Maps each supplier room to the best reference room using a multilingual cross-encoder. The service scores the full supplier×reference grid (M×N pairs), applies Platt calibration, and returns per-supplier ranked candidates plus an optional selected match when the top candidate clears the decision threshold.

Use when you have a canonical reference catalog and need per-supplier mapping decisions with confidence scores.

Key Parameters

• hotelId: Correlation id; echoed in the response
• chunkKey: Batch/chunk id for large property processing
• threshold: Minimum calibrated score for a positive match (predictedLabel == 1). Default: 0.5
• topK: Number of ranked candidates returned per supplier. Default: 3
• lowercase: Lowercase room text before scoring. Default: true
• includeLogitRaw: Include raw model logits on each candidate when true. Default: false
• references: Canonical reference rooms (array with id and name)
• suppliers: Supplier rooms to map (array with index i and name)

Per-supplier Result

status is mapped when the rank-1 candidate has predictedLabel == 1 and scoreCalibrated >= threshold; otherwise not_mapped with selected: null and candidates still populated.

To experiment with this API, use the sandbox key:
e2R4t6Y8u0i3O5p7A9s1D3f5G7h9J2k4