Recommendation platform technical method for thousand-person thousand-face personalized recommendation

Abstract

The invention provides a recommendation platform technical method for thousand-face personalized recommendation of thousands of people. The recommendation platform technical method comprises the following operation steps: step 1, defining an experiment; 2, dividing the definition experiment in the step 1 into an algorithm and AbTest; and step 3, the AbTest in the step 2 controls the recommendation platform to load at regular time, and controls each layer of configuration of the recommendation system to realize thousand-person and thousand-face personalized recommendation. The method can quickly support business demand iteration, reduce development cost, improve human efficiency, change a black box situation of a recommendation system, realize visual management of recommendation strategies, improve flexibility and stability of the recommendation system, support pluggable free configuration experiment strategies, provide an experiment training ground for algorithm and product students, and improve the development efficiency of the recommendation system. The method assists the experiment strategy to rapidly fall into the ground, enables the whole life cycle of the recommendation system, scientifically reflects the effect of the recommendation strategy through a powerful data analysis tool, and provides data support for the continuous optimization of the recommendation strategy.

Description

Technical Field

[0001] The present invention relates to the field of recommendation platforms, and in particular to a technical method for a recommendation platform that provides personalized recommendations for different people. Background Art

[0002] With the rapid development of the Internet, people's daily lives are increasingly closely connected with the Internet, such as listening to music, watching movies, shopping, reading, chatting, etc. At the same time, a huge amount of user and item data is constantly generated on the Internet every day, which makes it difficult for Internet users to quickly find the unknown information they need or are interested in. Therefore, personalized recommendation technology came into being and is constantly being innovated. Personalized recommendation technology aims to model the user's interest preferences based on the user's own characteristics, and then recommend items that meet the user's personalized preferences and have not been used yet.

[0003] In the existing recommendation platform technology methods, the experimental strategy logic code is highly coupled with the recommendation system architecture code, without modular division and management; a slight adjustment to the experimental strategy requires code modification and online service, resulting in poor system stability; the recommendation system lacks auxiliary functions, resulting in low system security; after the experimental strategy is online, it is impossible to conduct detailed analysis and optimization of the experimental strategy effect;

[0004] That is to say, in the existing technical recommendation system, various types of business codes are highly coupled, the system scalability and flexibility are poor, and the system risk is high; similar requirements are repetitively developed, the requirement iteration is slow, the development cost is high, and the efficiency is low; the adjustment of experimental strategy leads to frequent launch of the recommendation system, and the system stability and reliability are low; the recommendation experiment strategy is a complete black box, which is not easy for novices to understand and master, and it is extremely difficult to locate problems; the recommendation system has no indicators and performance monitoring, the indicators and performance fluctuate greatly, and there is no early warning; there are no powerful tools to accurately analyze the effects of experimental strategies. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the defects in the prior art, such as high system risk, high development cost and low efficiency caused by repeated development requirements, no indicators and performance monitoring, and extremely difficult to locate problems, and to provide a technical method for a recommendation platform that can provide personalized recommendations for thousands of people.

[0006] The present invention solves the above technical problems through the following technical solutions:

[0007] The present invention provides a recommendation platform technology method for providing personalized recommendations for different people, and the recommendation platform technology method includes the following steps:

[0008] Step 1: Define the experiment.

[0009] Step 2: Divide the defined experiment in step 1 into algorithm and AbTest;

[0010] In step 3 and step 2, AbTest controls the scheduled loading of the recommendation platform and the configuration of each layer of the recommendation system to achieve personalized recommendations for each individual.

[0011] In this technical solution, it is possible to quickly support the iteration of business needs, reduce development costs, improve labor efficiency, change the black box situation of the recommendation system, realize the visual management of recommendation strategies, improve the flexibility and stability of the recommendation system, support pluggable and freely configurable experimental strategies, provide an experimental training ground for algorithm and product students, help experimental strategies to be quickly implemented, empower the entire life cycle of the recommendation system, and scientifically reflect the effects of recommendation strategies through powerful data analysis tools, providing data support for the continuous optimization of recommendation strategies.

[0012] Preferably, the recommendation platform in step three includes a pre-online unit, an on-line unit and a post-online unit.

[0013] In this technical solution, the recommendation platform empowers the entire life cycle of the recommendation business before, on, and after it goes online, so as to achieve continuous optimization of the closed-loop process of the recommendation business and improve the flexibility, security, and stability of the recommendation system.

