ESR 1: Scaling Up Citizen Development with Recommender ChatbotsLissette Almonte Garcia
Universidad Autόnoma de Madrid (Spain)
The users of LCDPs (so called citizen developers) frequently lack a technical profile. This is a practical factor that hinders the applicability of LCDPs to create complex systems.
In order to support citizen developers to create applications beyond toy apps, we propose the concept of software development chatbots. We envision that these chatbots will be addressed in natural language to issue queries on how to achieve some goals (“I want the application to do X, Y and Z”), or how to perform some task within the current project (“How do I make the app to send an e-mail to all registered customers?”). The chatbots will include a query answering component, and will provide example fragments and templates. Such fragments will be extracted from repositories of existing application descriptions, using information retrieval (IR) techniques. The chatbots will be proactive as well, suggesting artefacts specifically designed for LCEPs and IDEs. For this purpose, chatbots will use conversational recommendation techniques that will exploit preferences of the target user and like-minded users, artefact attributes, and contextual (action-based) data.
The use of bots has been identified as a possible disruptive technology in software engineering, with high potential to improve developer performance through automation and natural interaction. Developers use bots, e.g., to automate deployment tasks, schedule tasks like sending reminders, integrate communication channels, or for customer support. Bots have also been proposed to access API documentation, to analyse software projects, or to assist in modelling activities using natural language (by our team). However, a system to build chatbots for domain-specific artefact recommendation – able to process queries in natural language and use information retrieval and machine learning techniques – is novel.
This project will develop novel concepts to create systems that combine recommendation, information retrieval and query answering for specific domains and platforms. The systems will be able to scale for recommendations in repositories of millions of artefacts, and will be embeddable in platforms like Lowcomotive, and social networks like Slack or Telegram. We target at empowering citizen developers to create more complex apps, and in these scenarios, we will target at improvements in development times in the order of 30%.