Industry Scale Semi-Supervised Learning for Natural Language Understanding

TL;DR

This paper develops a production-level semi-supervised learning pipeline for natural language understanding, exploring data selection strategies and comparing multiple SSL techniques to enhance intent classification and named entity recognition.

Contribution

It introduces a comprehensive SSL pipeline for NLU, evaluates data selection methods, and compares four SSL techniques in a production setting for large-scale NLU tasks.

Findings

Data selection significantly impacts SSL effectiveness.

Different SSL techniques have varying benefits for NLU tasks.

Guidelines are provided for choosing SSL methods in large-scale NLU.

Abstract

This paper presents a production Semi-Supervised Learning (SSL) pipeline based on the student-teacher framework, which leverages millions of unlabeled examples to improve Natural Language Understanding (NLU) tasks. We investigate two questions related to the use of unlabeled data in production SSL context: 1) how to select samples from a huge unlabeled data pool that are beneficial for SSL training, and 2) how do the selected data affect the performance of different state-of-the-art SSL techniques. We compare four widely used SSL techniques, Pseudo-Label (PL), Knowledge Distillation (KD), Virtual Adversarial Training (VAT) and Cross-View Training (CVT) in conjunction with two data selection methods including committee-based selection and submodular optimization based selection. We further examine the benefits and drawbacks of these techniques when applied to intent classification (IC) and named entity recognition (NER) tasks, and provide guidelines specifying when each of these methods might be beneficial to improve large scale NLU systems.