Recently, short video apps have become the target of public criticism. Due to the related short video content involving minors' bad information, the National Network Office successively talked about the fast-handed and the heads of the volcano videos of today's headlines, and made serious criticisms and ordered a comprehensive rectification. After that, WeChat and QQ will also suspend the direct play function of the short video APP during the short video remediation of the Internet. The related APPs include micro-vision, fast-hand, vibrato, watermelon video and so on.
In the new media era, the content produced by users every day is exploding. In these newly produced content, there is usually a large amount of bad content, which will make users feel resentful and even violate the law. At this time, these new production contents are needed. The review has been carried out, and the traditional manual review method has not kept up with the explosive growth of content. In the era of artificial intelligence, how to build a content open platform has become a hot spot for the industry.
On April 21st, the theme of "Artificial Intelligence: Content Review and Traffic Anti-Cheat" hosted by DataFun, Data 猿 and Microsoft Accelerator was held at Microsoft Asia Pacific R&D Group Headquarters. This event invites Liu Lu, a senior algorithm research and development engineer who has participated in the construction of 360 and other companies' content open platforms, Zhang Xiaoming, the head of the iQiyi traffic anti-cheating project, and the co-founder of Come Technology & CTO Liang Shu from the content, Traffic and other aspects share the era of artificial intelligence, how the content review and traffic anti-cheat of the platform is completed.
At the moment, there are three types of content open platforms: user production content, professional production content, and professional production content. Public comments and maps are typical user-produced content; Douban, known as a typical professional production content; WeChat public number, vibrato is a professional production content. The main problems and contradictions brought about by the strong sociality, strong randomness and strong operation of the content open platform are the contradiction between advertisers and consumers, the contradiction between content quality and content, and the contradiction between traffic and quality.
Liu Lu Senior Algorithm Research and Development Engineer
Liu Lu believes that building an effective credit system is the key to the content open platform to resolve these contradictions. A conventional credit system building structure involves content producers and users. After the content producer submits the content, he will do the personal credit and content review. The audit needs to use the big data risk control and credit rating function. If the content is not approved by the content producer, the content can be released after the two aspects are approved. Conduct AB testing and grading release, and after the content is released, a risk assessment is needed to determine whether the content needs to be recalled and set a corresponding recall strategy.
Liu Lu pointed out that in terms of content, it is necessary to set up level checkpoints. Low-credit content needs to be monitored by content trust, and high-credit content needs to be set up to be reported and monitored, so as to eliminate the non-compliant content of the content open platform to the greatest extent.
At the moment, the video website has formed a gray industry chain. Some video platforms default to brushing behavior, and more than 20% of the advertisements are false. At the same time, the process of brushing has been highly programmed, and the industry lacks reasonable supervision. For advertisers, traffic fraud is no different than fraud. Nowadays, major video websites are developing very rapidly. How to do traffic anti-cheating in the case of a large number of users? Zhang Xiaoming, the person in charge of the iQiyi traffic anti-cheating project, shared the "skills" and "dao" of iQiyi traffic anti-cheating.
Zhang Xiaoming, Research Fellow, Aiqiyi Traffic Anti-Cheat Project
In the industry, there is no uniform definition of traffic anti-cheating. In Zhang Xiaoming's view, traffic anti-cheat is to create non-user-generated or purposeful data, and there are two forms of machine cheating and human work. Among them, machine cheat includes imitating delivery log and calling interface. This form setting mode is relatively fixed, and the short-term effect is obvious. The disadvantages of people's work include WeChat group, QQ group and professional APP guided by special person to guide operation and share with users. The pattern is not fixed and the effect is related to the size of the organization.
Using these two modes for live broadcast popularity, website brushing ads, TV dramas, and merchandise brushing traffic will cause normal users to miscalculate the product's criteria, damage corporate reputation, inaccurate data analysis, and rising costs.
In Zhang Xiaoming's view, traffic anti-cheating has passive defensive, post-analysis, complex business scenarios, no general model, continuous old rules and new rules, how to conduct effective traffic anti-cheating? Based on past business experience, Zhang Xiaoming said that effective traffic anti-cheating needs to be prepared in the early stage: master the meaning of metadata in the delivery log or database, master the working mode and scene of the main business, avoid information islands, understand What is the purpose of cheating and good communication with the business department to confirm the cheats and to do confidential work. After the preparation is completed, IP reputation, security portrait, encrypted information detection, and device hardware information detection are performed on the technical side; rule statistics and machine learning are performed on the business side.
With the development of the mobile Internet, people's clothing, food, housing, and travel are all combined, which brings a lot of profit to the black industry. In the financial field, there are loan fraud, stolen card transactions, money laundering and other modes; in the e-commerce industry, there are flashing lists, fraudulent activities, etc.; in the game, there are black cards, hang-ups, etc.; in social aspects, there are obscene porn posts, fraud Advertising and other phenomena.
Liang Wei, Co-founder & CTO of Digital Technology
Liang Shuming said that in the face of these fraudulent behaviors, companies have taken corresponding measures, but most companies lack global risk data support and professional anti-fraud talent teams. With the inherent complexity of UGC content, the level of black production technology With the continuous improvement of professional level, enterprises are facing more and more challenges.
On the same day, based on past experience, Liang Shu shared some experiences in content anti-fraud. He said that UGC scenes include avatars, nicknames, postings, barrage, comments, personal descriptions, private messages, group chats, etc., UGC brings problems: scams, advertisements, pornography, horror and other texts, pictures, voice UGC content. The same content has different meanings in different scenarios. A single keyword blacklist or a single model can easily cause a lot of manslaughter, affecting the user experience, and use deep learning (RNN) and SVC good NLP technology to establish different aspects for semantic understanding. The model detects semantic environments from multiple latitudes. Char & Word Embedding automatically recognizes tens of thousands of sensitive word variants, WeChat/QQ/mobile phone number and other contact variants, and uses CRF-based sequence annotation to automatically identify and eliminate interference.
Nowadays, the scammers have a clear division of labor and have formed a huge black industrial chain. With the advent of the era of artificial intelligence, technologies such as big data and machine learning are bound to hit the black production.
The composite terminal is a typical representative of the connector (CONNECTOR), which is composed of a plug (female) and a socket (male) to realize the connection function. For the transfer of voltage, current and signal, this series is characterized by safe, reliable and stable contact, and is mainly used in electric power, communication, industrial control and other industries.
composite terminal block,component terminal block,composite terminal block adapter
Sichuan Xinlian electronic science and technology Company , https://www.sztmlchs.com