Documentation artifacts for conversational SRS in chatbots: a systematic review and a new meta-model proposal and validation

Context: Chatbots are complex applications due to their capacity to engage and maintain a conversation with humans. However, the conversational-related requirements of chatbots are hard to elicit, document, and test. Another challenge is the documentation, since there are not so many directions on how to register and test subjective requirements. 

Methods: We followed systematic literature review (SLR) guidelines and identified 42 studies that address the artifacts used by practitioners to document conversational-related requirements in literature. We also investigated what conversational requirements are addressed in requirements documentation.

Results: The main results indicate that UML diagrams, prototypes, tables of requirements, conversational flows, and scenarios are present in most chatbot documentation. Except for UML diagrams, those artifacts are used to document standard requirements or conversational requirements. In those artifacts, context-dependent behavior, assertivity, error handling, and human-like attitude are the most approached conversational requirements in the studies. In sequence, based on our findings, we proposed the conversational integrated map and validated it by conducting a 2-step questionnaire among software practitioners experience in requirements engineering and chatbot requirement's specification.

Conclusion: Future studies should investigate if existing artifacts are enough to address all complex aspects of chatbots' specific conversational requirements or require further adaptation. Future studies should investigate specific SRS needs for different types of softwares.