How Ubcoin Market modulates content using artificial intelligence
Ubcoin is developing now its own AI-based system based on deep neural networks to automatically moderate millions of variable content ads without a human factor and to minimize the legal and ethical risks for the business. Ubcoin is a unique crypto-to-goods exchange — a blockchain-based platform where everyone can buy or sell real goods for cryptocurrency. Ubcoin is conducting a public token sale now having raised more than 4 million dollars from private investors and some undisclosed amount from institutional ones — the Inventure fund and the Singapore-based Amereus fund. Ubcoin will be part of the Ubank mobile app that is pre-installed by Samsung in 10 countries and has over 16 mln installations worldwide. You can buy Ubcoin tokens with a discount now on https://ubcoin.io/en
Moderation of content on public marketplaces is an important task. Only a company that provides ethical and legitimate placement of content can generate steady growth and form a successful business model.
Lack of content moderation entails legal and image risks. For example, Facebook and eBay invested hundreds of millions of dollars into programs to track and block fake news. In the spring of 2018, a competition with a prize fund of $ 35,000 was held at the Kaggle site where 4551 teams from data analysis specialists solved the task of “Toxic Comment Classification”.
Three approaches to content moderation should be considered:
- Manual check. An expensive approach with low quality due to the “human factor”. Manual check is unacceptable in large-scale projects, such as the Ubcoin Market.
- Semi-automatic check and check based on rules. Semi-automatic check is only applicable to structured, highly specialized content. For platforms with high content variability, you can not manually set all the necessary rules.
- Approach based on machine learning. A forward-looking approach that uses statistical data to classify content. It is the approach, which is based on artificial intelligence that we use in the Ubcoin Market service, since it is devoid of the human factor and is resistant to the variability of content in the presence of a relevant training sample.
Tasks solved by artificial intelligence Ubcoin Market:
- Verification of content for compliance with the requirements of law. An artificial neural network detects content by category: drugs, weapons, pornography, etc., and either immediately blocks the advertisement or sends it to the moderator for a specialist.
- Ethical review. Ubcoin Market does not accept placement of materials containing toxic comments, racism or discrimination.
- Detect duplicate ads. Often, duplicate ads are placed by scammers, and therefore it is necessary to track copies of texts and images.
The principle of how artificial intelligence operates within the Ubcoin Market
The basis of artificial intelligence within the Ubcoin Market is established with three machine learning models.
- The processing of ad texts is carried out using a model based on a deep, bidirectional recurrent neural network and an algorithm for translating text data into a feature vector — Word2Vec.
- Deep neural networks with a ResNet-based architecture pre-trained on ImageNet images and additionally trained on their own training sample are used to process images and to reveal the content of the inscriptions on the image.
- The output uses the XGBoost algorithm, which is based on the composition of decision trees and takes into account variative data.
The Ubcoin Market team tests GAN algorithms (Generate adversarial network). GAN is an approach proposed by Google in 2014, in which not one neural network is used, but two that “teach each other.” Generative adversarial network will allow to reach a qualitatively new level in the classification of content.
Ubcoin Market will also use artificial intelligence for the tasks of personal recommendations and the fight against fraud. To do this, it is poised to collect data on the behavior of users on the platform and to build predictive models based on them.
The problem of models of machine learning is their “obsolescence” with time. In order for models to be relevant and give the best quality of prediction, a regular update of the training sample and model parameters is required. To do this, the Ubcoin Market team plans to use the data from both open source and work-related platforms, with markup on platforms such as “Amazon Mechanical Turk”.
The process of artificial intelligence Ubcoin Market is carried out as follows:
To train an artificial algorithm, a training sample is assembled, consisting of relevant examples. After “teaching” the algorithm, the input algorithm is supplied with text and graphic data. At the output, we get the value of a class that demonstrates whether the ad contains signs that indicate a violation of the rules.
Ubcoin Market implements state of the art approaches such as using technologies like artificial intelligence to ensure the company’s competitiveness in the 21st century.
Artificial intelligence based on deep neural networks allows Ubcoin Market to automatically moderate millions of variable content ads without a human factor and minimizes the legal and ethical risks for the business. You can buy Ubcoin tokens with a discount now on https://ubcoin.io/en