Its patented Blockchain technology, advanced algorithms and pioneering globally distributed network enable operators, carriers and vendors to eliminate business inefficiencies, operational challenges and financial issues in today's and tomorrow's telecoms market.

One of the key features of this ecosystem is the interactive arena that intelligently utilizes social networking tools and the marketplace to enable real-time communications, negotiations and instant decision making. In addition, it has a centralized forum to evaluate, negotiate and create new business opportunities, maximizing business opportunities and capitalizing/implementing the most urgent needs.

Challenge

One of the concerns of the marketing department was to prevent users from posting inappropriate content on the networks. However, we wanted to avoid having to go through a prior moderation process. Firstly, because of the cost involved, and secondly, because this type of filtering incentivizes users to disconnect from the social network.

The aim was to create an adult content filter capable of detecting adult content in a way that is transparent to all users who upload content. The filter should detect inappropriate content, regardless of whether it has been uploaded as an image or as text.

Solution

The system, consisting of a REST API, uses different services:

• Image filtering service: it is responsible for analyzing the images (classifying them using the Cloud Vision API). If the images pass the filter, it detects whether they contain text. If it contains text, it checks the text through the text filtering service and retrieves the response to be shown to the user.
• Text filtering service: responsible for text filtering. This aspect is highly complex because it involves the different APIs used to ensure that they are employed correctly. First, the system must check in the Google translation API if the text is in English (so that the NLP API can be used). Otherwise, it must be translated for use in the NLP API because the NLP API limits us in the sense that it only classifies text into categories if the text is in English.
   

From the existing API's, the following API's will be used:

1. Vision API allows developers to understand the content of an image by encapsulating Machine Learning models in an easy-to-use REST API. In this way, it quickly classifies images into thousands of categories, detects individual objects and faces, and reads printed words embedded in images. Vision API can categorize the image catalogue to add rich metadata to the images, perform moderation, identify sentiments, etc... In our case, we will use, on the one hand, the detection of adult or offensive content, but we will also analyze the image in search of text that could be problematic (using OCR techniques).
2. The Cloud Natural Language API provides developers with natural language understanding technologies, including sentiment analysis, entity analysis, entity sentiment analysis, content classification and syntax analysis. This API is part of the Cloud Machine Learning family but has limitations. It can only classify text into categories in English, so if a text comes in a different language, it won't work. That's why we need to use the Cloud Translation API first.
3. The Cloud Translation API provides a simple programmatic interface and a neural machine translation system that allows you to translate the strings you want in any supported language. This API is highly responsive, as it can be integrated into web pages and applications to provide fast and dynamic translation from source to target text (for instance, French to English). Language detection is also available in cases where the source language is unknown. The underlying technology is constantly updated to incorporate improvements from Google's research teams. As a result, we can deliver better translations and continue to add new languages and language pairs.

Other features of this solution can also be highlighted:

• The solution runs on the Kubernetes engine. So, an adaptive window image will be stored in the Container Registry.
• The service is running on the Kubernetes engine and traffic is forwarded to the container via the load balancer.
• API Vision API is used with a secure search.
• Cloud Natural Language API classifies content.
• Cloud Translation API is used to check the language of the text to be parsed and translated into English if the source language is not English.
• Firebase contains a dictionary of over 1.7k English words/expressions that are not allowed.
   

Results

After a month in the testing phase, the new system analyzed and filtered more than 500 comments, flagging some of them as inappropriate content. All filtered comments were manually reviewed and more than 90% contained inappropriate images or phrases that would have been rejected by any human moderator.

On the other hand, for the time being, no messages considered inappropriate or that should not have passed the content filter have been found among the unfiltered messages.

Therefore, after the results obtained by the filter, the client has decided that it is no longer necessary to review comments manually. This result represents a significant saving in the daily time spent by one person moderating comments on social networks.