18491877. Method for Training Large Language Models to Perform Query Intent Classification simplified abstract (GOOGLE LLC)

Method for Training Large Language Models to Perform Query Intent Classification

Organization Name

GOOGLE LLC

Inventor(s)

Krishna Pragash Srinivasan of Union City CA (US)

Michael Bendersky of Cupertino CA (US)

Anupam Samanta of Mountain View CA (US)

Lingrui Liao of Foster City CA (US)

Luca Bertelli of Redwood City CA (US)

Ming-Wei Chang of Redmond WA (US)

Iftekhar Naim of San Jose CA (US)

Siddhartha Brahma of San Jose CA (US)

Siamak Shakeri of New York NY (US)

Hongkun Yu of Redwood City CA (US)

John Nham of Fremont CA (US)

Karthik Raman of Sunnyvale CA (US)

Raphael Dominik Hoffmann of Los Altos CA (US)

Method for Training Large Language Models to Perform Query Intent Classification - A simplified explanation of the abstract

This abstract first appeared for US patent application 18491877 titled 'Method for Training Large Language Models to Perform Query Intent Classification

Simplified Explanation

The patent application describes a method for training query processing models using retrieval augmentation and multi-stage distillation.

• Unlabeled training examples of queries are obtained.
• A set of training examples is augmented with additional feature annotations to generate augmented training examples.
• A first query processing model annotates the retrieval augmented queries to generate inferred labels for the augmented training examples.
• A second query processing model is trained on the inferred labels, distilling the retrieval augmented model into a non-retrieval augmented model.
• The second model annotates the entire set of unlabeled training examples.
• Another stage of distillation trains a third query processing model using the entire set of unlabeled training examples without retrieval augmentation.

---

1. 1. 1. Potential Applications

• Natural language processing
• Information retrieval systems

1. 1. 1. Problems Solved

• Improving query intent classification accuracy
• Enhancing query processing models

1. 1. 1. Benefits

• Better performance of query processing models
• Increased efficiency in information retrieval tasks

1. 1. 1. Potential Commercial Applications
         1. Enhanced Query Processing Models for Information Retrieval Systems

1. 1. 1. Possible Prior Art

No prior art is known at this time.

---

1. 1. 1. 1. Unanswered Questions

1. 1. 1. How does the method handle noisy or irrelevant training examples?

The method does not specifically address how noisy or irrelevant training examples are filtered or handled during the training process.

1. 1. 1. What is the computational cost associated with training multiple query processing models in a multi-stage distillation process?

The patent application does not provide information on the computational resources required for training multiple models in the proposed method.

Original Abstract Submitted

Provided are computing systems, methods, and platforms that train query processing models, such as large language models, to perform query intent classification tasks by using retrieval augmentation and multi-stage distillation. Unlabeled training examples of queries may be obtained, and a set of the training examples may be augmented with additional feature annotations to generate augmented training examples. A first query processing model may annotate the retrieval augmented queries to generate inferred labels for the augmented training examples. A second query processing model may be trained on the inferred labels, distilling the query processing model that was trained with retrieval augmentation into a non-retrieval augmented query processing model. The second query processing model may annotate the entire set of unlabeled training examples. Another stage of distillation may train a third query processing model using the entire set of unlabeled training examples without retrieval augmentation.