SOCIAL NETWORKING
ADHD on Twitter
With the widespread use of social media, people share their real-time thoughts and feelings via interactions on these platforms, including those revolving around mental health problems. This can provide a new opportunity for researchers to collect health-related data to study and analyze mental disorders. However, as one of the most common mental disorders, there are few studies regarding the manifestations of attention-deficit/hyperactivity disorder (ADHD) on social media. This study aimed to examine and identify the different behavioral patterns and interactions of ADHD users on Twitter through text content and metadata of their posted tweets.
We revealed how
ADHD users behave and interact differently on Twitter compared to neurotypical
users. Based on these differences, researchers, psychiatrics, and clinicians can
use Twitter as a potentially powerful platform to monitor and study people with
ADHD, provide additional health care support to them, improve the diagnostic
criteria of ADHD, and design complementary tools for auto ADHD detection.
SOCIAL NETWORKING
Political Leaning Change on Twitter
This ongoing project is part of the COSINE project, spearheaded by the USC Information Sciences Institute and including Indiana University and the University of Notre Dame. The goal of COSINE is to create the first-of-its-kind cognitive agent simulation framework for studying multi-scale dynamic of social phenomena in online information environments.
data science
Data Crawler and Sentiment Analysis
This project is to analyze people’s behavior on Non-profit donation campaigns under moral foundation theory in Computational Social Science Lab (CSSL), USC.
In particular, I collected data from GoFundMe to create our own data set. Using Selenium, I first extract the URLs of each campaign from the index page. Then, using Request + BeautifulSoup, the crawler collects relevant information about each campaign (such as campaign name, fundraising team, etc.) from those URLs. The information of each campaign is saved in JSON format, and all data are saved into a plain text file for further analysis. Then, I applied sentimental analysis to the description of campaigns. The model scores the text for each emotion (14 in total), then labels (classifies) the text to the emotion with the highest score. [See code here]
NLP
Abstractive Text Summarization
As my graduation project, I focused on abstractive text summarization on Chinese single document. Compared with the extractive text summarization that extracts key sentences from the text as a summary, the summary results generated by the model of abstractive text summarization, which is based on deep learning, are more concise, and conform to people’s natural language cognition. However, abstractive text summarization often faces many problems, such as repeated words or unregistered words in the generated summaries.
In order to solve the above problems, I combined pointer-generator network with a deep reinforced model. This model uses Copy mechanism and new attention mechanism to reduce the appearance of repeated words and unregistered words, and uses reinforcement learning to improve its semantics and ROUGE score. Finally, the performance of the model is verified and analyzed through experiments.
DATA SCIENCE
Members’ Participation in SME-managed Online Brand Community
Nowadays, the increasing Internet leads to great changes in models and means of branding. By developing online brand communities, some SMEs achieve brand success. Existing studies on online brand community can be categorized into two aspects: studies on the motivation of members participating in online brand communities and studies on the impact of members’ participation. However, most of researchers took the communities established and managed by the advocates of strong brands as the study object. Only few of them concern the communities directly managed by the enterprises, especially by SMEs. Without the influence of strong brands, most of the SMEs are faced with the difficulties in recruiting community members and attracting members to participate in community regularly. However, in the real industry, some SMEs have successfully developed powerful online brand community and achieved branding. In this paper, MEIZU official forum is selected as the main study object. Based on literature review and grounded research, a theoretical model has been constructed. Through large-scale survey and data analysis, the hypotheses and relevant conclusions are verified. [See paper here]
DATA SCIENCE
Analysis and Optimization of Business Operation Data
In this project, I lead a team to build a model to analyze the sales behavior of various stores in the business circle and the association between consumers and stores in the business circle from the transaction data, so as to provide businesses with a feasible Improve or optimize the program. [See paper here(Chinese)]
INFORMATION SECURITY
Reversible Data Hiding in Encrypted Images
During a semester, our team explored several methods of reversible data hiding in encrypted images. We use python to reproduce multiple papers, compare the experimental results to prove the effectiveness of the paper method, analyze the influence of the parameters on the experimental results, and analyze the multi-layer embedding ability based on the UCID image library.
In particular, I reproduced the DHS_3 model, this model separate the process of image encryption and data hiding. However, because in the information hiding process, when the difference histogram is shifted for each sub-block, the difference between a certain pixel and a fixed pixel is calculated, and the peak pixel value is fixed at 0 and -1, So it is difficult to extend the algorithm to multiple layers. [See code here]
SOFTWARE DEVELOPING
Online Operation and Management System
During the internship in Jisuanke E-learning, I led a team to develop an online operation system for Go operation and management. We actively communicate with customers, design the database and system structure according to customer needs, and use Django+Vue for front-end and back-end separation development.
SOFTWARE developing
Practice-taking WeChat Mini Program
In this project, I led a team to develop a practice-taking WeChat Mini Program. This mini-program allows users to take practice and record their progress on LeetCode problems using WeChat. Users can also get personalized recommendations and search suggestions, post questions in the forum and discuss with other users.
This mini program has won the Second Price in North China Division, China Collegiate Computing Contest – WeChat Mini Program Application Development Contest 2020.
SOFTWARE DEVELOPING
Personal Website (class project)
This is a project for the HCI class. I built a personal website for Colin Ritman (a character in Black Mirror). I used pure JS+CSS in the frontend, and Java as the backend. This development is more focused on how to apply HCI design rules on the website, and how to enhance users’ experience and increase productivity.