Travelers often rely on both photos and written feedback to choose a hotel, but these content types have traditionally lived in separate worlds. Agoda has now bridged that gap with a pioneering multimodal content system that unifies hotel images and guest reviews through a shared topic taxonomy. This system enables seamless retrieval across over 700 million images and multilingual reviews, supported by offline enrichment and low-latency serving. The result is a richer, more intuitive travel discovery experience.

The Challenge of Disparate Content Types

For years, travel platforms have struggled to connect visual and textual data. A guest’s glowing review might mention a “spacious lobby,” but that same lobby could be hidden among hundreds of unrelated photos. Similarly, a beautiful pool image might never appear alongside reviews praising the poolside service. This fragmentation forces users to manually cross-reference information, which is time-consuming and often incomplete. Agoda recognized that to truly streamline travel planning, it needed to align images and reviews under a common framework.

A Unified Topic Taxonomy

The core innovation lies in a shared topic taxonomy—a structured set of categories that describe key aspects of a hotel experience, such as room cleanliness, breakfast quality, or pool amenities. Every image and every review segment is automatically tagged with relevant topics from this taxonomy. For example, a photo of a beachfront view gets tagged as “ocean view,” while a review sentence like “the staff was incredibly helpful” gets tagged as “service.”

How It Works

The process combines two major stages:

• Offline enrichment: Machine learning models analyze image content (using computer vision) and review text (using natural language processing) to assign topic tags. This pre-processing ensures that all data is enriched before serving, without affecting real-time performance.
• Low-latency serving: When a user searches for a specific topic—say, “spa”—the system retrieves both relevant images and review excerpts in milliseconds, presenting them side by side in search results or hotel pages.

This dual approach allows multimodal retrieval, meaning users can discover hotels based on any combination of visual and textual clues.

Scale and Performance

Agoda’s system handles an enormous volume of data: over 700 million images and reviews in multiple languages. The offline enrichment pipeline runs at scale, tagging new content as it arrives. To maintain speed, the serving layer relies on optimized indexing and caching strategies. The result is a system that can respond to complex queries—like “beachfront property with excellent breakfast reviews”—within fractions of a second.

This performance is critical for mobile users and those browsing during peak travel seasons. By keeping latency low, Agoda ensures that the multimodal experience feels seamless, not sluggish.

Benefits for Travel Discovery

The unified system transforms how users explore and compare hotels. Key advantages include:

1. More precise search results: A query for “cozy rooms with great views” now returns both matching photos and review snippets, giving users immediate visual and textual confirmation.
2. Reduced information overload: Instead of scrolling through endless galleries and review lists, users see only the most relevant pieces—filtered by the same topic.
3. Cross-language consistency: Because topics are language-agnostic, a review in Japanese about “清潔な部屋” (clean room) can be matched with an English-tagged image of a tidy bedroom. This breaks down language barriers in travel discovery.
4. Better decision confidence: Seeing a photo of the exact pool described in a positive review builds trust and reduces hesitation before booking.

Conclusion

Agoda’s multimodal content system represents a significant step forward in how online travel platforms organize and surface information. By aligning images and reviews with a shared topic taxonomy, the company has created a more intuitive discovery experience that saves time and delivers richer insights. With its robust scale and low-latency architecture, the system is poised to handle future growth in both content volume and user expectations. Travelers can now search with confidence, knowing that what they see and read will truly match.

This project, led by Leela Kumili, demonstrates how thoughtful data architecture can transform user experience on a global scale.