Shanhe Ding Email & Phone Number

San Jose, California, United States

San Diego, United States

I focused on refining and enhancing a generalized, backward-compatible embedding method, which involved:• Pioneering the utilization of a freeze-weight Word2Vec model over conTREE data• Crafting and validating a new metric, in tandem with an alignment method specifically for conTREE, fortifying our data analysis accuracy and system's robustness.

中国 上海市

I contributed significantly to the design and optimization of models and algorithms aimed at automating the analysis of resumes. Key areas of focus included the generation of embedding models, the development of clustering models, and the expansion of distant vocab and unsupervised vocab dictionaries.● Established auto-quality phrase extraction module ○ Used statistical method generating distant quality phrases and fused contrast learning with UCPhrase ○ Implemented phrase filtering methods including PMI, NER, and JS-divergence ○ Replaced ATTmap+CNN structure with last-hidden state + contrastive learning to enhance performance ○ Increased size of distant vocab dictionary 400% larger with 37% higher precision ○ Added unsupervised unseen vocab rating feature to the vanilla module● Designed hierarchical-based clustering model for large corpora ○ Implemented level-wised mini-batch K-means with restriction on inertia ○ Boosted the clustering speed 300% faster

Junior Research Assissant for Computer Science department

Master Of Science - Ms, Computer Science, 3.97

Bachelor Of Engineering - Be, Computer Science

Exchange Student, Computer Science, 5.5/6.0