Years from now, when we look back on 2021, we will surely find that this is a year full of changes.
This year, the epidemic did not dissipate as expected, and by the end of the year, various anti-epidemic actions were still unfolding; this year, the tide of Chinese stock listings in the United States was suspended this summer, and adhering to the latest data security standards became a new issue for many enterprises; or in this year, many subdivisions have undergone earth-shaking changes. For example, online education has passively contracted under the trend of "double subtraction" and has become a commemorative coin in the changing situation of the times.
There is only the industry under the era, and there is no era under the industry. Whether it is supporting the fight against the epidemic in the face of the long-standing epidemic, or the IPO suspended for the protection of national security, or the industrial adjustment made for the next generation of development models, there is a trend behind this, starting from various dimensions and assuming social responsibility in many aspects, which is becoming a compulsory course for the growth of enterprises today.
In particular, Internet companies in the mature stage have taken the lead in iterating. In the past year, there have been many news of large Internet companies rushing to help fight the epidemic, are keen on public welfare and support common prosperity. Only for common prosperity, after the state put forward the goal of "common prosperity", Ali and Tencent have officially announced the "common prosperity" fund of 100 billion yuan, and byteDance, Pinduoduo, JD.com and other leading Internet companies have also announced corresponding goals and plans.
However, one point that cannot be ignored is that for ordinary individuals, the guarantee of material life is only one side, and the safety and stability of spiritual civilization life is equally important - since the birth of the first batch of Internet companies in China around 2000, the digital services and content provided by Internet companies have become indispensable carriers of people's spiritual world. However, as the content that fills the spiritual world is unprecedentedly abundant, hidden violence, pornography, and false advertising are increasingly becoming risk factors.
It is precisely because of this that with the improvement of laws and regulations such as the Cybersecurity Law and the Data Security Law, as well as the new content attack methods that enterprises continue to face, the construction of digital content risk control has also become a must for many enterprises.
This time, internet companies as providers of digital services and content have chosen to stand in front of the stage to provide their own value for protecting people's spiritual lives.
Digital content risk control in iterations
Frankly speaking, digital content risk control is not a popular word. But the origin of this industry is closely related to the entire wave of digitalization.
Specifically, the emergence of harmful information on the Internet can be traced back to more than a decade ago when blogs and forums were just popular. At that time, various content channels were just becoming popular, and netizens were free to comment. "At that time, in the user community, there was already harmful information that would be spread through static pictures." Zhu Haoqi, general manager of NetEase Yidun, traced back to 36Kr.
However, in the early 2000s, even if harmful phenomena have emerged, the Internet content model has just taken shape, the public's awareness of clear cyberspace is not strong, and the technical means at that time are not as advanced as the current one, so digital content risk control has not become a topic of general concern.
Nowadays, with the spread of digital content channels, digital content risk control has become a standard that enterprises have to consider in the process of digital transformation.
First of all, from the perspective of industrial demand change, in 2019 alone, many well-known apps have been required to rectify due to community content problems, audio APPS have been removed due to content problems, and some luxury brands have also caused public opinion due to inappropriate speech or behavior. From another policy perspective, since the official implementation of the Cybersecurity Law in 2017, laws and regulations on security (such as the Data Security Law, the Personal Information Protection Law, etc.) have been continuously improved, and content risk control has become the focus of attention of enterprises like many security subdivisions. Moreover, with the increasing progress of public consciousness, creating a harmonious cyberspace environment is also expected.
The rise of the industry, manufacturers are particularly touched.
Zhu Haoqi joined NetEase in 2008 and was responsible for the construction of the content risk control system of NetEase Group for more than ten years. In his feelings, with the frequent occurrence of relevant cases, the trend of strict management of policies, and the ardent hope of the general public, more and more customers have taken the initiative to realize the significance of building digital content risk control capabilities. But going back, until 2016, there are still many customers who do not understand the value of content risk control products.
"At that time, many customers thought that they might be able to solve the problem by simply doing a keyword search function, but this obviously did not work well." He said.
Digital content risk control itself has a high degree of complexity, and it is necessary to cross many thresholds to do a good job. In summary, doing a good job in this business requires manufacturers to have sufficient business understanding, technical/product capabilities and long-term perseverance and accumulation of practical experience.
