Machine Learning Engineer2018 - 2018Fujitsu
Technologies: Python, TensorFlow, Spark, Selenium, Keras, Scikit-learn
- Built a crawler that collected the labels of a given phone number from a search engine's semi-open database. It processed millions of call detail records from tens of thousands of applicants for a subprime loan using Pandas.
- Manually selected features and classifiers with different imbalance dataset handle tricks to build a risk classification system and achieved 72% precision.
- Proposed an info-flow model based on a call detail record.
- Applied a model on a random-walk-based graphic neural network implemented by TensorFlow and Keras which is faster than a DNN approach based on the Tucker decomposition and achieved similar metrics, including a 76% F1-score and with 78% precision.
Research Assistant2015 - 2017Tsinghua University
- Collected reviews using Selenium from Amazon and applied a feature-extraction algorithm to find sentiment words and corresponding aspect words.
- Transformed raw reviews into VSM and built a binary sentiment analysis system and achieved 88% accuracy.
- Collected FAQs from an eCommerce website using Python.
- Extracted the subject, predicate, and object from the questions and answers.
- Built an automatic question-answering system based on similarity and applied it to an iBen robot.