[0014] Preferably, the pre-launch unit includes a performance stress testing module and an experimental Mock module.

[0015] In this technical solution, the pre-launch unit can be used to enable the recommendation platform to have performance stress testing and experimental mocks.

[0016] Preferably, the online unit includes a plug-in management module, a plug-in orchestration module, a feature management module, a model management module, a recommendation pool management module, an experiment management module and an experiment report module.

[0017] In this technical solution, the online unit can be used to enable the recommendation platform to have plug-in management function, plug-in orchestration function, feature management function, model management function, recommendation pool management function, experiment management function, and experiment report function.

[0018] Preferably, the post-launch unit includes an indicator monitoring module, a performance monitoring module, a link tracking module and a recommendation analysis module.

[0019] In this technical solution, the post-launch unit can be used to enable the recommendation platform to have indicator monitoring function, performance monitoring function, link tracking function, and recommendation analysis function.

[0020] Preferably, the recommendation system in step three includes an API layer unit, a recall layer unit, a coarse ranking layer unit, a fine ranking layer unit and a re-ranking layer unit.

[0021] In this technical solution, the entire life cycle of the recommendation system is enabled, and through powerful data analysis tools, the effectiveness of the recommendation strategy is scientifically reflected, providing data support for the continuous optimization of the recommendation strategy.

[0022] Preferably, the API layer unit includes a business line one module, a business line two module and a business line N module.

[0023] Preferably, the recall layer unit includes a recall module, a filtering module and a fusion module.

[0024] Preferably, the coarse-grained layer unit comprises a coarse-grained model module, a feature module 1, a score weighting module 1 and a material cutoff module 1.

[0025] Preferably, the refined ranking layer unit includes a refined ranking model module, a second feature module, a second score weighting module and a second material truncation module.

[0026] On the basis of being in accordance with the common sense in the art, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present invention.

[0027] The positive and progressive effects of the present invention are:

[0028] The present invention can quickly support the iteration of business needs, reduce development costs, improve labor efficiency, change the black box situation of the recommendation system, realize the visual management of recommendation strategies, improve the flexibility and stability of the recommendation system, support pluggable and freely configurable experimental strategies, provide experimental training grounds for algorithm and product students, help experimental strategies to be quickly implemented, and enable the full life cycle of the recommendation system. Through powerful data analysis tools, it can scientifically reflect the effects of recommendation strategies and provide data support for the continuous optimization of recommendation strategies.

[0029] At the same time, the present invention enables the recommendation service in its entire life cycle before, during and after it goes online, thus realizing the closed-loop process of continuous optimization of the recommendation service and improving the flexibility, security and stability of the recommendation system.

[0030] Further, the present invention closely integrates the recommendation platform with the Abtest diversion service to achieve visualization, automation, and one-stop management of recommendation strategies;

[0031] Furthermore, the present invention supports the free assembly of various business plug-ins, realizes the reuse of business codes, reduces development costs, and improves labor efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is an overall schematic diagram of the technical method of the recommendation platform for providing personalized recommendations for each individual according to an embodiment of the present invention.

[0033] Figure 2A method flow chart of the technical method of a recommendation platform for providing personalized recommendations to different people in an embodiment of the present invention.

[0034] Figure 3 for Figure 1 The schematic diagram of the recommendation platform showing the technical method of the recommendation platform for providing personalized recommendations for each individual.

[0035] Figure 4 for Figure 1 The recommendation system schematic diagram of the recommendation platform technical method for providing personalized recommendations for thousands of people is shown.

[0036] Description of Reference Numerals

[0037] 1. Pre-launch unit; 11. Performance stress testing module; 12. Experimental Mock module;

[0038] 2. Online unit; 21. Plugin management module; 22. Plugin orchestration module; 23. Feature management module; 24. Model management module; 25. Recommendation pool management module; 26. Experiment management module; 27. Experiment report module;

[0039] 3. Post-launch unit; 31. Indicator monitoring module; 32. Performance monitoring module; 33. Link tracking module; 34. Recommendation analysis module;

[0040] 4. API layer unit; 41. Business line one module; 42. Business line two module; 43. Business line N module;

[0041] 5. Recall layer unit; 51. Recall module; 52. Filter module; 53. Fusion module;

[0042] 6. Coarse-grained layer unit; 61. Coarse-grained model module; 62. Feature module 1; 63. Score weighting module 1; 64. Material cutoff module 1;

[0043] 7. Refined sorting layer unit; 71. Refined sorting model module; 72. Feature module 2; 73. Score weighting module 2; 74. Material cutoff module 2;

[0044] 8. Rearrangement layer unit; 81. Weighting module; 82. De-weighting module; 83. Forced insertion module; 84. Scattering module; 85. Operation position module. DETAILED DESCRIPTION

[0045] The present invention is further described below by way of examples, but the present invention is not limited to the scope of the examples.

