From "E -commerce Law" to "Personal Information Protection Law"，Then go to the latest "Recommended Management Regulations on Internet Information Service Algorithms"，my country's legal norm system on algorithm governance is gradually established，In various algorithms，Personalized recommendation algorithm is the most closely connected with people's daily life，It is also one of the types of algorithms that cause more controversy。

With the emergence of many web platforms based on Web2.0 technology，The technical threshold for the production and dissemination of information of ordinary users is greatly reduced，The total number of social information is also increased geometric level。However, in the face of massive information，Relying on manual editing or fixed rules (such as click volume、timeline、Keywords) Based information distribution method，But can only achieve "three thousand weak water takes one scoop"，The quality of information screening is guaranteed，But most of the incremental information is not effective because it cannot effectively reach the user，The "28 Laws" in the information market will not change substantially due to the growth of the supply side。

Personalized recommendation algorithm is exactly born to solve the above problems。Personalized recommendation service，Network service providers can screen in massive information based on user preferences，and give priority to the information content that meets user needs。This method of information distribution has significantly improved the matching efficiency of both parties to the supply and demand，The total amount of information exposure in the unit time can be greatly increased，Create conditions to activate more market segments。The active of market segments bet365 Play online games will undoubtedly expand the overall capacity of the market，While creating a lot of employment opportunities for the society，Continuously enrich the supply of goods and services in the market。For example，Video bloggers in the short video platform、The self -media in the information platform, the media, etc., all need to rely on personalized recommendation algorithms，To achieve precise matching of supply and demand in massive information and vast people，and get economic income。other，From the perspective of information receiving party，Personalized recommendation algorithm also helps to gain information with real value，Reduce energy consumption in information retrieval。Good recommendation algorithm will also be based on the user's existing preferences，Expand its information boundary，to discover or stimulate new information requirements。

Under an ideal state，A well -operated personalized recommendation algorithm will be the information provider、Information receiver and the information platform brings a good pattern for the three parties。However, the reality is not always the case，The intrinsic risk of personalized recommendation algorithms cannot be ignored。In practice，Algorithm service provider who master the dominant power of the rules may be due to lack of technical capabilities，or the consideration of its own interests，Make the relevant subjects and even the social public interest negatively affected due to the imbalance of algorithms，The most typical of which is to obtain stronger user stickiness，"Information Cocoon Room" effect caused by users to recommend similar information。In fact，It is to avoid users who are at passive acceptance status when they cannot make a meaningful choice.，Legislators have given their algorithm interpretation right。Legislators hope to increase the transparency of the algorithm operation bet365 Play online games mechanism，Users have the ability to feedback the effect of the algorithm based on their own wishes，Therefore, the negative impact of personalized recommendation on it。

Understand the basic principles and operating mechanisms of the algorithm，Breaking the algorithm black box、Establish an important foundation for the trust of algorithm。The underlying logic of the personalized recommendation algorithm can be summarized as the following steps: Collection information -establishment of a label system according to information，Forms to users、Information、Channel and other dimensions —— Select the crowd according to the mapping relationship circle between labels，Determine push information content -continuously adjusting the excellent algorithm according to the feedback of the push effect。

First，Personalized recommendation with the crowd rather than the individual。The purpose of using personalized recommendations is to promote value conversion through the accurate matching of demand and supply，If you buy a product、Improve video broadcast rate and information clicks, etc.，and only when this conversion reaches a certain scale，can the information supplier gain meaningful benefit。So，When setting a personalized recommendation algorithm，Dynamic adjustment of information "accuracy" and "recall" according to actual needs，Neither can over -compress the crowd in order to pursue extreme accuracy，You cannot simply sacrifice the accuracy of information for the expansion of the crowd。

