Kang Chun Fan
Hi my name is Kang Chun Fan. My goal is to become a backend engineer or a data engineer, therefore, I am currently woking on to enhance my ability to handle backend services also taking courses related to artificial intellegence. As for a long term schedule, I am planning to do some projects related to backend handling, especially in the part to manage large-scale data, which might be useful in machine learning too.
Full-time Research Assistant
Computational Finance and Data Analytics Lab, Academia Sinica | Prof .Chuan-Ju Wang
• Domain-Specific Sentiment Lexicon Construction of User Generated Reviews
- Processed raw reviews from Yelp and TripAdvisor with NLP methods to improve efficiency for matrix calculation.
- Constructed the co-ocurrence network to produce the embedding vectors.
- Evaluated the similarity of embedding vectors for candidate sentiment words to build the domain specific
• Financial Engineering Prediction on Company Default Probabilities with Deep Learning (Cooperation with National University of Singapore)
- Predicted the probability a company will bankrupt or be acquired within 1 month to 36 months.
- Designed the LSTM neural networks with tensorflow based on python to predict real world financial data.
- Enhanced the accuracy up to over 90% to predict 1 month before event, and over 80% to 3 months before event.
R&D Intern, CMoney Financial Software Company
- Increased proficiency in C# and object oriented skills through employee training.
- Built billboards with C# to watch stock prices instantly.
- Implemented packet switching system to simulate the congestion of data interchange when stock market opens.
Master of Computer Science
University of California, Santa Cruz
B.S. in Engineering Science & Ocean Engineering
National Taiwan University
Programming: Python, TensorFlow, C++, C#, Java
Software: MATLAB, Linux/Unix, LaTeX
Kang-Chun Fan, Chun-Hsiang Wang, ”RELEX: REpresentation Learning for Sentiment LEXicon Genera- tion from User Reviews”, BigNet 2018 Workshop in Lyon
- The framework requires no additional annotation of seed words or external lexicons.
- Produced embedding representations of lexicons to enhance the usability of further Machine Learning research.
- Applied to a variety of datasets from different domains to construct domain-specifc sentiment lexicons.