Clarify the thinking of Internet information service algorithm

In recent years, Internet information service algorithms have played an important role in accelerating information dissemination, prosperous digital economy, and promoting social development. At the same time, unreasonable applications have also affected normal dissemination order, market order, and social order. Information cocoon houses, big data killing, algorithm discrimination and other algorithm applications have emerged endlessly, which has aroused widespread society. On September 17, 2021, the National Internet Information Office and other nine ministries and commissions jointly issued the "Guiding Opinions on Strengthening the Comprehensive Governance of the Internet Information Service Algorithm" (hereinafter referred to as "Opinions"). Systemic and comprehensive norms requirements. 2022年3月1日起，《互联网信息服务算法推荐管理规定》（以下简称《规定》）正式施行，支撑《规定》中算法备案工作的互联网信息服务算法备案系统也同期正式上线运行，标志着my country's comprehensive management of Internet information service algorithms has entered a new stage of rule of law, normalization, and long -term.

The "Regulations" uses Internet information service algorithm recommendation activities as the starting point. In terms of regulatory objects, the "Regulations" specifically clarified the object of algorithm governance to provide information to the user with information to recommend information to the user. The algorithm is divided into five categories: generating synthetic classes, personalized push class, sorting selection, retrieval filtering, and scheduling decision -making categories. In terms of regulatory content, the "Regulations" clearly requires that algorithm recommendation service providers should implement the responsibility of the main body of the algorithm, establish and improve the safety management system and technical measures of the algorithm, adhere to the mainstream value orientation, effectively protect the user's rights and interests, and promote the application of algorithm. The "Regulations" made clear requirements for the behavior of the five major algorithms from two aspects: information service specifications and protection of user equity protection. In terms of regulatory system, the "Regulations" have established scientific supervision systems that cover measures such as algorithm grading classification, algorithm filing management, algorithm supervision and inspection, algorithm risk monitoring, algorithm safety assessment, etc., and strive to sample both the symptoms and gradually form a long -term mechanism for algorithm governance.

The "Regulations" divide the Internet information service recommendation algorithm into five categories, and set targeted differentiated terms for each algorithm. Based on the specific content of the "Regulations", the possible governance ideas of these five categories of algorithms are explained.

The first is to generate the governance of the synthetic algorithm. The synthetic algorithm is a tool -type algorithm. The key to its governance lies in "fixed borders, clear signs, and control of the source." The first is to ensure the reasonable scope of application of the algorithm, and the urging algorithm recommendation service provider shall not use the generating synthetic algorithm to generate composite bad information. Article 7 of the "Regulations" clearly states that "must not be generated for synthetic false news information." Secondly, it is required to make a significant identification of generating synthetic data, ensure the user's right to know, and prevent illegal and bad data that generate synthesis in the spread of network space. Article 9 of the "Regulations" states that "if an algorithm that is found to be obviously identified to generate synthetic information, it shall be made before the transmission can be continued." Finally, the source of the source of such tool -type algorithms. Synthetic algorithm service providers shall effectively perform the filing procedures in accordance with the requirements of the "Regulations". Those who fail to perform the filing procedures shall not provide the production of synthetic services to the outside world. In addition, it is necessary to strengthen the detection and traceability technical research on the production of synthetic content, and to ensure the closed loop of the supervision of the synthetic algorithm.

The second is the governance of a personalized push class algorithm. Personalized push class algorithms are distributed algorithms. The key to its governance is the full process management of the protection of the distribution mechanism to the protection of user equity. The first is to require algorithm recommendation service providers to strengthen the supervision and management of the source of manuscripts. The "Regulations" clearly states that "the information that is not allowed to use algorithms to recommend service communication laws and administrative regulations" and "news information released by units within the scope of non -state regulations." Algorithm recommendation service providers should establish a sound manuscript source management mechanism to prevent illegal and bad content from entering the source pool. The second is to strengthen the supervision and management of the content distribution mechanism. In response to the user model and user label, artificial intervention mechanism and model optimization goals involved in the distribution mechanism, the "Regulations" requires that "the keywords of illegal and bad information must be recorded into the user's interest point or as a user label and push the information" " Optimize the transparency and interpretability of the rules such as search, sorting, selection, push, display, etc., avoid adverse effects on users, prevent and reduce disputes "," must not set up inducing user obsessed, excessive consumption, etc. violating laws and regulations or violations of ethics and morality Algorithm model. Finally, the implementation of the protection of user rights and interests. The "Regulations" clearly requires the algorithm recommendation service provider to effectively protect the legal rights and interests of the user's right to know and the right to choose, including "the situation where the user provides the algorithm recommendation service in a significant way, and publicize the basic principles of the basic principle of the algorithm recommendation service in an appropriate manner. , Objective intentions and main operational mechanisms, "" Options to provide users who do not target their personal characteristics, or provide users with convenient closure of algorithm recommendation services "and so on. In addition, for special groups such as minors and elderly people, the "Regulations" also put forward targeted rights protection requirements, and the specific implementation of the supervision process should be verified. The "Regulations" is committed to the management of the entire process of pushing information services through the source of the manuscript, algorithm recommendation model, manual intervention, and user independent selection, and build an integrated, mass, and individual network information content ecology, realize the content of information content, realize the content of information content Height, width and temperature. The third is the governance of sorting selection algorithms. The selection algorithm of the sorting is a model model. The key to its governance is to strengthen guidance management. The first is to strengthen the governance of the content ecology of the list page. The "Regulations" requires that "algorithm recommendation service providers should strengthen the algorithm recommendation service version of the noodle page ecological management", "on the first screen, hot search, selection, list category, pop -up window and other key links, actively present information that meets the mainstream value -oriented information "". The second is to require algorithm recommendation service providers to open the ranking basis and the rules of the list to ensure the fairness, justice and transparency of the list. Finally, strengthen the review of the list intervention mechanism, and check whether there is a phenomenon such as "manipulation list" and "control hot search or selection" prohibited in the "Regulations".

