The scenery of the sea is very good, beware of being trapped in the "cultural" reef. Follow【Rongyun Global Internet Communication Cloud】Learn more <br>Recently, overseas e-commerce, one of the largest online shopping sites in Southeast Asia, has caused a lot of uproar. The platform featured a transgender influencer with mixed reviews and a woman in a wheelchair as the protagonists of the ad, which is believed to mirror the Thai royal family who spends years in wheelchairs and discriminate against people with disabilities.
Although the platform has removed the advertisement and made a solemn apology, the people who felt hurt did not buy it. The advertising turmoil not only alerted the Prime Minister of Thailand, but also led to the boycott of the platform by the military, prosecution by government departments, and even an official response from the Chinese embassy in Thailand.
As the Thai government spokesman emphasized in response to this incident, "content creators should be careful about the content they publish online." Risky content that should have been filtered out in the review was not checked due to cultural differences, and never intended to cause such a huge wave.
And companies going overseas will face such "acclimatization", and content review has become a link that cannot be ignored.
As a communication cloud partner of 300,000+ apps worldwide, Rongyun is not only based on the communication network (SD-CAN) and self-developed optimal link scheduling algorithm in 233 countries and regions around the world, but also provides users with safe and reliable services. Communication cloud services, and also cooperate with professional partners to create overseas content auditing solutions , providing multilingual intelligent content risk control services for global social and pan-entertainment applications.
The first level of overseas nuggets to overcome cultural "acclimatization"
For companies going overseas, sensitivity to local culture and knowledge reserves are necessary conditions for risk control. In Southeast Asia, which covers multiple religious cultures, more caution is needed, especially in the social realm. With the rapid development of the industry, social products are characterized by a large number, variety, rich forms, and diverse content. With the surge in traffic and interests, they are also accompanied by a series of regulatory risks. Content security prevention, control, and clean-up, various links All need to be strictly checked.
Social products have higher requirements and challenges for content review capabilities, mainly in the following aspects:
1. The complexity of regulatory requirements varies from country to country
For overseas markets, under the influence of complex factors such as law, religion, culture, and language, the compliance of overseas apps must meet both local laws and local cultural habits, which brings new challenges to content review.
Taking the European and American markets as an example, Internet content adopts a strict classification system, and there are strict protection strategies for minors of different age groups such as 3 years old, 13 years old, and 17 years old, and racial speech is expressly prohibited; in the Middle East , Religious elements such as crosses and pigs are extremely sensitive; in Japan and Thailand , it is necessary to avoid expressing dissatisfaction with the imperial system and the royal family, and South Korea’s political remarks against the North Korean government are also strictly censored locally.
2. The need to protect the brand image
Today, when online social networking has become mainstream, how to create a clean and healthy social environment for users is a top priority that affects the brand image, which in turn affects the success of the app.
According to the survey, after users engage in interaction in a healthy environment, the retention rate will increase by 3 times, the session length will increase by 60%, and after being harassed by the Internet, 15% of users will turn to other social platforms.
Different countries and scenarios Different types of overseas content risks and complicated details
Cultural differences are reflected in various aspects. For example, the laughter of Southeast Asian countries is very different from that in China. In terms of atmosphere, their live broadcasts will appear more "noisy". In terms of content auditing, in addition to the differences brought by different regions, even in the same region, different application scenarios also face great differences in auditing.
For example, in social scenarios, words such as Kill and Shoot are not allowed and must be identified and avoided; while they appear in game scenarios, it is reasonable.
For another example, in game nicknames, companies generally set special rules for the use of words such as Admin and Operator, because they are worried that users will disguise themselves as software operators and bring risks such as fraud; they do not need to be special when they appear in chat. Notice.
This requires a professional and comprehensive content review service. For apps with overseas social attributes, the risk types of content mainly include pornography-related, political-related, drug-related, violence-related, race, discrimination, religion, minors, abuse, pull Human advertisements , etc., the above are the objects that the network regulatory authorities of various countries focus on.
Focus on overseas social professional, comprehensive risk control strategy
On top of the dual-core capabilities of IM instant messaging and RTC real-time audio and video, Rongyun adds X elements - the sum of peripheral capabilities of communication, fully empowering developers. And "audit" is one of the X capabilities of Rongyun's key layout.
