Neural Ned Bøt

Inspiration

1. The Great Commission
2. The global need for conversations about Jesus
3. The online opportunity to engage with people through conversation.
4. The thousands of conversations that have already been had about Jesus online.
5. The hundreds of salvations because of those conversations.

What it does

Stage 1: Understand conversations about Jesus, thanks to the population of the Evangelism Corpus with annotated ‘Got Questions’ data and the creation of a baseline NLU model.

How we built it

1. Identified intents & entities in a spiritual conversation
2. Labeled ‘Got Questions’ data with intents & entities
3. Labeled ‘Got Questions’ data with Bible verses
4. Translated labeled data into Corpus format
5. Created training & test sets for NLU
6. Trained & tested NLU

Challenges we ran into

• Labeling data is boring
• Difficult to parse JSON into proper format
• Training the NLU takes a long time

Accomplishments that we're proud of

• We successfully labeled over 200 GotQuestions messages and responses
• We were able to create a couple of semi-working NLUs in the short time that we had

What we learned

• Creating a NLU that works well takes a very long time
• Time flies

What's next for Neural Ned Bøt

1. Develop annotation feedback loop to allow for faster annotation process
2. Integrate NLU into Intelligent Conversation Engine
3. Promote Evangelism Corpus to allow for other use cases

Built With

• jupyter-notebook