The fourth is to retrieve the governance of filtering algorithms. The retrieval of the filter algorithm is a screening algorithm. The key to its governance is the management of the feature library and label library. The first is to strengthen the validity assessment of the feature library and label library. Article 9 of the "Regulations" clearly states that "algorithm recommendation service providers should strengthen information security management, establish and improve the characteristic library for identifying illegal and bad information, and improve the standards, rules and procedures of the storage." The second is to strengthen the supervision of retrieval and filter results. Some algorithms recommend service providers to retrieve the filtering algorithm to filter the relevant content of the interests or sort and adjust the content of the interests, so as to achieve the purpose of competitors to suppress or self -preferential treatment. Article 15 of the "Regulations" clearly states that "algorithm recommendation service providers shall not use algorithms to perform unreasonable restrictions on other Internet information service providers, or hinder and destroy the normal operation of the Internet information services provided by its legally provided Behavior".

Fifth, the governance of scheduling decision -making algorithms. The scheduling decision -making algorithm is an automated decision -making algorithm. The key to its governance is the protection of the legal rights and interests of the algorithm. The first is to strengthen the governance of the chaotic chaos. Algorithm recommendation service providers cannot be guided by economic interests, ignore the humanistic care of workers, and dispatch workers based on the decision -making results of the algorithm. Article 20 of the "Regulations" clearly states that "algorithm recommendation service providers provide work dispatching services to workers, they shall protect the workers' legitimate rights and interests such as labor remuneration, rest and vacation, and establish and improve platform order allocation, reward composition and payment, work, work Time, reward and punishment and other related algorithms ". The second is to strengthen the governance of discrimination against algorithms. Article 21 of the "Regulations" clearly states that algorithm recommendation service providers "shall not use illegal behaviors such as consumer preferences and trading habits, and use algorithm to implement unreasonable differences in transaction conditions such as transaction prices."

The "Opinions" clearly put forward the goal of the comprehensive governance of the Internet information service algorithm: using about three years or so, gradually establish a comprehensive governance pattern of algorithm security with a sound governance mechanism, a complete regulatory system, and an algorithm ecological specification. In terms of governance mechanisms, the security governance policies and regulations of the Internet information service algorithm are formulated. The main powers and responsibilities of algorithm safety governance are clear, and the governance structure is efficiently operated. In terms of regulatory system, it is innovatively constructed to form a multi -dimensional supervision system such as algorithm safety risk monitoring, algorithm safety assessment, scientific and technological ethics review, algorithm filing management, and illegal acts involving algorithms. In terms of algorithm ecology, the advocates algorithm is correct, full of positive energy, fairness and fairness, openness and transparency of algorithms, safe and controllable algorithm development, and independent innovation, and effectively prevent the hidden risks brought about by the abuse of algorithms. The comprehensive management of Internet information service algorithms is a long and arduous task that requires joint efforts from all walks of life. On the one hand, it is necessary to strengthen technological innovation and development, and overcome the inherent security risks of the algorithm itself from the technical level. On the other hand, the regulatory departments, enterprises, industry organizations, and the public are required to participate in multi -directional participation, strive to explore new measures and measures for comprehensive governance of Internet information service algorithms, jointly promote the application of ecological governance of algorithms, and create a clear network environment.

(The author is a researcher at the Institute of Computing Technology of the Chinese Academy of Sciences)

Source: China Social Science Network-Journal of Social Sciences of China

Author: Shen Huawei

- END -