Tickets

Loading…

Aditya Kiran

Improving search relevance in hyperlocal food delivery using (small) language models

Submitted Jun 3, 2024

Introduction

The ability to accurately understand and serve customer search queries is critical to Swiggy. This need is amplified in food delivery platforms operating in India due to the wide variety of languages, cuisines and tastes. Our platform alone offers millions of items from hundreds of thousands of restaurants across India. Not only do Indian dish names have a tremendous amount of regional variety, international cuisine served in India is typically customised to the Indian palate. This variety is reflected in search queries where these can be a combination of words including dish names, dish categories, cuisine, preparation style, occasions, dietary preferences, just to name a few. In Indian food delivery space, often the queries are an amalgamation of multiple intents related to dish preference, dietary preference, time slot of the day and the customer’s latent mood of food preference at that point of search session within the hyperlocal space. Given the diversity of dishes across multiple restaurants in the country, language models play a huge role in understanding the relevancy between search query and the corresponding dish names suitable for the query.

This talk will focus on our journey in leveraging the language models for improving the search relevance.

Target audience

The primary audience for this talk includes the data science practioners interested in leveraging language models and fine-tuning them for search use cases to improve the overall relevance.

Outline

  1. Era of language models evolution
  2. Understanding the customer search queries
  3. Data curation and strategies for handling unseen queries
  4. Supervised and unsupervised fine-tuning approaches
  5. Evaluation and error analysis

Impact

This work mainly includes to improve the relevance of broader intent search queries in our platform. We primarily focus on

  1. Search relevance metrics
  2. Latency and inference pipeline of LM based retrieval

Attaching the reference blog
https://bytes.swiggy.com/improving-search-relevance-in-hyperlocal-food-delivery-using-small-language-models-ecda2acc24e6

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hybrid Access Ticket

Hosted by

All about data science and machine learning

Supported by

Gold Sponsor

Atlassian unleashes the potential of every team. Our agile & DevOps, IT service management and work management software helps teams organize, discuss, and compl

Silver Sponsor