The "Thirteenth Five-Year Plan for National Industry Development" clearly pointed out that it is necessary to accelerate the integration of key areas through the promotion of "Internet +" entrepreneurial innovation and the application of digital creativity in social networks, and vigorously develop new entertainment marketing models such as virtual social networking. Digital creative ecosystem. As a result, the "Internet +" new model of social applications has officially emerged, which has also led to the mass launch of diversified social software such as live video and virtual dating.
Among them, the diversification of social software is the marketing core of the industry ecosystem. It is not limited to the diversified gameplay of social software itself, but also involves the diversification of user groups and social content, such as video, audio, emoji and other chat modes. With the continuous growth of downloads of this type of social software and the continuous accumulation of active users, it is also accompanied by a higher degree of various online business risks.
In response to the trend of illicit property, pre-blocking and content supervision
According to the 2020 social network software industry research and analysis report, the social network software industry has a market size of 120 billion, a customer population of 60 million, a usage analysis of 90 million, and a compound annual growth rate of 14%. It can be seen that the social network software industry has a large scale of coverage, a high proportion of service users, and the huge market scale shows considerable demand for business opportunities.
With the further improvement of national policies, social software is also innovating and changing according to different user needs, and continuously optimizing and upgrading its own products and service quality for different demand scenarios. But it is worth noting that under the diversified development of social software, some pain points that are inevitable for social software platforms have also been highlighted.
At present, social software mainly has problems such as content violations, ad diversion, network fraud, and user experience impact. The reason for this is that in addition to content violations that users may involve themselves, it is more due to active black production and multi-channel scalping of cash to attract money... This has caused a series of risk control loopholes, which seriously affected the normal operation and commercial value of social software. , And bring huge economic loss.
Sumei Technology has been committed to providing professional AI business risk control services for global customers for five years. It has currently served more than 1,000 customers worldwide. In addition to the social field, it also covers banking, insurance, aviation, e-commerce, video, education, Games and other fields, and maintain an annual growth of more than 300%, and the renewal rate of old customers exceeds 130%.
Seek doctors first, and combined with the current operating status of social software and existing problems, Shumei Technology Black Production Research Institute found the following problems:
Bulk account maintenance: The virtual machine performs a large number of account registrations, and sends a large number of advertisements/product promotion information to the user without the user's permission.
Content violations: The content (pictures, videos, audios) posted on chat windows, avatars, Moments, and live broadcasts involve violations and affect user experience;
Advertising diversion: There is information leakage, which makes users' personal information effective data for criminals to commit fraud. The black property makes profit from the one-stop service of "draining script + station software + account reselling";
Internet fraud: Aiming at target user groups with a certain economic foundation and emotional needs, the black industry uses frequent chats and the temptation to draw cakes to deceive.
In essence, the source of these problems is mainly due to the high-frequency activity of black industry on social software platforms for their own selfish desires. They use a mature and complete clear division of labor to steal commercial benefits using various methods to accelerate the decline in the revenue of the software platform. To seek rapid growth in personal income.
After in-depth research, the Black Industry Research Institute of Shumei Technology also found that the black industry currently uses 7*24 hours of real-time monitoring, and has carried out a global layout for fraudulent activities, and the number of people is also growing rapidly. The "White Paper on E-commerce Ecological Security" shows that, according to estimates, the number of black industry employees reached 1.5 million, and the market scale reached 100 billion. Therefore, the large-scale crimes committed by black companies should receive the attention and attention of social software.
How to solve the above problems? Sumei Technology believes that on social software platforms, whether it is the user’s APP startup, registration, and login links, or in the context of content production scenarios such as pictures, audio, and video in the dynamic and chat windows of the square, it cannot supervise the human review of the social software platform. In all aspects, defense mechanisms such as device risk SDK and verification codes can be deployed through AI to achieve intelligent interception and real-time reminders in advance.
At the same time, by deploying a full-stack real-time intelligent risk control engine-Skynet, and a full-stack intelligent content recognition engine-Tianjing, similar risk control problems can be solved efficiently. The advantage of SUMEI Technology is that it strengthens the level of barriers in the software risk control business link and focuses on cracking down on a series of illegal behaviors of black production, which can maintain the normal order of social software operations and provide users with Come for a better social experience.
Apply AI intelligent risk control engine to fully guarantee the security of online social content
In the final analysis, it is a question worth pondering how to develop the new business format of industry services in a sound manner and realize the social network software industry to move towards a higher quality and fairer direction.
In recent years, with the increasing popularity of social software by teenagers, the “Z era” people’s ability to realize their social needs, entertainment needs, and consumer demand with high intensity has ignited the bold innovations of the late wavers on pan-entertainment platforms. Therefore, UGC content security on social software is one of the primary concerns of the social industry.
