Research Data Management Development Fund (2022)

We summarise the results in the policy brief articles into charts and figures, which are shared in the CUHK Research Data Repository to the public. These graphical outputs give the readers an update about the shift of economic and educational inequalities in the main GBA cities. They also provide implications on the urbanisation and returns to education, which helps understand the challenges and barriers in the current GBA labour market.

The funding was used to hire student helper and support part of the salary of research staff in working in this project. They were trained to use the CUHK library resources to collect data and manage the online data. From this project, they also learnt how to handle sensitive data and handle with data privacy issues.

For this reason, we propose the first Cantonese Knowledge-Driven Dialogue Dataset for General Use (CK-DDD) in Hong Kong, which collects the information in multi-turn conversations from various sources. Now our corpus contains over 800,000 conversational datasets obtained from over a dozen sources in different business sectors. These datasets will be processed and then make accessible to the public via our online repository. CK-DDD will escalate the utilization of Cantonese dialogue systems in various commercial applications, for example, Live AI chatbot for sales inquiry.

The corpus will be constantly updated and expanded if the usage of Cantonese becomes increasingly popular, and the information provided will be more sentimental and reliable. Eventually, we will organize this dialogue dataset as self-sustainable and freely accessible information to the public for research and development purpose

Our team is working on a system that can hold conversations in Cantonese. We’ve created a special database called the Cantonese Knowledge-driven Dialogue Dataset for REStaurants (KddRES), which will be available online for everyone. This database contains 800 conversations from 10 different restaurants, each with its own unique style. We’re committed to regularly updating this dataset to ensure it remains a valuable resource for all.

Our project has three main goals. First, we’ll share the KddRES dataset with the public, making it a valuable resource. Second, we’re developing a dialogue system that can be customized for any restaurant, which is great news for small and medium-sized eateries. Third, we’ll publish a research paper to tell the academic community about our innovative approach.

To make sure the data we collect is top-notch, we’ve hosted and organized a series of workshops and meetings. We’re keeping a close eye on data quality and keeping thorough records.

This project is backed by CUHK research data management tools, ensuring our data is well-managed and accessible for future use. We’re using Github to share our project’s structure and CUHK data center for downloading the dataset. So, it’s not just about creating a unique dialogue system for restaurants – it’s also about promoting good data management practices.

For the task of fake news detection, compared with traditional news websites, social networks have a unique feature of news transmission topology. How to use powerful methods like GNN to model the content feature and topology of news is an important direction in this research field. We aim to detect fake news by generating critical diffusion structures that determine the veracity of the news. Besides, datasets used in the existing work are old and do not contain the latest hot events in recent years. Therefore, we will collect an up-to-date fake news dataset on social networks to ensure that detection methods remain viable in the new era.

Yusong Zhang, Kun Xie, Xingyi Zhang, Xiangyu Dong, Sibo Wang:
KPG: Key Propagation Graph Generator for Rumor Detection based on Reinforcement Learning. CoRR abs/2405.13094 (2024)

Our research involves creating neural network models with the DeepMD kit, a machine-learning toolkit designed for training artificial neural networks using computationally intensive reference data. The models we develop are stored as graph files on researchdata.cuhk.edu.hk. Despite the large amount of data required for training, the final model files are compact, consisting mainly of the network weights. This compactness is similar to large-language models like ChatGPT, which, although trained on extensive datasets, are ultimately defined by their weights.

These models are easily accessible to other researchers via our data repository, simplifying the process for those interested in applying our methods to their own molecular dynamics studies. This accessibility not only fosters collaboration but also accelerates advancements in the field by providing a ready-to-use resource that can be seamlessly integrated into diverse projects.

2)DeepMD model of water-oil interfaces: https://doi.org/10.48668/WVQDAK

3)DeepMD model of water-oil interfaces with catalyst: https://doi.org/10.48668/OCLLON

The data provides global historical and future climate information, useful for dynamical downscaling projections of the Earth’s future climate, atmospheric environment, hydrology, agriculture, wind power, etc.

README files describing the raw data files, as well as R sample scripts for easier data handling and visualization were developed and included in the repository. This enables the general public, including those who may not have strong coding and climate background, to easily access and utilize our data for downstream input. The improvement in access to our data may enhance public knowledge in climate change and visualization of similar kinds of data.

During the development of this project, we have also gained experience in making our data more user-friendly to those unfamiliar with climate data processing. The platform been useful for sharing of our data, and our experience may also be helpful for some courses from the Earth and Environmental Sciences Programme, as well as in other educational projects, such as e-learning material development.

The dataset is a valuable resource for researchers studying Chinese media, politics, and society. We have already used this dataset to produce journal articles. The following article was published in 2023, and more articles are under review:
Guo, J., Huang, X., & Fang, K. (2023). Authoritarian environmentalism as reflected in the journalistic sourcing of climate change reporting in China. Environmental Communication, 17(5), 502-517.

I also presented the dataset and relevant research at major international conferences including the International Communication Association conference and the Association for Asian Studies conference. Scholars studying China’s media and society show great interest in this dataset and we are developing potential collaborations in using this dataset for future research.

The RDM Development Fund has helped secure raw data of this project, and a robust system was developed to ensure data integrity and accessibility for future use. It also paid for both human coders (student helpers) and generative AI tools to analyze and clean the data. Student helpers gain valuable experience in data processing and analysis, contributing to their professional development. It also supported the attendance of international conferences, broadening the reach and impact of the project.