Deng Pan Email & Phone Number

Hangzhou, Zhejiang, China

Participated in optimization of the recommendation system on China's largest e-commerce platform, primarily focusing on optimizing deep learning algorithms for recall, ranking. Conducted large-scale multi-GPU training on billion-scale data.Implemented ranking improvements. Integrated various cutting-edge deep learning models including DIN, SENet, MMOE, and PLE, while conducting feature selection and dynamic adjustments based on relevance and personalization.Enhanced recall stage, added a new video recall pathway using SwingI2I., improving relevance and consistency between content and product recommendation.Achieved a 2.56% AUC improvement through ranking model upgrades. After two months of continuous testing, average three-second retention stability increased by 3.62%, achieving full deployment online.

Beijing, China

Created a model to detect event based on supervised contrastive learning; trained model on dialogues datasets.Ran BERT model as baseline with accuracy of 67% and assessed results to find key point of Event Detection task.Modified sampling method of trigger words using category information; applied concept of "class centroid" to augment trigger words as training samples; improved effect of contrastive learning by 30%.Conducted experiments; with optimized sampling method, model performed 12% better than baseline.

Beijing, China

Focused on POI (point of interest) strategy for Didi's map including removement, revision and addition of POI; generated 9k+ pieces of valid POI information within 3 months, directing following manual work.Processed million-level user data with Spark, filtering valid data with search terms and date for POI algorithm; Developed and ran routine tasks on Linux; Applied Git for streamlined version control and team code reviews.Designed deep learning models based on BERT to extract semantic information of users' retrieving words ; predicted users' potential target POI and devised algorithm to identify 1k+ mistakenly removed POI.Implemented KNN algorithm with Python to discover and mark confusable POI pairs on Didi's map; wrote scripts to automate routine tasks of digging confusable information, identifying 3k+ confusable POI pairs as reminder for users.

Hangzhou, Zhejiang, China

Developed a network maintenance tool with Java to monitor situation of Ethernet switches in company.Utilized Springboot for back-end development and MySQL to build database.Collected customer opinions of China Air using Python crawler and analyzed with deep learning based NLP models.Wrote Python scripts to automate routine tasks of public opinion analysis, analyzing 30k+ online comments.

Master Of Science - Ms, Electrical And Electronics Engineering

Study track: Computer Engineering

Bachelor Of Science, Computer Science