Source URL: https://cloud.google.com/blog/topics/customers/glance-builds-gemini-powered-knowledge-graph-with-google-cloud/
Source: Cloud Blog
Title: From news to insights: Glance leverages Google Cloud to build a Gemini-powered Content Knowledge Graph (CKG)
Feedly Summary: In today’s hyperconnected world, delivering personalized content at scale requires more than just aggregating information – it demands deep understanding of context, relationships, and user preferences. Glance, a leading content discovery platform that delivers personalized, real-time content experiences on mobile lock screens across the globe, serves over 300 million users worldwide across 450 million devices. Beyond news aggregation, Glance curates diverse content including entertainment, sports, gaming, shopping, and lifestyle content, making every glance at the phone screen meaningful and engaging.
However, with the exponential growth of digital content, particularly the overwhelming volume of daily news articles, Glance faced a critical challenge: how to effectively navigate this information while maintaining the personalized, contextual experiences users expect. The existing search and content discovery capabilities needed significant enhancement to uncover emerging trends, improve search relevance, provide deeper context, and most importantly, deliver truly personalized content recommendations that resonate with individual user preferences and behaviors.
Glance partnered with Google Cloud Consulting team to build a sophisticated Content Knowledge Graph (CKG) that addresses these challenges head-on. This solution leverages Google Cloud’s advanced AI and data processing capabilities, including Gemini models, BigQuery, Vertex AI, and Google Cloud partner Neo4j, to ingest, process, extract, standardize, classify, and structure news data into a dynamic network of interconnected entities and relationships. The Content Knowledge Graph has dramatically improved search relevance, enhanced personalized content discovery, provided deeper contextual insights, increased user engagement, and improved scalability and efficiency.
aside_block
Tackling the content discovery and personalization challenge
Glance’s mission extends far beyond traditional content aggregation. As a platform serving hundreds of millions of users globally, Glance must deliver hyper-personalized content experiences that anticipate user interests and adapt to evolving preferences in real-time. The challenge was multifaceted:
Uncovering trends at scale – Identifying emerging topics, viral content, and the complex relationships between different content categories across news, entertainment, sports, and lifestyle domains
Enhancing personalized search – Improving the accuracy and relevance of search results based on individual user behavior patterns, preferences, and contextual signals
Providing intelligent context – Offering users deeper understanding not just of individual pieces of content, but how different stories, events, and topics connect across the broader content ecosystem
Scaling personalization – Delivering these sophisticated experiences across 300+ million users while maintaining real-time responsiveness and relevance
Manually analyzing and connecting the dots within this vast, multi-dimensional content landscape was simply not scalable. Glance needed an intelligent, automated approach that could understand content at both granular and contextual levels while powering personalized recommendations at unprecedented scale.
Building a Gemini-powered content intelligence engine
Glance and Google Cloud Consulting team collaborated to architect and implement a comprehensive Content Knowledge Graph that transforms how content is understood, connected, and delivered. This sophisticated system leverages the full spectrum of Google Cloud’s AI and data processing capabilities:
Architecture of Content Processing for Knowledge Graph Creation
Intelligent content ingestion and processing – The system ingests content from diverse sources beyond news, including entertainment articles, sports updates, lifestyle content, and trending topics, storing them in BigQuery for efficient, scalable processing.
Advanced entity extraction and relationship mapping – Using Gemini foundational models, the system extracts key entities (people, organizations, locations, events, brands) and identifies complex relationships between them across different content categories.
Entity standardization and knowledge linking – Extracted entities are normalized using Gemini’s advanced language understanding and linked to authoritative knowledge sources like Wikipedia, ensuring consistency and enabling sophisticated cross-content analysis.
Multi-dimensional content classification – Gemini foundation models classify content into granular categories using the IAB content taxonomy while also identifying sentiment, urgency, and relevance signals for personalization.
Intelligent content summarization and tagging – Gemini generates contextual tags, compelling short headlines, and category labels, enabling users to quickly grasp content essence while powering recommendation algorithms.
Dynamic Knowledge Graph construction – The extracted information is structured into a Neo4j graph database, creating a living, breathing network of interconnected entities, topics, relationships, and user interaction patterns.
Real-time trend analysis and prediction – The system integrates with external trend APIs and analyzes user engagement patterns to identify and predict trending topics, providing actionable insights for content curation.
Interactive analytics dashboard – NeoDash powers an interactive dashboard for monitoring trending content, analyzing entity relationships, and visualizing content performance across different user segments.
Diagram 2:
How entities extracted from one news article help identify related news articles
Engineering for Global Scale and Performance
Glance enhanced the Content Knowledge Graph solution to handle 50,000+ daily articles across multiple content categories. The engineering team implemented several critical optimizations:
Event-driven architecture transformation – Migrating to a Kafka-based event-driven architecture with intelligent retries and asynchronous operations resulted in 4x throughput improvement, enabling real-time content processing at massive scale.
