Agmis launches all-in-one AI platform for automated marketplace content moderation
Leading Baltic technology developer Agmis has launched a new AI powered platform for automated marketplace content moderation. Aimed at marketplaces, ecommerce and classified advert websites, EasyCMT titled platform uses AI to moderate visual, textual and video content. The platform automatically identifies content that goes against the policy guidelines, preventing it from appearing on the site.
EasyCMT platform was developed by Agmis product spinoff EasyFlow, which specializes in retail Computer Vision applications. By using AI for image analysis, the content moderation platform automatically identifies unwanted content in user uploaded pictures or videos, including nudity, weapons, drugs, and hate symbols. The platform can be augmented to detect watermarks, excessive usage of text and other unwanted aspects, further improving marketplace user experience. EasyCMT also comes with inbuilt tools for background removal and image enhancement.
In addition to image and video moderation, the platform analyzes textual content, preventing hate speech, offensive comments or potentially fraudulent postings from making it onto the website. For improved user experience, the platform also scans for ALL CAPS postings or usage of unwanted symbols.
“Marketplaces are the most lively corners of the internet. Think of a real world town market, where sellers are jousting for buyer attention. With fast growth of the user-base, unmoderated marketplaces are quickly overrun by spam and unwanted content. Poor user experience is the main reason why users choose to shop elsewhere” – says head of product development at EasyFlow Simas Jokubauskas.
According to Simas Jokubauskas, the EasyCMT aims to address two main challenges for marketplace owners. Firstly, the platform helps to keep unwanted and harmful content off the site. Secondly, it nudges third-party sellers to adhere to website-wide content policy guidelines for a unified and fluid user experience.
“Marketplace owners are faced with a balancing act between fast-paced growth and over-policing. Serving visual, textual and video mediums, our content moderation platform workflows bring the best of both worlds, with the ability to prevent unwanted content from appearing online, as well as reducing the manual oversight burden from human content moderation” – adds Simas Jokubauskas.
Head of product development at EasyFlow Simas Jokubauskas
With EasyFlow content moderation platform all user-generated entries are assigned a policy breach probability score. In a fully automated mode, AI prevents all unwanted content from appearing on the site automatically. Human moderators can overrule these decisions later on. Operating in an assisted mode, the platform will only block the content where it is highly certain that it breaches the policy of the marketplace. In the cases where it cannot make a decision with absolute certainty, the platform will prioritize suspicious entries for human moderators to review. In a manual mode of operations, the platform will not prevent any content from appearing. Yet it will prioritize the item review list for human moderators, with the most suspicious postings coming first in line.
“These workflows make a better use of human content moderator resources, shifting their attention to postings where their review is most needed, eliminating the requirement to manually check every content item on the website” – says Simas Jokubauskas.
EasyFlow has already made its mark in the world of ‘offline’ retail. EasyFlow team are the authors of Computer Vision powered applications for in-store analytics, including apps for shelf-level stock management, planogram compliance, checkout queue management and other tools for ‘brick and mortar’ retailers to improve their store performance. ScanWatch – an application for unmanned checkout counters – automatically identifies scanned unpackaged products (e.g. vegetables or baked goods) without the need to select them from a self-checkout picklist menu. It also helps to prevent unintentional product misscaning and mitigate checkout fraud.
“It’s only natural that with our experience in real-world image recognition, we are bringing this expertise into the digital retail realm” – says Simas Jokubauskas.