Szu-Min Chen Email & Phone Number

Sunnyvale, Ca, Us

System Infrastructure

San Francisco, California, Us

Search Infrastructure Team

San Francisco, Ca, Us

- Owned entity disambiguation project, developed algorithms to automatically identify ambiguous queries from more than 2 millions search logs collected from Hive.- Established a dataset with 3000+ data by devising a work flow for efficient and automatic label update using Python and built the fronted using React-Redux.- Boosted 10% accuracy by building tools to quantify and improve the performance of current model on capturing entities in users' search queries.- Controlled the quality of entity extraction by detecting anomaly model behavior via developing a data pipeline with Airflow.

Taipei City, Taipei City, Tw

- Worked on visual storytelling, building a system using Pytorch that generates a fluent story given a stream of images. I proposed an inter-sentence diverse beam search algorithm to improve the expressiveness and diversity of generated stories. This project is published as NAACL workshop paper.- Dealt with sentiment analysis task about predicting valance-arousal score of a word/phrase. I designed a model that automatically gets several images that correspond to each word from the Internet to augment the training data, also incorporate attention and word embedding to enhance accuracy. Published as IJCNLP shared task paper.- Developed a conversation sentiment classifier with Pytorch which takes into account the user switch and emotion flows in the real conversational data. The results surpassed state-of-the-art by 9.9%.

Armonk, New York, Ny, Us

Worked on several projects and also collaborated with large teams include engineers and consultants.- Programmed a robot on embedded Linux system and connects it with the cloud platform using Node.js. Implemented several functionalities such as text-to-speech and motion control. It can takes commands in natural language and also be able to control by Android app.- Developed large-scale intellectual conversation systems using Java , Javascript and deployed them on the cloud platform for multiple airline and retailing companies in Taiwan. Combined machine learning and linguistic rules to improve user experience and also incorporated the concept of BPMN model to enhance stability.

Master Of Science - Ms, Computer Science

Bachelor Of Science - Bs, Computer Science