NING GUO

Query Intention Detection

- Developed a binary classification model to identify video-intent queries using Transformer-based architecture (fine-tuned task layer).

- Leveraged large-scale search logs and clickstream data for training, applying effective sampling strategies.

- Evaluated model performance using ROC-AUC, precision-recall, and online A/B testing, improving triggering accuracy for video results.

Video Trigger Decision System

Design and deployed a tree-based ML model for real-time video trigger decisions on Bing search.

- Built feature pipelines using search logs and LLM-labeled data for both query and content representation.

- Improved video engagement metrics by enhancing result relevance and reducing low-quality triggers.

LLM-based Quality Measurement Pipeline

- Built a scalable quality monitoring system using large language models (LLM) and fine-tuned SLMs to measure search result quality.

- Automated daily scraping, inference, and reporting pipelines, providing actionable insights into model and UI degradation.

- Played a key role in integrating quality metrics into product decision-making loops.

Optimizing ranking results of commodity queries with purchasing needs in the commodity vertical search field by building and iterating trigger and ranking machine learning and deep learning models and algorithms. Such as Train LTR model, evaluate model based on NDCG, DNN multi-class query classification model for classification tasks.

Collecting and analyzing the delivery drone operation data everyday, using data visualization tool PowerBI to analysis and show these results.