Better Netizen: Using GCN Networks

Our project runs under a regional NGO which works as a hub to support our daily maintenance. Our first pilot is granted with funding from Future Education Foundation High School Seed Sponsorship. Through cooperation and coordination, synergies among domestic minority and international students in order to promote a more unified regional voice and cultivate an ecosystem for youth development. Our next stage is developing a BetterNetizen app for applying our algorithm to better support and protect the netizen community.

The GCN model, Adaptive Graph Convolutional Neural Networks, using data science and machine learning to make social media a peaceful and race-neutral place, helped us gain the first prize. In the future, we would like to increase our influence in applying technology (in particular, machine learning and data science) to promoting solidarity and fighting against racism.

The expeditious growth in usage of social media platforms and blogging websites passed 3.8 billion marks of active users that use text as a prominent means for interactive communication. In the past few years, there has been a significant rise in toxic and hateful content on various social media platforms. Recently Black Lives Matter and Anti-Asian Hate movement came into the picture again causing an avalanche of user-generated responses on the internet.

A fraction of users use discriminatory communication intended to insult and intimidate specific groups or individuals due to their gender, race, sexual orientation, or other characteristics that have been an obstructive byproduct of the growth of social media.

While efforts to educate about racial justice and counter hate have been made via social media campaigns (e.g., the #BLM campaign), but their success, effectiveness, and reach remain unclear. Moreover, online hate speech has a severe negative impact on the victims, often deteriorating their mental health and causing anxiety (Saha et al., 2019).

Thus, it is critical to utilize an effective method to evaluate and detect hateful users. After detection, censors can take necessary measures to warn or punish haters to prevent haters from spreading hate speech.

Our BetterNetizen project, a new GCN model, Adaptive Graph Convolutional Neural Networks, can address the problem that existing GCN models treat edges in the network unweighted, which cannot fully capture the structural information and leads to suboptimal performance. By implementing the new GCN model, online social platforms can utilize our proposed method to evaluate and detect hateful racist users. This not only applies to social media, but also to any form of text related to racism. The censorship system will be built in through machine learning using the GCN model. 

Online social platforms can utilize our proposed method to evaluate and detect hateful users. After detection, staff can take necessary measures to warn or punish people giving hateful speeches online, thus preventing these people from damaging online racial equity. In the meantime, the model can also be used to assess the social impact on different types of hate speech. Technology helps to bring convenience and happiness to humanity and also brings negative influences. Hate speech is exactly an issue coming with Informatics technology. We hope that the proposed method will solve the problem and, in the future, provide a clean and friendly Internet world to all users.

Our website (https://www.chssaofficial.com/) will serve as a resource for organizations and individuals working to combat hate in their communities. Our next pilot is building 2B support and app development to better apply our algorithm. Through the website, people can download open source and apply the source to their own websites so that the censoring method can work.

We specifically target these two groups of people:

• People delivering hate speech online: people who spread negative speech, promote violence, and attack or threaten others on the Internet should be punished and educated. Our algorithm can help find these people accurately. Then we may prevent them from giving any speech within a certain amount of time or provide educational resources.
• People affected by racial hate speeches online: People who experience discrimination and exclusion (social, political, and economic) because of unequal power relationships across economic, political, social, and cultural dimensions are more likely vulnerable to hate speeches. Online hate speech has a severe negative impact on the victims, often deteriorating their mental health and causing anxiety. To protect people affected by hate speeches is also to combat racial discrimination, protect countries’ cultural diversity and promote equality and justice to all.

The new model can detect hateful users in social networks. Different from existing GCN models, AdaGCN treats weights of edges in the social network trainable and learns transformation matrices simultaneously. To prevent overfitting, AdaGCN has a leave-one-out loss function using label propagation to provide an extra supervised signal for the training process of our model. On a dataset of 100,386 Twitter users, the AdaGCN model demonstrates a robust performance of AUC score and 0.474 ± 0.018 F1 scores, surpassing all the baselines. Strictly censoring racist hateful speeches giving hateful speeches will create a friendly online environment and promote racial equity.

Our project runs under a regional NGO which works as a hub to support our daily maintenance. Our first pilot is granted with funding from Future Education Foundation High School Seed Sponsorship. 

Highlights:

• Aims to amplify the voices of international students, build a united community and provide resources, solutions, and a safe space.
• Contribute to a better understanding and development of a common approach to youth empowerment in the region by strengthening mechanisms for information-sharing and exchange of best practices
• Establish and maximize the use of new and existing virtual platforms for information-sharing of experiences of cultural exchange.
• Won the first prize in the Technologies Promote Education Equity competition held by Future Education Foundation in 2020 and secured a scholarship of 5,000 dollars.

Our next stage is developing a BetterNetizen app for applying our algorithm to better support the netizen community and deploying the app in at least one community.

Catherine Jin (Main Leader), Ian Chen, Tony Wang