1. Insights
  2. AI Data
  3. Case Study

Training a recommendation engine through image classification

When our client required classification of 500 images within an exceptionally tight timeframe, they turned to TELUS International's AI Data Solutions for our team's proven ability to deliver high-quality annotation quickly and at scale.

Monitor floating on digital background
  • Share on X
  • Share on Facebook
  • Share via email

The challenge

Our client, a global leader in consumer devices, ecommerce, music and more, required defined segmentation, text caption generation and imagery classification for 500 images to train their recommendation engine.

Recommendation engines use machine learning to predict the likelihood that a user will prefer a particular item or service. They deliver personalized suggestions based on a user’s previous actions and the actions of similar users to help them find more of what they like, or relevant alternatives.

The project images to be classified contained foods, beverages and related items like utensils, plates and containers. A single image often contained multiple food types that needed to be identified, such as eggs, toast and bacon on a plate.

The client turned to TELUS International’s AI Data Solutions and our team’s proven ability to deliver high-quality annotation at scale within an exceptionally tight timeframe.

The TELUS International solution

For image classification purposes, the team of annotators used the interactive segmentation feature in our proprietary computer vision platform, Ground Truth Studios. This reduced manual effort while annotating the first group of images. The platform was then able to use the trained dataset to automate the labeling and classification processes for the duration of the project.

The results

TELUS International was able to deliver the proof of concept phase of the project over an accelerated 10-day period. Quality assurance was done using a combination of automated processes and a team of dedicated experts to deliver quality scores of 96 - 97%. By implementing our automation and quality features, we were able to realize an effort reduction metric of 31%.

Check out our solutions

Test and improve your machine learning models via our global AI Community of 1 million+ annotators and linguists.

Learn more