Help retrieval-augmented generation systems answer individual medical expense insurance coverage questions accurately using policy evidence.

Evaluate whether models can find the right premium fashion products using multi-constraint queries and explainable relevance judgments.

Labeled images of fighting and walking in public scenes support behavior recognition and monitoring of potential incidents.

Improve translation of financing offers, payment plans, and checkout options for installment purchases of appliances, electronics, and furniture.

Connect shopper needs with small appliance product pages and matching evidence to support retrieval training and evaluation for retail.

A high-quality image dataset for detecting leftover areas on cafeteria trays.

Retrieve business hotel catalog details with verifiable source evidence, from room configurations to rates and cancellation terms.

An audio dataset covering front desk service dialogues, supporting the training of intent classification models.

This dataset focuses on barcode image recognition and security protection applications.

Train models to use tools reliably and answer with evidence through inventory and batch lookup trajectories grounded in specialty food retail.

Train models to turn application materials and policy sources into complete, evidence-traceable reviews of personal loan decline decisions.

Turn luxury hotel dining descriptions into structured catalog attributes for smarter information extraction and catalog management.

A high-quality text conversation dataset containing hotel ratings, reviews, and replies.

The Restaurant Onsite Call and Reminder Chinese Speech Recognition Audio Dataset focuses on enhancing the intelligent interactive experience in the retail e-commerce industry.

Train models to recognize the intent behind guest messages with contextual examples from multi-turn hotel service conversations.

Match extended-stay guests' facility and space requirements to hotel rooms and suites for more useful lodging search and recommendations.

Train models to answer questions about building fire systems using source material, while recognizing when the evidence is insufficient.

This dataset presents the structure of tableware placement under different conditions in image form. It is used to enhance the intelligence level of dining management and automated placement systems.

Evaluate whether a system can find the right service, identify the right merchant, and support its answer with evidence from multi-merchant catalogs.

Train and evaluate roasted coffee product retrieval with retail catalog attributes paired with natural-language queries.