CHANG-YU TAI

747-***-**** ! ad16hh@r.postjobfree.com Ï Los Angeles, CA 90006 è Chang-Yu Tai ¥ github google scholar EDUCATION

The Ohio State University (OSU) Columbus, OH

M.S. in Computer Science and Engineering (CSE) GPA: 3.8 Aug 2021 -Graduated: May 2023 National Taiwan University (NTU) Taipei, Taiwan

MS. B.S. in Chemistry (Chem.) Sep 2011 -Graduated: Jan 2018 EXPERIENCE

Eccalon - Machine Learning Engineer May 2023 – Present

• Collected basketball statistics, including the identification of shots, blocks, assists events, and player roles, as well as team identification.

• Implemented an algorithm to estimate the homography between 3D court positions and 2D frame positions for basketball and football games. Twitter - Machine Learning Engineer Intern, CFFM Team May 2022 – August 2022

• Implemented hyperbolic-based KGE score function in PBG package and pre-trained large-scale KGE for 5 billion users.

• Introduced hyperbolic-based KGE, containing debiased information, to Twitter’s recommendation model, achieving 0.7% RCE improvement.

• Introduced Product Quantization for embedding compression, achieving 88.5% storage reduction while maintaining on-par performance. Academia Sinica - NLP Research Assistant, supervised by Dr. Lun-Wei Ku Jan 2019 – Jul 2021

• Conference Committee Reviewer: ACL 2021-2023, EMNLP 2021-2023, AAAI 2023-2024. External Reviewer: SIGIR 2022, KDD 2022, ICDM 2022.

• Conducted research independently on Knowledge Graph based Recommendation System, Content-based Recommendation System

• Improved the content-based news recommendation models with cacaFly’s (Best Digital Marketing Agency in Taiwan), achieving averaged 21.5% Click-Through Rate and 6.3% Duration improvement for a total of 30 days on the cacafly’s online news platform.

• Implemented the personalized advertisement generation models with E.SUN bank, Taiwan’s 1st in the World’s Best Banks 2021, and achieving average 6.9% Click-Through Rate improvement for a total of 60 days on the E.SUN’s online advertisement platform.

• Participated in 2019 Formosa Grand Challenge—Talk to AI hosted by hosted by the Ministry of Science and Technology (MOST) Taiwan. and won 1st place at the QA part by using Pytorch Transformer with a new algorithm proposal named “Two Times Squad“. - [Link] RESEARCH PUBLICATIONS

Exploring Chain-of-Thought Style Prompting for Text-to-SQL. C.-Y. Tai (1st author), H. Sun. - [Link] EMNLP 2023 Long

• Studied how to enhance the reasoning ability of LLMs for text-to-SQL parsing through chain-of-thought style promptings. Roll Up Your Sleeves: Working with a Collaborative and Engaging Task-Oriented Dialogue System. - [Demo] SIGDIAL 2023 Demo

• Developed a chit-chat feature for a task-oriented dialogue system, aimed at enhancing user satisfaction.

• Leveraged data processing technologies, including Docker, Spark, Elasticsearch, PostgreSQL, CI/CD and Amazon EC2, to build remote modules. Hyperbolic Disentangled Representation for Fine-Grained Aspect Extraction. C.-Y. Tai (1st author), M.-Y.Li, L.-W. Ku. - [Code] AAAI 2022 Long

• Introduced hyperbolic space information to capture relation between seed word and review segment for Aspect inference. Knowledge Based Hyperbolic Propagation. C.-Y. Tai (1st author), C.-K.Hua, L.-W. Ku. - [Code] SIGIR 2021 Short

• Introduced hyperbolic space ebmedding, which effectively encodes hierarchical structure information, to GNN-based model.

• Developed hyperbolic-based GNN recommendation model by Pytorch, achieving 96.2 % Precision improvement on Amazon-book. User-Centric Path Reasoning towards Explainable Recommendation. C.-Y. Tai (1st author), L.-W. Ku. - [Code] SIGIR 2021 Long

• Developed a Multi-View Policy Network model in Pytorch for reasoning user’s decision making process.

• Improved model by utilizing a panoramic view of the user’s portfolio while reasoning, achieving 36.9 % Recall improvement on Movie-Lens. MVIN: Learning Multi-view Items For Recommendation. C.-Y. Tai (1st author), M.-R. Wu, L.-W. Ku. - [Code] SIGIR 2020 Long

• Built a user-centric GNN-based model in Tensorflow, which explicitly highlights user interesting nodes on KG by attaining them on user portfolio information, achieving 5.9% AUC improvement on Amazon-book dataset PROJECT

XHS Search Map - [GitHub] Aug 2023 – Dec 2023

• Implemented Entity-Link functionality for location information and extracted keywords from influencer posts.

• Employed React as the server-side framework and seamlessly integrated MySQL with Express to show the results. Public Opinion Analysis on Social Media - [GitHub] Jan 2021 – Mar 2021

• Collected Twitter information via Twitter API and prepossessed the whole data to generate individual user info by spaCy and Scala.

• Generated social media trends and keywords via the Aspect Extraction algorithm and analyzed overall social trends via Spark. SKILLS

Programing Languages (In order of proficiency): Python, Scala, Java, C++, C Tools: Pytorch, Tensorflow, AWS, Docker, Kubernetes, Spark, Scala, Notebook, MySQL, Elasticsearch, D3, Node.js, React, Scikit-Learn, OpenCV

Contact this candidate