Day #143 - Chatbot Reading and Implementation Notes

Reading #1 - Link1
Key Point #1 - Closed Domain Based Chatbot

Key Point #1 - NLP Based Intent / Entity / Response System

• Intent
• Entity
• Response / Action

Key Point #2 - Knowledge Based Question and Answering

• Convert Query into Database Query
• User Query to Structured Database Query

Paper #1 - A Survey of Available Corpora for Building
Data-Driven Dialogue Systems

Datasets - The Second Dialog State Tracking Challenge - DSTC1 (Williams et al., 2013)
The ATIS Spoken Language Systems Pilot Corpus

Summary of Key Points

• For Dialogue based systems Speech Recognition, State Tracker, Response Tracker, Speech Synthesizer are different components
• Goal Driven Dialogue Systems - ML to identify intention / heavily hand crafted rules
• Learning Dialogue System Components - Probabilistic Models, Discriminative tasks, Using Supervised learning
• Dialogue State Tracker - K Dialogue state based tracker
• Ranking based systems, Encoding word by word, P(Action / Dialogue Tracker)
• Different corpus information and details

Paper #2 - MACHINE LEARNING FOR DIALOG STATE TRACKING

• Discriminative methods exploiting ML to implement Sequence to Sequence models
• Estimate Likelihood, Prior probability
• n-gram based/ word based Dialogue State Tracker
• Rule Based / RNN based

Paper #3 - Smart Reply: Automated Response Suggestion for Email 

• LSTM Neural network to process and predict likely response
• Semi-Supervised learning approach
• Feed forward network suggests whether or not to suggest
• Semantic Intent Clustering (Partition responses into semantic clusters)
• Key tasks - Tokenization / Normalization / Salutation removal / Remove infrequent words (Personal names, Phone numbers)

Paper #4 - AliMe Chat: A Sequence to Sequence and Rerank based Chatbot Engine

• Identify Intent
• Check Knowledge Base
• Answer Generation
• Attentive Seq2Seq model

Paper #5 - Evaluating Natural Language Understanding Services for Conversational Question Answering Systems

• Request text analysis
• Determine Contextual Information
• Response Message Generation from Knowledge Base

Chatbot Product Ideas

• Based on the person, Gender age assign appropriate Style, Tone, Personality
• Localized chatbot implementation
• Customized Response Generation based on Knowledge base and contextual information
• Leverage Data Store (Interaction History) / Knowledge Base (Content Management System)
• SMS-based chatbot

Technical Ideas

• Named Entity Recognition
• Sentiment Analysis
• Part of Speech Tagging
• Dependency Parsing

Projects - Deep Learning for Chatbots, Part 2 – Implementing a Retrieval-Based Model in Tensorflow, chatbot-retrieval

Happy Learning!!!