In this regard, Rongyun cooperates with leading professional partners in the industry to help developers in social scenarios to accurately identify illegal content such as pornography, violence and terror in text, pictures, audio and video in instant messaging and real-time audio and video services. Meet the needs of all scenarios; provide services such as auditing, marking, and custom configuration to meet enterprise needs through digital means, and create triple security guarantees for business, content, and users .
(Content review and detection category, click to enlarge)
- Support Chinese (Simplified), Chinese (Traditional), English, Japanese, Korean, Thai, Vietnamese, Malay, Indonesian, Hindi (Latin Hindi), Filipino, Arabic, Turkish, Russian, Text and audio review in 24 standard languages: German, French, Italian, Spanish, Portuguese, Swedish, Polish, Greek, Norwegian, Dutch.
- Supports text detection for 5 native dialects in the Middle East, Egypt, Sharm, North Africa, Gulf, and Sudan, as well as Bengali, Urdu, Marathi, and Assamese in India.
1. Text review
Global deployment of text auditing, the audit results are responded within 100ms, strategies can be formulated according to the audit standards of different overseas countries, and expressions such as standard language, dialect, slang, abbreviations, social cryptic language, etc. can be supported, and contextual semantic recognition can be used to avoid false detection, And keep up with the latest policies and current political news of various countries, and constantly update the algorithm model.
(text review example)
2. Image review
Use advanced target detection and content recognition algorithms to detect illegal content in pictures in real time, support cartoon scene and real scene recognition, support complex scenes such as "picture-in-picture", avatar thumbnails, blurred pictures, etc. etc. have strong detection ability.
(image review example)
3. Audio Audit
Focusing on social and other scenarios, it can accurately identify and review 23 languages, and use the ASR+ text model and voiceprint model to comprehensively judge content involving pornography, abuse, violence, terrorism, politics, advertising, hate speech, and minors. Effective Resist the risk of audio content violations, with an accuracy rate of over 95%.
(Applicable scenarios for audio review)
4. Video review
Accurately identify video content related to pornography, politics, violence, terrorism, advertising and other video content in various scenarios, resist violation risks, ensure video content security and compliance, support intelligent definition of frame extraction strategies , and improve auditing while ensuring detection accuracy Efficiency, cost saving, integration of EAST, CRNN, efficient-net, Resnet and other models, avoiding manslaughter and missed detection of a single model, and high recognition accuracy.
5. Multilingual Minor Protection Policy
Identifying and detecting content that violates minors is a major feature of overseas auditing in Rongyun's content auditing program.
On the user side, the age, face and tone of minors can be recognized;
In terms of content, it blocks sensitive content such as vulgar pornography, bloody violence, ideological poisoning, and personal information for minors, helps the platform to establish an intelligent content recommendation program for minors, and protects the privacy of minors through various strategies.
(Minor Protection Policy)
Practical case
A domestic social app uses Rongyun's complete domestic audit service. When the business enters the global development stage, domestic audit alone is not enough to support complex scene requirements, so the overseas text + video + picture audit service is opened, which solves three problems in one fell swoop. Big pain point:
❓ In the process of user chat, comprehensively judge and block sensitive words in combination with language, context, variant, etc.
☑ Solution: Multilingual text review, intelligently block sensitive words.
❓Identify videos involving pornography and violence, especially for scenes that are difficult to identify, such as picture-in-picture pornography, the edges of pictures, and sensitive parts exposed under blurry light.
☑ Scheme: Identify sensitive content through automatic frame capture-submission for review-image review service; among them, the image review service solves complex scene problems through targeted data enhancement.
❓In the process of image recognition, in addition to domestic pornography, violence, and politics, some countries also require pictures of tongue sticking out, smoking, and foot pictures, and recognize complex scenes, such as long pictures, posters, and picture-in-picture , comics, etc.
☑ Scheme: The image review service enables fine-grained models such as tongue out, feet, kisses, pictures and texts, pornography, and politics to identify sensitive images.
For social applications, the issue of content security is a life-and-death issue that cannot be overemphasized at any time. Rongyun joins hands with professional platforms in the industry to help developers going overseas to communicate smoothly and worry-free globally, face the broad market, and take good care of the "security checkpoint".
**粗体** _斜体_ [链接](http://example.com) `代码` - 列表 > 引用。你还可以使用@来通知其他用户。