Towards Automatic Classification of Design Decisions from Developer Conversations

Documentation of architecture design decisions can be cumbersome, though it is critical to ongoing software maintenance and integration of new features. In practice, developers often discuss design decisions in online communication tools, such as Slack, Microsoft Teams or Gitter. In this paper, we introduce a design decision bot for Slack to reduce the workload for documenting design decisions. The bot heralds the integration of machine learning and language processing to decision recording. Using available communication tools and interactions (Slack and Slack bots) allows developers to document design decisions mid-stride without diverting attention from other development tasks. The bot utilises a "transformer" for decision classification, avoiding rigid and onerous interactions between the bot and developers. The transformer builds on Bi-directional Encoder Representations from Transformers (BERT), a machine learning technique for Natural Language Processing (NLP). In this paper, we present the theoretical concepts of the bot with early implementation details and the results of preliminary evaluations.