Experience

Fake reviews and extreme opinion detection based on Yelp

Fake reviews detection model based on graph features, accuracy up to 0.88
Fint-tuning on ULMFit model with the accuracy of binary sentiment analysis up to 97%, better than the best method available with 95% accuracy
Innovative definition of extreme opinion and model transfer to five classification problems, with accuracy to 79.55%

Face Recognition API

In Linux Docker configuration deep learning environment
Develop deep learning models for face detection(MTCNN), Face alignment and Face Recognition(FaceNet),image quality assessment, age recognition and gender recognition
API was developed based on Tornado framework

Data analysis for Chinese Football Association Super League(CSL)

Data processing and data cleaning
Explore player game-style distribution and find abnormal players by clustering
Build time series model(eg. LSTM) to predict the outcome of the game, accuracy up to 63.2%

Google Sites

Report abuse