Page 81 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Computer Science & Engineering

and later apply machine learning and data mining models to it. The goal of our project is to totally struct Stack Overflow
to an HIN so that we can learn the knowledge embedding and optimize use experience.

Natural Language Processing for Social Media and E-Commerce Data

Supervisor: SONG Yangqiu / CSE
Student: DU Donghong / ELEC Course: UROP1100, Spring

Natural language processing is a hot topic nowadays with broader and broader applications in daily life where social
media and e-Commerce data plays a role. This report basically describe what I experience this innovative field under the
instructions of supervisor as well as some other learning process for basics of NLP. First part is experience on utilizing
various name entity recognition API to deal with the data and second part is about a simple recurrent network learned
from online source.

Supervisor: SONG Yangqiu / CSE
Student: BAI Jiaxin / COGBM Course: UROP1100, Spring
UROP2100, Summer
SUN Xinyu / COSC Course: UROP1100, Spring
UROP2100, Summer

This summer we continue working on our original plan. Jiaxin focus on Multiplex Network Embedding for Selectional
Preference. MNE is a way to use dense vectors to represent the nodes of a multitype knowledge graph. Meanwhile the
dependency graphs are multitype knowledge graph, with words as nodes and dependency types as edge types. As a
result, we can get the word embedding by applying MNE to dependency graphs. We evaluated the embeddings with SP-
10K dataset. Xinyu focus on grammar correction and helped Hongming to search for the relevant materials. We had a
basic model based on Robsut Word Reocginiton via Semi-Character Recurrent Neural Network. And here we have one
example:

From this example, we can see that the network’s performance is acceptable and this network is very easy. If we add
more layers, it will perform much better. The details will be explained in the later part.

Fair and Efficient Cache Management for In-memory Big Data Systems
Supervisor: WANG Wei / CSE
Student: CHENG Xiaohe / COSC Course: UROP1100, Spring

71