DeepCare Chatbot - Generating answers to customers using Deep Learning and NLP
summary
Prototype deep-care chatbot generates replies to customer tweets using a rule filter plus a neural generator, trained from an AT&T scrape of 1.3 million tweets reduced to 100,000 collected examples and then to 15,000 training examples after preprocessing. Pipeline uses TensorFlow seq2seq with spaCy features (lemmas, POS, dependency, negation) and a small web demo; results sometimes produce plausible social‑selling replies but still make grammar and logic errors and misattribute products.
AI-written summary · about
chapters (10)
- 0:00DeepCare Chatbot intro
- 3:29What is a chatbot
- 6:00Twitter data selection
- 8:06Two-stage model approach
- 13:13Pre-processing techniques
- 17:48NLP rules for answers
- 23:25Seq2Seq training
- 27:45TensorFlow implementation
- 29:12Conclusions and limits
- 30:07Q&A on hyperparameters
related talks
32mChatBot Learning Assistant with Large Language Models
41mConversational Agents at Scale: Retrieval and Generative approaches
58mA Primer on AI Powered Conversational Experiences and Language Models
01h 03mIt's All About AI
13mExploring AI-Powered Chatbots and Their Transformative Role
24mSmart Reply: Learning a Model of Conversation from Data
01h 21mCreating Ethical Generative AI for Enterprises
19mThe Future of AI