Xiaohan Ding

Welcome to my website! I recently graduated from University of Massachusetts Amherst with a Master of Science degree in Computer Science, and I am currently working at Amazon as an Applied Science in the Automatic Speech Recognition field.
Interests: My interest lies in the intersection of Scocial Science, Statistics, and Computer Science. I like to ask the question, what makes the society it is today and how can we make it better. How to quantify “fairness” and design frameworks and algorithms that account for this “fairness”. In the world where everything becomes to connect to the internet, how can we design systems and artificial agents which act in the best interest of human and make people better off.

Research Projects

Analyzing the BlackLivesMatter movement through Tweets 2018 Spring
The hashtag #BlackLivesMatter has become a symbol of the anti-police brutality and anti-racism against the Black American movement on Twitter. The work reported in this paper aims to characterize the dynamic developments of the #BlackLivesMatter hashtag and its counter creation #AllLivesMatter on Twitter from 2014 to 2018, by utilizing Natural Language Processing (NLP) techniques.
I am very grateful to be advised by Prof. Brendan O’Connor on this project.
Amazon Alexa Automatic ASR Error Correction 2019 Summer
This project is done while I was an Applied Scientist Intern at Amazon. Alexa is the voice-controlled smart assistant developed by Amazon. The Automatic Speech Recognition (ASR) system for Alexa already achieves high accuracy. However, it suffers from long-tail errors such as new commands or long sentences. Thus, we want to develop a model that can automatically detect and correct possible transcription errors made by upstream ASR systems. During my internship, I developed a transformer-based deep neural network model that improved word error rate (WER) by 3.5% compared to the current Alexa production model.
HaydnModels: Symbolic Music Composition with Ensemble Models 2018 Fall
The goal of this project is to build model(s) that can compose string quartets like the famous composer Haydn. We achived this by utilizing neural network’s power to learn patterns. Our final model(s) are built upon the LSTM structure and we construct the models in a way that each model learns one aspect of the music composition.

More About Me

In order to feed myself after college I started learning how to cook, and it has quickly became one of my favourite things to do. I love trying recipes from different cultures and countries. I learned dishes from the Mideast, the Mediterranean, Peru, India, Thai, Korea and many more. Check out some of my favourite dishes here. The experience of cooking food from places other than my home country showed me that many good flavors are international. I think just like life, we are all sharing the experience of existing on this planet.

I grew up playing piano and attending drawing classes, both I still love to do. Recently, I have been induging myself in the rap music world. I love the poeticism of the rap lyrics, and the love and hope behind them.