Skip to content

Latest commit

 

History

History
95 lines (59 loc) · 3.65 KB

README.md

File metadata and controls

95 lines (59 loc) · 3.65 KB
Raw

Bugbot - Designing a Chatbot to Elicit Contextual Information of Software Requirements

While user feedback has become a valuable resource to elicit system requirements and identify issues, a common lack of contextual information in feedback challenges development teams. Previous research has primarily focused on text analysis through natural language processing to identify topics or trends in user feedback and capture requirements by summarizing the most common user subjects. However, such approaches fail to improve the general quality of collected user feedback and the amount of contextual information included. This chatbot prototype uses Rasa Open Source to elicit contextual information by interacting with users during a conversation.

The chatbot was developed for a german streaming application. Therefore it currently only supports conversations in german.

👷‍ Installation

To install please first install rasa and its dependencies on your system following the official documentation.

The chatbot also uses a knowledge base to infer information using the elastic stack. If you want to use this functionality you need to run a separate elastic container as described in the development section.

🤖 How to run

After cloning the repo run Use rasa train in the project directory to train a model a new model.

Then to run custom action set up your action server in a separate terminal window running

rasa run actions --actions actions.actions

There are some cusotm actions that require connections to external services specifically ValidatePlaybackIssueForm and SubmitIncidentAction. To run these you need to setup your own elastic stack or use a database to connect to. See the development section for more instructions.

To run the Bot use (the debug flag is optional):

docker run -p 8000:8000 rasa/duckling
rasa shell --debug

To run the Bot under http://localhost:5005. Use html/dev/index.html for a simple webpage to access the bot.

rasa run -m models --enable-api --cors "*" --debug

👩‍💻 Project Structure

data/stories/ - contains stories

data/rules/ - contains rules

data/nlu/ - contains nlu training data

data/nlu/responses - contains chatbot responses

html/dev - contains a webinterface to interact with the chatbot run http://localhost:5005

domain.yml - the domain file, including bot response templates

config.yml - training configurations for the NLU pipeline and policy ensemble

actions - contains custom action code

Development

To run custom actions locally, put a file called .env in the root of your local directory with values for the following environment variables. While the chatbot works generally without them it is recommended to setup up the external services to see the full functionality.

In our development environment the elastic stack is hidden behind a proxy. To use your own elastic stack either follow the same approach or review and change actions/elastic/ to match your requirements. This is WIP.

We save our incident reports using Airtable. To connect to airtable review their documentation and actions/util/create_incident_report. The incident table used needs to include all fields of create_incident_report.

ELASTIC_HOST=#Domain of the elastic host
ELASTIC_ADMIN=#Elastic login
ELASTIC_PSWD=#Elastic password

AIRTABLE_BASE_ID=#Base ID of your airtable project
AIRTABLE_API_KEY=#API key of your airtable account
AIRTABLE_TABLE=#Table name

Credits

This project is part of my master thesis at the Universität Hamburg.