First of all, from the perspective of technological evolution, when blogs were just emerging in the past, the technology related to digital content risk control was mainly based on text recognition, and the recognition ability for pictures was not high. In recent years, with the evolution of artificial intelligence technology, chip processing power and other technologies, coupled with the improvement of the underlying capabilities of cloud services, it has made it possible for products to remain stable in a high concurrency environment. The accumulation of such technical capabilities is one of the prerequisites for doing a good job in content risk control products.
In addition, another barrier to digital content risk control is a deep understanding of the customer's business.
In this field, in addition to the red line clearly stipulated by laws and regulations, the judgment of other content risks needs to be strongly combined with user scenarios. The difference between digital content risk control and other security products also lies in the product characteristics that are easier to judge due to the low threshold of understanding. The consequence of this feature is that many customer-side business personnel can put forward their own judgment dimensions for content risk control according to their own understanding, making it difficult for manufacturers to reach consensus with customers. In this case, digital content risk control manufacturers need to have a deep understanding of customer business scenarios in order to be targeted.
In summary, digital content risk control not only requires the technical capabilities of manufacturers, but also challenges manufacturers' understanding of scenarios and business experience. This also brings an interesting phenomenon from the side - even if the industry demand is gradually rising, compared with other security-related products and solutions, there are still not many manufacturers with digital content risk control as the main business.
There may be fewer manufacturers with both technical strength and scene understanding. From the overall observation of the industry, the leading Internet companies that started from the content field are the providers of digital content risk control capabilities.
NetEase Easy Shield: Optimize the content security system, so that enterprises can focus on business innovation
NetEase has attached importance to content risk control since providing blogs and mail products.
Early on, its security team focused on building content risk control capabilities to solve the corresponding problems in its own content products. In 2011, with the continuous improvement of the importance of security, NetEase's security department was officially set up as an independent department. In 2016, after the actual polishing of its own business scenarios, NetEase Security Department began to provide external services, and Yi Dun was also established, and content risk control is an important capability provided by it to the outside world.
In the past, due to insufficient technical capabilities, the industry often relied too much on manual review to ensure content security. With the increasing diversification of Internet products, enterprises have to publish hundreds of thousands or even millions of posts every day, the task of content moderation has increased sharply every day, and strong reliance on manual methods is doomed to not last. Therefore, efficient and safe content review has become an urgent proposition.
Therefore, in response to such needs, Yi Dun proposed a set of its own optimization schemes according to experience. Specifically, it proposed the content security guarantee system of "machine identification + security policy + manual review" earlier in the industry.
The core idea of this set of solutions is to standardize the recognition ability through technology in the face of a large number of non-standard scenarios, and at the same time combine with manual work to make the result judgment relatively customized. Zhu Haoqi introduced that this set of solutions means the improvement of efficiency and the reduction of traditional labor costs for customers, and for Yi Dun, it can also make products more standardized and more suitable for external services.
It can be seen that in the content security assurance system of "machine identification + security policy + manual review", "machine identification" undertakes the part of standardizing the identification ability, and "manual review" is a relatively qualitative part. Then, the "security strategy" is like a bridge in the middle, which not only helps the human team to feed back the characteristics and knowledge it processes to the machine, making the machine recognition more "smart", but also continuously filters the data after the machine recognition according to some policy tools, so that its result output is more accurate.
In fact, before 2012, the initial team of NetEase Yidun began to polish the guarantee system of "machine identification + security strategy + manual review" internally; in 2016, NetEase Yidun officially served externally, continuously exporting this concept to the outside world and continuously iteratively optimizing; around 2019, "this guarantee system has become a common solution path in the industry." Yi Dun introduced.
In practice, for example, a smart hardware head manufacturer has a voice assistant product with a monthly life of more than 100 million. In the face of the astronomical magnitude of the Q&A corpus and the massive data added every day, it is necessary not only to consider cost control, but also to balance the amount of data, processing speed and accuracy, while facing the challenges brought by the changing content wind direction.
In other words, in the face of this situation, the ideal content risk control product needs to achieve a combination of efficiency and effect. After careful selection and evaluation, it purchased e-Shield's digital content risk control products and services to apply e-Shield's one-stop content security solution for human-machine collaboration to the business.
First of all, Yi Dun's intelligent machine audit system will conduct semantic analysis of the text based on deep learning and large-scale sample training, and distinguish the possibilities of pornography, vulgarity, advertising, politics, violent terrorism, insults, and irrigation that may exist. Through Yi Dun's algorithm capabilities, it can automatically and standardize the pre-processing of Q&A corpus text according to the type of violation, so that the bad and harmful content is automatically active. Machines solve ubiquitous audit problems, increasing efficiency and reducing costs. It is understood that Yi Dun will regularly optimize and iterate on the model, and retrain the model by collecting false and missed judgment samples to ensure the continuous iteration and evolution of the machine.
