubernetes, and coordinating distributed database access.

本仓库收录了论文《微服务的浪潮是否正在逆转？追踪开发者社区中针对微服务架构（Microservice Architectures）的情感倾向》的数据集与源代码。

摘要：本研究旨在探究大型企业所报告的微服务（Microservices）采纳率下滑现象，是否同样在开发者社区中有所体现，并重点聚焦其关联挑战。我们分析了近十年间的20000条HackerNews平台帖子，并运用多款语言模型与Transformer（Transformer）模型，对这些讨论中针对微服务架构的情感倾向进行分类。为保障研究结果的可靠性，我们基于本数据集对多款最先进模型展开评估，并选取情感分类表现最优的模型。
尽管RoBERTa与EASTER等传统方法在三分类情感分析任务中表现欠佳，但GPT-5.4取得了最佳效果，其未加权F1值达到73%，略优于其前代模型GPT-4.1；而GPT-4.1的表现仍优于GPT-5-mini。
我们的分析显示，负面情感呈稳步上升趋势，在2021至2023年间达到峰值，且负面情感占比持续高于正面情感。我们将这一趋势部分归因于疫情期间开发者团队的快速扩张，这使得更多开发者接触到复杂的分布式微服务架构。此外，开发者还面临诸多微服务特有的挑战，例如使用Kubernetes部署与编排服务，以及协调分布式数据库访问。