In 2018, the entertainment and audiovisual industry first felt a wave of regulatory cold wave, a large number of well-known pan-entertainment products were subject to regulatory interview. Content security has become the industry has to pay attention to the Top problem, the construction of audio-visual system can not get rid of content security, audio-visual industry practitioners also need to pay more and more attention to content security. LiveVideoStack email interviewed Zhu Haoqi, CTO of netease Cloud ESHIELD, a leading domestic content security manufacturer, to share the technical status quo and future challenges of content security.
Planning/LiveVideoStack
LiveVideoStack: Please briefly introduce yourself and your responsibilities? What’s the special story compared to the rest of the industry?
Zhu Haoqi: netease Cloud Security (ESHIELD) is a one-stop security service under netease Cloud, which was originated from the security department of netease. In the past 20 years, we have mainly provided security risk control support for all netease products, covering content security, mobile security, network security and business security.
Netease cloud security and other security vendor, we have been business scenario driven security, for example content security team is encountered when the UGC product such as netease blog content of hazardous waste are organized a large-scale attack, under the netease group of high-level decision-making, quickly set up, we should be the first big domestic companies to set up this kind of team. I am mainly responsible for the technical research and development of cloud security services in E-Shield, especially the content security services related to intelligent identification and the business security services based on big data.
LiveVideoStack: it is said that from netease news to netease cloud music comment culture, you are doing support behind, can you share some of your deep impression? Or something fun?
Zhu Haoqi: The information technology revolution has made the social division of labor and coordination more and more close, let alone within the same company. Therefore, as the basic security support department of netease products, we just did what we should do. But there are a lot of details that need to be discussed and understood when helping products build content security systems. For example, netease has the feature of building buildings, but there is a possibility that some floors have problems while others have problems. Then we need to discuss with the product to increase the function of building demolition, so that we can remove (delete) all floors that we think have problems. This is actually a big change for the whole follow-up system, and the company did invest a lot of resources to upgrade the architecture.
In addition, cloud music at the beginning of the design drew lessons from the problem, the design does not allow multi-storey buildings. (Two layers at most). But with Cloud Music, we went a step further, and in addition to cleaning up harmful content, we also tried to do some “operational needs” functions, such as identifying and cleaning up content such as meaningless flooding and abusive content that disrupts the community atmosphere. The result is not bad, in fact, netizens on the Internet will follow the ethos of the community and adjust their behavior.
LiveVideoStack: The goal of LiveVideoStack community is to gather the first technical community in China focusing on multimedia audiovisual industry. Since its establishment, we have indeed verified the huge demand of the industry and engineers in this industry. What special insights does netease Cloud Shield have in the audiovisual field?
Mr. Zhu: We are very optimistic about the development of the audio-visual industry and the future of the multimedia audio-visual technology community represented by LiveVideoStack. Netease has always had explorative products in entertainment multimedia, but Yi Dun feels the outbreak of audiovisual industry from a special side. Essentially our content security service is content recognition. We currently identify about a billion pieces of information a day. Since our inception, we have seen an explosion in the proportion of images, videos and audio.
Our conversations with our service customers demonstrate that the content industry continues to evolve towards richer media and social interaction, and constantly expand new experiences for users. In the past two years, three trends have been highlighted in the industry. The first is vertical penetration. Audio-visual interaction is being embedded in tourism, finance, e-commerce and other industries as a standard function. In addition, the integration of different media forms is speeding up. New audio-visual content is constantly added to media, and short videos are launched in music applications. Pan-media and pan-entertainment has become a consensus. Finally, more multimedia forms, more interaction and mass content also bring greater and more complex content security supervision pressure to audio-visual platforms, content security must be intelligent.
LiveVideoStack: Many of our members are audiovisual platforms, and they are really interested in how to build an efficient content security system. Do you have any suggestions?
Zhu Haoqi: In terms of technology, intelligent recognition technology based on natural language processing and deep learning technology should be adopted as soon as possible to transform the traditional content recognition technology represented by keywords and feature database to ensure the coverage of text, pictures, voice, especially live video. In terms of products, attention should be paid to the coverage of content release, especially UGC interactive interface. In addition to comments, blogs, albums, bullet screens, etc., signatures such as nicknames, profile pictures and other corners should not be neglected. In terms of strategy, we should establish a precise and clear audit system and response process for harmful information, and make emergency response.
LiveVideoStack: From text, image, video and audio, the difficulty of understanding information is also increasing, can you share some of the technical exploration that you and your team are doing? Like NLP, voice print recognition.
