MODCLUB

Inspiration

I had been thinking about developing a decentralized moderation platform for a while, back in 2021, and was looking forward to the DFINITY Internet Computer launch. I knew that the IC developer community was going to come up with a lot of great applications based on user-generated content and I wanted to create something that would be useful to the community. I saw a way that we could contribute by developing a solution that would help developers focus on their product, and make the IC a platform where Web3 citizens could work, create, and connect safely. This solution evolved from an initial idea to become MODCLUB.

A central challenge for Web2 platforms that work with user-generated content is how to set and enforce rules to maintain the integrity of their culture and user experience. The need for moderation and Proof of Humanity (POH) on decentralized platforms is even more important since spammers and trolls have access to unlimited verified identities. There is an increased risk of bot swarms because it is easier to create a wallet address on Web3 than it is to create a traditional email or social media account. MODCLUB’s hackathon submission is an AI Abusive Language Detector. We can leverage AI to enhance our moderation solution, acting as a first layer of defense. Our AI can detect a wide range of abusive language that can be immediately rejected without human review.

We hope MODCLUB and our Supernova Hackaton submission will help the IC developer community to continue to focus on building great products!

Welcome to the club!

What it does

MODCLUB’s mission is to clean up Web3.

MODCLUB provides a Moderation-as-a-Service and Proof of Humanity solution for Web3 applications. Moderation-as-a-Service allows an application team to protect users and maintain trust within their community while “outsourcing” the responsibility and labor involved in the daily business of moderation. Just as a company might use an API to handle or execute a task or deliver a service they don’t want to build out themselves (e.g. maps, search, user management, etc.), applications deployed on the IC can integrate with MODCLUB to comprehensively handle their content moderation activities. Moderators can earn rewards, and developers can control and review how their content is vetted at a cost far lower than doing it themselves. Over time, we aim to build a deep pool of trusted moderators and customer applications so the former can be well rewarded and, later, well served, as well as an AI layer to automate the bulk of the work. MODCLUB’s POH ensures an actual human is associated with an account. Our POH process is flexible, modular, transferable and can seamlessly integrate with other decentralized applications. Our AI Abusive Language model is a radial basis function Support Vector Machine. It’s hyperparameters are optimized using a Gaussian Process regression. The features input into the model are unigrams and bigrams and the inverse document frequency are the feature values. Example features are regularized by the square root of the examples sum of feature values in order to handle larger sentences with many non abusive words. There is also a custom lemmatization preprocessing applied to the text, which maps terms to higher probability terms with the same root word.

How we built it

MODCLUB platform was built completely on-chain. All functionality from frontend to backend is hosted on the IC in order to achieve a fully decentralized solution. We have a multi-asset canister storage system that consists of a series of canisters all written in the Motoko language. Our moderation process employs the Schelling Coin mechanism. And our POH process was built to be modular. We built out an AI Abusive Language model leveraging AWS. First our lambda function auto pulls our UGC data from the canisters and passes them to SQS. The second lambda function uses the AI algorithm to decide the abusive score of the content. The same lambda function then passes the data to the third lambda function using SQS. Finally the last lambda function makes the final decision to accept or reject the content based on the specific threshold levels we have set and sends it back to our canister.

Challenges we ran into

Working on any type of AI solution is challenging. Given our deadline we were not able to implement the solution on-chain. To circumvent this issue we leveraged an off-chain solution using AWS, while still designing our own abusive language AI model. Another challenge was finetuning the sensitivity of the model. It took some time to reach the right configuration for approving and rejecting content accurately. Our final challenge was restructuring our task distribution system so that we could pre-process content by AI, having only uncertain results be evaluated by our moderators.

Accomplishments that we're proud of

Our AI can help MODCLUB scale to handle a large amount of data for content moderation as we grow. We are very proud and excited to bring AI to MODCLUB. This is just the first step as we plan to bring a full fledged AI solution on-chain with more support and functionality for not just text, but images, videos and Proof of Humanity.

What we learned

We learned a lot as a team during our Supernova Hackathon experience. We learned how to build an off-chain integration with MODCLUB, and how to create an AI model for abusive language detection!

What's next for MODCLUB

We have lots in store for MODCLUB’s future! We are committed to improving and bringing our AI solutions on-chain and expanding its capabilities. All big players in the space use a combination of AI, assisted by human moderators and we intend to do the same, while leveraging the power of the IC. As MODCLUB expands, we intend to target not only the global content moderation solutions market, but also the global data labeling services, and the KYC and AML market. These markets will collectively account for $107.21 billion in spending worldwide as of 2031. Our POH solution will be scaled up to handle the demands of KYC and AML. And beyond UGC content moderation MODCLUB will expand to offering data-labeling.
MODCLUB will expand its customer application base beyond the DFINITY ecosystem. The platform is designed in such a way that it will be able to integrate with not only applications built on the DFINITY Internet Computer, but also cross-chain applications and traditional web applications.