Efficient Natural Language Response Suggestion for Smart Reply

Abstract

This paper presents a computationally efficient machine-learned method for natural language response suggestion. Feed-forward neural networks using n-gram embedding features encode messages into vectors which are optimized to give message-response pairs a high dot-product value. An optimized search finds response suggestions. The method is evaluated in a large-scale commercial e-mail application, Inbox by Gmail. Compared to a sequence-to-sequence approach, the new system achieves the same quality at a small fraction of the computational requirements and latency.

References in corpus (7)

• Sequence to Sequence Learning with Neural Networks
• Asymmetric LSH (ALSH) for Sublinear Time Maximum Inner Product Search (MIPS)
• A Network-based End-to-End Trainable Task-oriented Dialogue System
• Smart Reply: Automated Response Suggestion for Email
• End-to-end LSTM-based dialog control optimized with supervised and reinforcement learning
• Swivel: Improving Embeddings by Noticing What's Missing
• Conversational Contextual Cues: The Case of Personalization and History for Response Ranking

Cited by in corpus (52)

• Wizard of Wikipedia: Knowledge-Powered Conversational agents
• Multi-Stage Document Ranking with BERT
• Information Retrieval: Recent Advances and Beyond
• Social Media Polarization and Echo Chambers in the Context of COVID-19: Case Study
• Semantic Models for the First-stage Retrieval: A Comprehensive Review
• Pretrained Transformers for Text Ranking: BERT and Beyond
• DeCLUTR: Deep Contrastive Learning for Unsupervised Textual Representations
• RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering
• End-to-End Retrieval in Continuous Space
• Training Neural Response Selection for Task-Oriented Dialogue Systems
• TSDAE: Using Transformer-based Sequential Denoising Auto-Encoder for Unsupervised Sentence Embedding Learning
• SimCSE: Simple Contrastive Learning of Sentence Embeddings
• DiPair: Fast and Accurate Distillation for Trillion-Scale Text Matching and Pair Modeling
• Neural Passage Retrieval with Improved Negative Contrast
• Neural Code Search Revisited: Enhancing Code Snippet Retrieval through Natural Language Intent
• Information Leakage in Embedding Models
• Who Wrote this? How Smart Replies Impact Language and Agency in the Workplace
• OCC: A Smart Reply System for Efficient In-App Communications
• Unsupervised Contextual Paraphrase Generation using Lexical Control and Reinforcement Learning
• LAReQA: Language-agnostic answer retrieval from a multilingual pool
• Diversifying Reply Suggestions using a Matching-Conditional Variational Autoencoder
• A Repository of Conversational Datasets
• CheerBots: Chatbots toward Empathy and Emotionusing Reinforcement Learning
• PRANK: motion Prediction based on RANKing
• LightningDOT: Pre-training Visual-Semantic Embeddings for Real-Time Image-Text Retrieval
• Self-Supervised Contrastive BERT Fine-tuning for Fusion-based Reviewed-Item Retrieval
• AspectCSE: Sentence Embeddings for Aspect-based Semantic Textual Similarity Using Contrastive Learning and Structured Knowledge
• Learning Dense Representations of Phrases at Scale
• Learning Federated Representations and Recommendations with Limited Negatives
• Preserving Semantics in Textual Adversarial Attacks
• VacancySBERT: the approach for representation of titles and skills for semantic similarity search in the recruitment domain
• Towards Personalized and Semantic Retrieval: An End-to-End Solution for E-commerce Search via Embedding Learning
• Time to Take Emoji Seriously: They Vastly Improve Casual Conversational Models
• Investigating Societal Biases in a Poetry Composition System
• Instance-based Transfer Learning for Multilingual Deep Retrieval
• Lucene for Approximate Nearest-Neighbors Search on Arbitrary Dense Vectors
• Understanding Chat Messages for Sticker Recommendation in Messaging Apps
• Beyond "How may I help you?": Assisting Customer Service Agents with Proactive Responses
• Neural Retrieval for Question Answering with Cross-Attention Supervised Data Augmentation
• Toward Interpretability of Dual-Encoder Models for Dialogue Response Suggestions
• DialogueCSE: Dialogue-based Contrastive Learning of Sentence Embeddings
• Dialogue Distillation: Open-Domain Dialogue Augmentation Using Unpaired Data
• Measuring the Quality of Answers in Political Q&As with Large Language Models
• A Dataset and Baselines for Multilingual Reply Suggestion
• Language Scaling for Universal Suggested Replies Model
• Exploring Low-Cost Transformer Model Compression for Large-Scale Commercial Reply Suggestions
• Measuring a Texts Fairness Dimensions Using Machine Learning Based on Social Psychological Factors
• A Conditional Generative Matching Model for Multi-lingual Reply Suggestion
• COUGH: A Challenge Dataset and Models for COVID-19 FAQ Retrieval
• Exploiting Twitter as Source of Large Corpora of Weakly Similar Pairs for Semantic Sentence Embeddings
• Put Chatbot into Its Interlocutor's Shoes: New Framework to Learn Chatbot Responding with Intention
• Generating multiple-choice questions for medical question answering with distractors and cue-masking