Mr. Zhu: It is true that information is becoming more difficult to understand, but the effective technology has been updated in recent years. On the one hand, we try to understand the increasingly complex multimedia content by means of the integration of various technologies. For example, in the process of picture analysis, we will use the combination of multiple deep networks to extract the semantic information of pictures. Meanwhile, we will use NLP technology to understand the emotion and meaning of the text information captured by OCR. Finally, the comprehensive understanding of images and characters is used to determine the information effectively. On the other hand, due to the increasing content of Internet video, we use video analysis technology to better interpret the content of streaming media on the basis of static image processing. In addition, voice print recognition technology is also an important research topic in the field of security, and we are also trying to use end-to-end voice print technology to assist identity authentication.
LiveVideoStack: What has been the fun of working in security for so many years?
Zhu Haoqi: Security is a serious and rigorous industry, a small problem may lead to a serious crisis of the enterprise, in fact, should not talk about fun (compared with games, social networking, e-commerce and other fields). Chances are you’re having fun at the expense of others (it’s fun to research safety issues, learn the details of what happened, think about remedies, explore new risk points, etc., but that’s usually after a product has been notified of a safety problem). Of course, the sense of achievement brought by helping customers solve security problems is very strong, and the battle of wits and courage with black production is also considered to be happy in pain. We’re trying to solve something that’s “against humanity.” It’s hard to do, but it’s fun to do.
LiveVideoStack: can you briefly review and forecast the history and future of security systems?
Zhu Haoqi: The history of information security before the invention of computers was mainly in the field of cryptography, such as encrypted telegraphy and so on. With the development of computer technology, the field of information security is expanding, and now it has included network security, system security, mobile security, business security, content security and so on. At the same time, the threat of information security is also constantly upgrading, security incidents occur frequently, and the challenge to security work is getting bigger and bigger. I think we will see the same trend in the future: both the scope of security and the threat of security will increase.
LiveVideoStack: Security departments tend to have a low profile in enterprises, and for many management, security is a cost center rather than a profit center. But safety is no small matter, a small mistake can bring huge losses to the company and customers. What do you think of security as a special industry? Memorable stories to share.
Zhu Haoqi: The threshold of security industry is relatively high, and there is indeed a thankless suspect, so the circle is not as broad as other information technology industry. But with the occurrence of security incidents, more and more business leaders and government officials have begun to realize the importance of information security, and the relevant laws and regulations have been constantly improved. I think the status of the security industry is bound to rise over time. After scale, the security industry can also have commercial exploration, which can not only help companies solve security problems, but also become a source of profits.
LiveVideoStack: Inside netease, is security a profit center? It’s opening up internal technology to the outside, right? Why the move?
Nicholas qi: specifically, netease internal security work is divided into two parts, one is the basic technology research serves for the group, is a commercial security products easy to shield team, commercial department is a profit center, it is the netease 20 years of accumulated some internal technology for enterprises in need, the formation of commercial service. I think it’s a logical thing to do. On the one hand, the technology needs to be implemented in more scenarios to prove and maintain its advanced nature. On the other hand, the commercialization of high-threshold security services can quickly meet the needs of the security market, help enterprises in need to quickly realize the construction of security system, reduce security risks, improve the efficiency of enterprises and reduce their costs.
LiveVideoStack: Now all industries are talking about AI, what is the status quo of AI application in the field of security? What is the industry optimistic exploration direction?
Zhu Haoqi: In fact, AI technology has been around for a long time, but in recent years, with the help of new algorithms and new high-performance hardware, the effect has made significant progress. In certain areas, it can even achieve more accurate and efficient results than human. We are currently applying AI technology to content security and business security, using better models to identify content and identify risky behavior. As for the direction of AI exploration, there are too many, such as speech recognition, translation, robot, face recognition, automatic driving and so on. It can be said that the hot direction with high capital investment is promising.
LiveVideoStack: in order to better distinguish between normal users and threatened users, users need to do portrait analysis, at the same time how to protect users’ privacy, what work have you and your team done?
Zhu Haoqi: Actually, the user portrait only needs the data of each dimension of the user, but it does not need the plaintext data. As long as the data is consistent for the model. So we are on the one hand, the data secrecy are established within the department, according to the data importance hierarchy permissions, another aspect as far as possible the real data to do the desensitization treatment, such as mobile phone Numbers, hash directly into another value, can’t reverse, people can not understand, even if such data reveal that don’t cause the problem of privacy.
LiveVideoStack: NOT long ago at MWC in Spain, 5G was hot. Expect 4K and even 8K video to become ubiquitous in the near future. Will this create new challenges for security services?
Zhu haoqi: The theoretical speed increase of 5G will definitely bring benefits for bigger and clearer content transmission, but it’s ok for content security, there won’t be too much pressure. However, the popularization of 5G will greatly increase the number of access devices, which may further reduce the threshold for hackers to attack, leading to a significant increase in security costs.