Then, Yi Dun went through a professional review to solve the complex problem. Specifically, after the Q&A corpus text has been reviewed by the machine, it is handed over to a human review team for secondary marking, intervenes when the machine is unable to make a decision, and corrects the decision. The advantage of this is that it can meet the non-standard but urgent business requirements of the customer, for example, when the project was implemented, the customer put forward the need to identify the positive and negative content of the government and the compliance of laws and regulations. Taking the compliance of the Advertising Law as an example, Yi Dun helped them comply with relevant laws and regulations, and eliminated the Q&A with the suspicion of "recommendation" in the knowledge Q&A on medical treatment, health care and health care, so as to make the recommendation of the company's intelligent products more neutral and objective.
At the same time, e-Dun's review platform also pays attention to the efficiency of massive content processing. For example, in the cooperation, customers can access the "intelligent audit platform" of E-Shield through API, process massive digital content on multiple lines, and dynamically carry out audit data query, quality inspection task issuance and other operations. It is understood that compared with the traditional audit platform, the "intelligent audit platform" of Yidun will assign tasks to AI and auditors, and help auditors get rid of complicated operations and further improve human efficiency through interactive design such as "automatic skip review", "high-volume word tool", "keyboard shortcuts", "full-screen review", "multi-person review" and so on.
Through this set of content security system that pays attention to both effect and efficiency, Yidun has digested the historical data accumulated by customers for a long time in a few months, and the average accuracy rate of Q&A corpus review is more than 99.8%. Equally important, while filtering out bad and harmful content, Yi Dun also re-labeled the Q&A corpus text after the machine review through the manual review team, and selected high-quality popular science Q&A to help customers improve the content value of their products.
In terms of more universal effect judgment, the cumulative detection data of Yidun digital content risk control products has reached 1,000 billion+, the average audit efficiency has increased by 10 times, and the anti-fraud recognition rate has reached 99%+.
Speaking of the differentiated advantages of Yidun as a whole, Zhu Haoqi feels that compared with independent manufacturers, Yidun's products are more stable and reliable, while compared with other head Internet companies, Yidun's solutions are more flexible and scenario-based.
To achieve these characteristics, it is still necessary to return to this set of "human-machine collaboration" concept, as well as the basic human-machine collaboration factor of Yi Dun. "We have been polishing our own content scenes for a long time, and it is precisely because NetEase has made so many content products that our content security capabilities can continue to be tested, providing customers with reliable services, so that customers can focus on their own business innovation (without worrying about digital content risk control issues." Zhu Haoqi said frankly.
The responsibility of the head company: to build an ecology and continue to protect the spiritual world of the people
Today, as the industry leader, Yi Dun also sees the demand for digital content risk control in the rapid growth.
For example, in the past, it was often digital native enterprises such as the Internet that attached importance to the construction of content risk control. However, in recent years, with the progress of digital transformation of the whole industry, more and more enterprises in traditional industries have begun to find Yidun.
"In addition to Internet customers, we now have a large number of retail, bancassurance and automotive customers." Zhu Haoqi said.
When demand gradually rises, supporting capacity should also keep up in time. However, the non-standard digital content risk control scenario, as well as the current situation that cannot be completely freed from artificial labor, have reduced the speed of industry development.
In this regard, Yi Dun's approach is to draw a meticulous ability tree for the profession of auditors by promoting the improvement of the construction of safety industry system standards, and also deeply participating in the arrangement of relevant vocational training courses, and promoting them to improve their own capabilities.
(The National Occupational Skill Standard for Network and Information Security Administrators, which Yi Dun co-drafted, has been officially released in 2020)
To be honest, in the first impression of many people, auditors belong to the entry-level work of the Internet, and the barriers are not high. However, Yi Dun tells us through his own 20 years of relevant experience that auditors are a profession with high requirements for comprehensive ability, and not everyone can be casually employed.
"The ideal auditor needs to have knowledge reserves from all walks of life, but also has a certain programming ability, and needs to feed the machine and deploy strategies at work." Zhu Haoqi believes that it is unscientific to rudely recognize the auditor profession as a "labor-intensive" industry, and even sometimes if the auditor's ability is not in place, the effect of the audit may be even less accurate than that of the machine. Moreover, in addition to the knowledge accumulation and programming ability, a high-level auditor must also have a deep understanding of his own business, so as to set reasonable judgment boundaries according to business requirements.