Graph database optimization – Neo4j query optimization and indexing strategies drastically reduced query response times from seconds to milliseconds, enabling real-time content recommendations.
Kubernetes-native deployment – Moving to managed Google Kubernetes Engine (GKE) with auto-scaling capabilities improved system reliability and resource utilization.
Strategic performance optimization – Applying the 80-20 principle to focus on high-impact optimizations, coupled with Redis caching and Cloud Spanner for critical data, reduced processing latency by 80% and boosted recommendation coverage from under 60% to over 85%.
Intelligent load balancing – Implementing smart load distribution across processing pipelines ensured consistent performance even during viral content spikes and peak traffic periods.
Business impact
The Content Knowledge Graph has delivered measurable improvements across key business metrics:
Enhanced content discovery performance – CKG-powered content recommendations boosted Cards per Session (CPS) by 24%, directly improving user engagement and platform stickiness.
Significantly increased user engagement – More relevant and contextually aware content delivery resulted in higher click-through rates and a 5% increase in swiping sessions, particularly in the related content sections.
Real-time trend intelligence – Users now discover trending topics instantly across news, entertainment, sports, and lifestyle categories, with faster trend detection compared to previous systems.
Data-driven content strategy – The CKG provides comprehensive, actionable insights into content performance, user preferences, and emerging trends, enabling data-driven editorial and curation decisions.
Global scalability and efficiency – The cloud-native architecture seamlessly handles Glance’s ever growing global content pool while maintaining cost efficiency and performance.
Shaping the Future of content discovery
The Glance Content Knowledge Graph transforms raw, unstructured content into a sophisticated, interconnected knowledge network, and has has empowered Glance to deliver truly engaging content experiences that anticipate user needs.The solution’s success lies not just in its technical sophistication, but in its ability to enhance the human experience of content discovery – making every interaction with Glance more meaningful, relevant, and engaging.
As content continues to proliferate and user expectations for personalization grow, the principles and technologies demonstrated in this project provide a blueprint for the future of intelligent content platforms. We’re excited to see how Glance leverages this powerful foundation to further innovate in personalized content discovery and set new standards for user experience in the digital content ecosystem.
AI Summary and Description: Yes
**Summary:** The text details Glance’s innovative approach to personalized content delivery using a Content Knowledge Graph built in partnership with Google Cloud. By leveraging advanced AI technologies, this solution significantly enhances content discovery and personalization for over 300 million users. The implications for AI and cloud computing security are notable, as the system navigates large volumes of data and user interactions while maintaining relevance and responsiveness.
**Detailed Description:**
The provided text discusses the implementation of a sophisticated Content Knowledge Graph (CKG) by Glance, a content discovery platform. This initiative aims to transform content delivery through deep contextual understanding and advanced AI capabilities. Here are the key elements and insights from the text:
* **Personalization Challenge:** Glance addresses the challenge of delivering hyper-personalized experiences amidst increasing digital content and user expectations. The core objectives included:
– Uncovering emerging trends and viral content
– Enhancing the relevance of search results based on user behavior
– Providing intelligent context for content relationships
– Scaling personalized experiences for a vast user base
* **Partnership with Google Cloud:** Glance collaborated with Google Cloud Consulting to leverage state-of-the-art AI and data processing tools, including:
– Gemini models for advanced content understanding
– BigQuery for efficient data processing
– Neo4j for managing the Content Knowledge Graph
* **Core Features of the Content Knowledge Graph:**
– Intelligent ingestion from various content sources for structured processing
– Extraction of entities (people, events, etc.) and relationship mapping using AI
– Real-time trend analysis and prediction based on user engagement
– Dynamic knowledge graph construction that builds a network of interconnected content
* **Scalability and Performance Optimizations:**
– Utilization of an event-driven architecture for real-time processing
– Optimization of the graph database for faster query responses
– Deployment on Google Kubernetes Engine to ensure reliability and scalability
– Strategic performance optimizations reducing latency and improving recommendation coverage
* **Business Impact:**
– The implementation has led to measurable improvements in user engagement and content discovery, such as a 24% increase in Cards per Session and a 5% rise in swiping sessions.
– A data-driven content strategy is now enabled, allowing for informed editorial decisions based on user preferences and engagement data.
* **Future Implications:**
– This project not only showcases technical advancements but also sets a standard for future intelligent content platforms, emphasizing the alignment of AI capabilities with user-centric content experiences.
– It illustrates potential challenges in ensuring security and compliance while managing vast amounts of user data and interactions.
In summary, the CKG initiative exemplifies how advanced AI and cloud architecture can transform user experiences in content discovery while highlighting considerations for security and personalization in the evolving digital landscape.