Multimodal Content System

Agoda built a Multimodal Content System to bridge the gap between siloed image and review data. This system uses a topic-based clustering approach to programmatically link visual content with textual feedback, providing users with a richer, contextual understanding of hotel properties. It involves enriching images and reviews with tags, mapping these to central topics, and serving curated content packages for low-latency retrieval. This post details the technical implementation of the image classification, review tag extraction, sentiment aggregation, and the overall data pipeline orchestration using PySpark and Kubeflow, with serving via Couchbase.