In a word, an ideal auditor should be like a security strategy person, completing the established responsibilities while becoming a bridge to feed the machine.
In order to achieve this goal with the industry, Yi Dun has been promoting the relevant state departments to formulate professional standards for the audit industry in the past few years. In terms of effectiveness, all auditors now have to undergo standardized training before they can take up their posts. At the same time, Yi Dun has also worked with relevant state departments and industry associations to formulate relevant courses.
Of course, opening up the education and training link is only the first step, and it is also necessary to implement this matter in a down-to-earth manner. In this regard, Yi Dun also cooperates with institutions such as the Disabled Persons' Federation to train more talents for the industry while also promoting the employment rate of the disabled. According to the company, there are many trained auditors who have been employed within Yi Dun and other Party A customers.
When the number of high-quality talents is increased, the service quality of digital content risk control will also be improved. Moreover, as the social value of content risk control is more affirmed, one trend that can be seen is that more and more small and medium-sized enterprises will also keep up with the pace of head companies and join the guardian action of people's spiritual civilization life.
Industry dynamics first became proof. For example, in December 2020, the 2020 China Internet Creditworthiness Conference with the theme of "Trustworthiness, Mutual Trust, Practice and Practice Together - Jointly Promoting the Construction of Online Creditworthiness" was successfully held in Qufu, Shandong Province, which aims to promote the construction of online creditworthiness, advocate network civilization, and help the construction of a comprehensive network governance system and the development of the digital economy from the perspective of strengthening self-discipline in the Internet industry. In July 2021, the 2021 China Internet Integrity Conference with the theme of "Gathering Strength and Winning Development - Building a New Era of Online Integrity" was successfully held in Changsha, Hunan Province, and more than 500 industry guests jointly sought the way of network integrity in the new era and jointly built and shared a new space for network integrity.
As a pioneer in content risk control, Yi Dun is also convinced of this trend. Similarly, in order to guard the online content space with these small and medium-sized enterprises, Yi Dun launched the "Qingfeng Action". In this action, Yidun will open a number of core functions related to content risk control for free for all kinds of small and medium-sized enterprises, hoping to make these emerging enterprises have the ability and willingness to participate in the protection of cyberspace civilization from day one.
For example, in the actual scope of capability openness, e-Shield will provide such customers with text, picture, audio and video content detection modules in content security, as well as functions such as registration, login protection, and user account identification in business security. As can be seen here, the content security module and the business security module are combined here. Zhu Haoqi explained that this is also because small and medium-sized customers may be brought by the same account at the same time to bring "wool" and "release harmful content" two types of negative impact, so the combination of content security and business security is a solution derived from the needs of enterprises, but also the substantive connotation of "digital content risk control".
Up to now, Yidun has served thousands of paying customers, including many well-known enterprises and institutions such as People's Network, Ministry of Foreign Affairs, Huatai Securities, China CITIC Bank, OPPO, vivo, Didi, Zhihu, B station, etc., and the service indirectly covers more than 99% of netizens. It is believed that with the development of the industry, more enterprises will choose to entrust complex digital content risk control issues to professional manufacturers such as Yi Dun to deal with.
When it comes to the observation of future trends as a whole, Zhu Haoqi feels that the general trend of digital content risk control is beyond doubt, and the industry needs to continue to coordinate the balance of efficiency and effect. But there is a little expectation that Yi Dun and himself hope to achieve - the goal of digital content risk control is to protect the harmony of people's spiritual world, so on the basis of meeting compliance requirements, perhaps everyone can discuss and collide with the most appropriate safety scale, so as to provide more colorful content forms above the red line.
"Security is often cold, but in the field of digital content risk control, maybe we can be more humane. On top of the requirements of compliance, we help everyone have a better content ecological experience. Zhu Haoqi said.
The lone traveler is fast, the multitude is far away. Whether it is the strict requirements for polishing products, or the tireless cooperation in breaking the circle, up to 20 years of hard work has shown that NetEase Yidun's long-term goal is to create a better life with scientific and technological innovation, and to meet the user's yearning and pursuit of a better life. It is believed that this mission is also the social responsibility that under the background of the times, head manufacturers need to continue to undertake and have the ability to continue.