Case Study: How The Times of India Brings Real-Time Personalization to 1,500+ Daily News Stories


Published October 29, 2024

AI in the Newsroom is a series of case studies from the Online News Association (ONA) that highlight specific ways journalists are building and using AI tools.

The Times of India built Signals—an in-house recommendation system that personalizes content based on user preferences, recent news trends, and engagement data. This case study outlines their experience creating a scalable, adaptable, and audience-centered personalization model that has transformed readers’ interactions with news content.

Opportunity

As one of India’s largest news organizations, The Times of India faced a critical challenge in the digital age: maintaining reader engagement while publishing over 1,500 stories daily. With declining SEO-driven traffic and limited subscription potential, the newsroom needed a solution to retain audience attention and differentiate its platform. The organization identified an opportunity to create a dynamic personalization system to improve content discovery, enhance ad targeting, and balance multiple revenue streams without overwhelming readers.

Solution

The Times of India built Signals, an advanced recommendation system that revolutionized their approach to content distribution.

At its core, Signals employs a collaborative filtering model that moves beyond traditional content categorization. Rather than relying on static content labels or tags, the system analyzes user behavior across the platform in real time, creating dynamic clusters based on consumption patterns. This approach captures nuanced relationships between content and user preferences that might not be apparent through conventional categorization methods.

Understanding the rapid pace of news cycles, they implemented a unique ‘forget’ mechanism within Signals. This innovative feature regularly resets user preferences to reflect current behaviors and breaking news interests, allowing content recommendations to evolve with the news cycle. The mechanism prevents readers from being confined to past topics while ensuring recommendations stay relevant to ongoing events and changing interests.

The technical foundation of Signals required building a custom, real-time data pipeline after the team discovered that traditional analytics tools, like Google Analytics, couldn’t meet their needs. This proprietary infrastructure enabled low-latency data processing, collecting comprehensive clickstream data such as time on page, scroll depth, device type, and user location. This information feeds into the model instantaneously, allowing for real-time recommendation updates.

In terms of user experience, Signals marked a significant departure from traditional news website design. The team transformed their widget-based layout into an algorithm-driven continuous feed, mimicking the familiar dynamics of social media platforms. This redesign removed the constraints of traditional widget boxes, allowing for more flexible content presentation while collecting granular engagement data, including user pauses and scroll behaviors.

The editorial judgment remains central to Signals’ operation. The system carefully balances algorithmic recommendations with editorial oversight to prevent filter bubbles. Editors maintain control over key placements, particularly the top five articles, while the system ensures a healthy mix of user preferences, trending stories, and editorially selected content.

For revenue optimization, they implemented an innovative ‘mutual fund’ approach to content monetization. This unique system treats different revenue types—advertisements, affiliate links, and subscriptions—as distinct asset classes. For each user, Signals calculates optimal risk-reward ratios, tailoring content recommendations to maximize engagement and revenue without overwhelming users with irrelevant content.

The system also adapts to the natural rhythms of news production. Understanding that content flow fluctuates throughout the day—with high volume during morning print edition releases and slower periods in the afternoon—Signals automatically adjusts its recommendations. During low-volume periods, the system intelligently surfaces relevant archived content, maintaining engagement even when new content is limited.

Results so far

According to Ritvvij Parrikh, Senior Director of Product at Times Internet, The Times of India has achieved an 85% improvement in website click-through rates and an over 30% increase in app engagement since implementing Signals.

The system has particularly excelled at recirculating valuable archived content, with 50% of personalized recommendation views coming from stories over two days old.

The sophisticated infrastructure underlying Signals has also enabled insight into user behavior. The team can now measure engagement with remarkable precision, tracking not just basic metrics but detailed interaction patterns. The system weighs content based on intentional user actions rather than passive views, reducing position bias and ensuring that articles users actively seek to receive appropriate priority.

Lessons learned

The Times of India’s development and implementation of Signals offers numerous insights into the complexities of building a personalization engine for news.

Real-time adaptability is essential in news delivery, unlike e-commerce or social platforms where user interests remain relatively stable. The system’s ‘forget’ mechanism ensures recommendations stay relevant to the rapidly changing news cycles.

Custom infrastructure investment is fundamental, not optional. Traditional analytics tools couldn’t support real-time personalization needs, making proprietary development necessary despite resource requirements. This foundation enables the sophisticated AI applications that power Signals.

Cross-functional collaboration drives success. Signals’ effectiveness relies heavily on cooperation between editorial, product, and engineering teams. Editorial guidance proved crucial in developing moderation rules, filtering outdated content, and identifying evergreen stories. This cross-departmental support enables the system to maintain high editorial standards while pushing technical boundaries.

Keep learning with TOI

For more insights into The Times of India’s approach, listen to the Newsroom Robots podcast episode with Ritvvij Parrikh, where he shares an in-depth look at building Signals.

This resource is part of the AI in the Newsroom series. Read other case studies you might have missed:

 

Related Topics
Author
Nikita Roy
ICFJ Knight Fellow and Newsroom Robots Podcast Host