New AI Method Improves Healthcare Search Intent with Query Clustering and Session Context

Classifying the intent behind healthcare search queries is notoriously difficult due to ambiguous language and limited labeled data. Prior work used user click behavior and pairwise loss functions to learn query representations, but many health queries have multiple intents, leading to contradictory click patterns. Global aggregate statistics can also misalign with session-specific intents. The team from Emory and industry researchers tackled these issues with a multi-pronged approach.

Their solution clusters similar queries to improve representation learning, then applies a custom loss function designed to handle multiple intents simultaneously. They introduce a concordance rate (CR) score to quantify how much a query's global intent aligns with its session-level intent. Experiments on a proprietary Health Search dataset and the public TripClick benchmark show their method boosts both clustering quality and downstream intent classification accuracy, outperforming existing baselines. This work promises more relevant healthcare information delivery in search engines.