Build a Q&A Bot with DeepLearning4J
summary
Builds a Java Q&A bot using DeepLearning4J with an autoencoder plus classifier architecture: 1,024‑dim input layers, ~2,500‑word vocabulary, compressed latent code fed to a softmax output. Training runs many epochs (CPU or GPU) and yields about 85–86% accuracy on the FAQ dataset. Produces a saved model used from a Spring Boot servlet and Microsoft Bot Framework front end; code and CSV datasets are on GitHub, ready to reproduce. automatic encoder
AI-written summary · about
chapters (13)
- 0:00Start
- 6:58Neural network architecture
- 8:55What happens inside a neuron
- 10:23The role of activation functions
- 18:53Gradient descent is not perfect!
- 24:25Building and training a FAQ model
- 25:18Fingerprint the data with an auto-encoder
- 27:26Relate the fingerprint to an answer
- 34:06Encoding text as a bag of words
- 36:01Encode answers
- 36:42Map neurons to answers
- 40:44Answering a question
- 42:12You too can use deep learning
related talks
01h 01mDeep Learning and Java
44mDeep Learning with DeepLearning4J and Spring Boot
31mMachine Learning-Enabled Chatbots
33mBuilding the next-generation Conversational AI with Python and Deep Learning
50mBuilding Deep Neural Networks using Deep Learning for Java(DL4J)
01h 51mDeep Language Modeling for Question Answering using Keras
52mBuild Smarter Knowledge Management Systems with AI Technology
39mTackling the Limits of Deep Learning