Second，The source of the label is multi -dimensional。The label system is the "soul" of a personalized recommendation algorithm，is the direction marker guidance of information distribution。The most familiar user portrait of people，Essentially based on user identity information、Social attribute、Behavioral habit、Category preference、Geographical bet365 best casino games location and other multi -dimensional information，The label system generated through induction and interpretation，Each label is a dimension of understanding and understanding users。In order to form a mapping relationship between information and users，Personalized label not only requires portraits from users，You need to perform portraits around the information itself，Finally formed multi -dimensional、A three -dimensional label system。Taking the personalized recommendation algorithm of the e -commerce platform as an example，The commonly used label type also includes: commodity dimension，If the leaf category of the product、Price range、Price range、Target user population characteristics, etc.; store dimension，If the shop positioning、The area where it is located、Store type, etc.; Channel dimension，Ru search、Promotional Activities、live broadcast、Recommended。other，The characteristics of the external environment may also become a consideration factors affecting personalized recommendation results，If climate change、holidays、Large sports events, etc.。

third，The results of personalized recommendation come from the mapping relationship between tags，The mapping rules are the key to the final decision to recommend the information content，It may come from algorithm service providers for society、Market、Social and other rules of insights，If the consumer is purchasing a category of products，Usually interested in category B products。Information publishers may also choose according to specific needs，If advertisers can use marketing tools，Choose the corresponding label according to the characteristics of goods or marketing activities to circle the target group，To achieve a more accurate release effect。Of course，The mapping rules in practice are often more complicated，bet365 live casino games Behind each recommendation result，It may involve a number of tabs with different weight labels。Algorithm engineers need comprehensive accuracy、Fineness、Diversity、Novelty、Fineness、Real -time indicators，Based on the actual effects, the weight of the label and its mapping rules are dynamically adjusted。

According to Article 24 of the Personal Information Protection Law，Personalized recommendation for information content also belongs to the field of automation decision -making。But with the credit report、Different algorithms used in credit，Personalized recommendation algorithm in most scenarios is to distribute information from users，It will not directly change the rights and obligations between users and other subjects，Then directly affect its personal property rights and interests。So，Algorithm grading classification management principles proposed based on the "Recommended Management Regulations on the Internet Information Service Algorithm，The setting of the explanation path of the personalized recommendation algorithm shall consider the operation mechanism of the algorithm、Use scene，and may be possible for users、Effects caused by related subjects such as algorithm service providers。After all, algorithm explanation itself is just a means，Help users better geographical algorithm operation mechanism，Therefore, it is capable of effectively feedback the algorithm effect through its own behavior，Finally achieve a win -win situation and establish a trust relationship，is the right to set the algorithm for the legislator、The ultimate goal of refusing right。Specifically，It can focus on the following three dimensions。

First，Data dimension。If the algorithm is compared to the precisely running machine，So the data is the energy of providing power for the machine。So，Discovery of data types based on personalized recommendation algorithms，Bet365 app download It will help users understand the operating logic of algorithm from the bottom layer，Especially for the introduction of personal information usage，The relationship (or related) relationship between the personal information used by the user authorized by the user can also be，Resolving users' concerns about personal information security，and constraint on the abuse of personal information on algorithm providers。

Second，Decision Missile。Mainly disclose the label system to the user through a reasonable range，and explain the mapping relationship between the personalized recommendation results and the label，Let the user get reasonable expectations for the source of the recommendation results。But what needs to be explained is，Because the label system is the core of the operation of a personalized recommendation algorithm，Business secrets involved in algorithm service providers，If it is completely insightful by other market entities or the public，Not only will it cause direct damage to its commercial interests，It may also cause the algorithm to be used maliciously，Destroy normal business order。So，Information disclosure of decision -making dimension，It should allow algorithm service providers to perform desensitization treatment，and focus on explaining the basic operation logic of the algorithm。

third，Effect dimension。In order to allow users to work before、Normally and after and afterwards, multiple dimensions more fully understand the operating mechanism of personalized recommendation algorithms，Algorithm service providers can in a simple and easy -to -understand way，Explain the feedback effect of users。If the user is "not interested" after the user's identification of specific information content，The informed algorithm will "No longer recommend similar products"，Under the premise of not leaking business secrets，It can also further refine the bet365 Play online games granularity of the effect，If the user is "not interested" after the user's identification of specific information content。

(Author is a senior expert in Alibaba Group Policy and Regulations Research Office)