Therefore, in response to the regulatory needs of target customers, from the scene application side, Sumei Technology is based on domestic leading AI technology (GAN, TDNN, LSTM, RNN, etc.) and the global SaaS risk control service network, and deeply integrates the full-stack model system. Real-time defense and precise identification provide a comprehensive and safe high-factor risk control solution for social software platforms.
In terms of risk control capabilities, the global SaaS AI risk defense network has identified and intercepted business risk behaviors more than 100 million times a day. In terms of data volume, the current Sumei global anti-fraud service has implemented 6 service clusters around the world, protecting 3 billion users worldwide, 110 million daily interception risk behaviors, cumulative equipment coverage 5 billion, and daily anti-fraud visits 100 Billion, with a cumulative visit volume of 6 trillion.
In terms of business cooperation, Sumei Technology insists on putting customer needs first, deploying domestic and overseas business through self-developed AI technology, and continuously breaking through the number of existing service users in the billions.
First of all, at the location where users register and log in to the software platform, use the SUMMIT-full-stack real-time intelligent risk control engine for supervision. It is possible to detect black production beforehand, increase the cost of black production and store accounts, and block 1,000 accounts every day.
In terms of technical advantages, for APP startup, user registration, login, and then to specific social business behaviors, in this full-process scenario, Sumei Technology adopted a full-path real-time deployment control strategy.
In addition, Shumei Technology has also realized the closed loop of the whole process of strategy evolution in core technology. Such as offensive and defensive research (black production intelligence analysis and path reproduction), strategy design (law discovery feature extraction), strategy research and development (data mining model training), strategy verification (offline verification and online grayscale verification), strategy online (effective in seconds) ), strategic operation (interception monitoring and early warning and case sampling detection), case analysis (missing killing and manslaughter case analysis) full-flow strategy closed loop.
For example, at the user registration end, once data is accessed to the registration interface, two auditing methods can be generated. One is the release of normal users, and the other is that high-risk accounts are refused release and recommended for manual review. Such accounts can be immediately blocked once they have released advertising information in batches and repeatedly.
At the same time, a time-domain correlation network is established based on mobile phone numbers, equipment, and IP, and unsupervised algorithms such as community discovery and risk communication can be used to discover black production groups and identify potential new fraud threats. By establishing data models (risk data, RNN time series scoring model, community discovery model) and using account risk levels to optimize product operation strategies, activity ROI can be improved.
Secondly, in scenarios such as user private chats, square dynamics, and video matching pre-cutting frames, Sumi Tianjing-intelligent recognition can be used to intercept, reduce the pressure of manual review, reduce manual review costs by 80%, and improve posting efficiency. The compliance content was increased to 99.9%, which significantly improved the user experience.
From the perspective of image recognition on the Moments/Square dynamics, the text content has the characteristics of diversity and complexity, from basic variants, homophones, adjective pronouns, to tens of thousands of leader names and false advertising variants, not only The quantity volume is numerous, and the illegal forms are treacherous and concealed, which poses a very big challenge to the "smartness" of the content recognition system, but Sumei Technology can solve them one by one.
In terms of technical strength, SUMMIT's intelligent text recognition engine is based on in-depth research on natural language, combined with machine learning technology of deep learning, namely fast text and NLP semantic analysis, HMM's intelligent word segmentation, and it is used for hundreds of billions of sample databases, Multi-model training for millions of illegal words and tens of thousands of variant types. So as to achieve multi-dimensional collaborative AI of content, behavior, and portrait, accurately and effectively identify illegal content.
At the same time, Sumei Technology can also dynamically monitor the trend of violations in real time, realize second-level updates of the lexicon of violations, and hour-level updates of incremental models to deal with the risk of new content violations that may be added to the Internet at any time. And helped large e-commerce social platforms reject 4.44 million text violations and 1.05 million photo violations.
Third, in scenarios such as one-to-one voice matching and group chat rooms, SUMEI audio streams can be used for real-time monitoring; in scenarios such as one-to-one video matching, SUMEI video streams can be used for real-time monitoring services. In this way, it can effectively avoid the risk of uncontrollable violations of social software in situations that cannot be fully covered by human review, leading to software removal. At the same time, through the machine review + review configuration strategy, 30,000 hours of audio and 7,000 hours of video can be filtered every day to effectively optimize the user experience.
In terms of core advantages, through the highly integrated deep learning technology, Sumei Technology can also use speech signal processing delay neural network (TDNN) and long short-term memory network (LSTM) to reduce noise and reverberation of original audio, and adopt Intelligent speech recognition engine, voiceprint recognition engine, timbre classification engine to get the recognition result and gender label, timbre label and voiceprint information.
Through the implementation of the above-mentioned Sumei AI business risk control overall solution, the content security protection level of social software can be greatly improved, and the customer content risk control system can work efficiently, thereby effectively saving labor costs.