[0046] Figures 1 to 4The structure diagram of an embodiment of the recommendation platform technology method of the present invention is shown. The recommendation platform technology method of the recommendation platform technology method of the present invention includes the following steps:

[0047] Step 1: Define the experiment.

[0048] Step 2: Divide the defined experiment in step 1 into algorithm and AbTest;

[0049] In step 3 and step 2, AbTest controls the scheduled loading of the recommendation platform and the configuration of each layer of the recommendation system to achieve personalized recommendations for each individual.

[0050] In this technical solution, it is possible to quickly support the iteration of business needs, reduce development costs, improve labor efficiency, change the black box situation of the recommendation system, realize the visual management of recommendation strategies, improve the flexibility and stability of the recommendation system, support pluggable and freely configurable experimental strategies, provide an experimental training ground for algorithm and product students, help experimental strategies to be quickly implemented, empower the entire life cycle of the recommendation system, and scientifically reflect the effects of recommendation strategies through powerful data analysis tools, providing data support for the continuous optimization of recommendation strategies.

[0051] AbTest refers to AB testing, which is to develop two plans for the same goal. In the same time dimension, user groups with the same or similar composition randomly use one plan, collect user experience data and business data from each group, and finally evaluate the best version based on significance test analysis and formally adopt it.

[0052] The recommendation platform in step three includes a pre-online unit 1, an on-line unit 2 and a post-online unit 3.

[0053] In this technical solution, the recommendation platform empowers the entire life cycle of the recommendation business before, on, and after it goes online, so as to achieve continuous optimization of the closed-loop process of the recommendation business and improve the flexibility, security, and stability of the recommendation system.

[0054] The pre-launch unit 1 includes a performance stress testing module 11 and an experimental Mock module 12 .

[0055] In this technical solution, the pre-launch unit 1 can be used to enable the recommendation platform to have performance stress testing and experimental Mock.

[0056] In experiments, Mock usually refers to a simulated behavior. It is mainly used in the testing process. For some objects that are not easy to construct or obtain, a virtual object is used to create them for testing.

[0057] Mock enables testing to be performed independently of external systems or components by simulating the behavior of objects. Mock objects can replace real objects to simulate the behavior of objects. This testing method can help developers and testers perform effective testing when real objects are unavailable or difficult to build.

[0058] The online unit 2 includes a plug-in management module 21, a plug-in arrangement module 22, a feature management module 23, a model management module 24, a recommendation pool management module 25, an experiment management module 26 and an experiment report module 27.

[0059] In this technical solution, by using the online unit 2, the recommendation platform can have plug-in management function, plug-in arrangement function, feature management function, model management function, recommendation pool management function, experiment management function, and experiment report function.

[0060] The post-launch unit 3 includes an indicator monitoring module 31 , a performance monitoring module 32 , a link tracking module 33 and a recommendation analysis module 34 .

[0061] In this technical solution, the post-online unit 3 can be used to enable the recommendation platform to have indicator monitoring function, performance monitoring function, link tracking function, and recommendation analysis function.

[0062] The recommendation platform is presented to users in the form of a web page, supporting page menu and button-level permission control to ensure the security of the recommendation system;

[0063] The recommendation platform backend records various operations on web pages in a log format to make the operation traceable.

[0064] A web page is a file organized in HTML format on the World Wide Web. When searching for information through the World Wide Web, it appears in the form of an information page, which may include graphics, text, sound, video and other information.

[0065] The recommendation platform data is persistently stored in MySQL to ensure data reliability and integrity. MySQL data is loaded into AbTest memory in seconds, closely integrating the recommendation platform with AbTest diversion to achieve rapid distribution of experimental strategies.

[0066] MySQL is an open source relational database management system RDBMS. The MySQL database system uses the most commonly used database management language SQL for database management;

[0067] Because MySQL is open source, anyone can download it under the General Public License and modify it according to their own needs.

[0068] The recommendation system in step three includes an API layer unit 4 , a recall layer unit 5 , a coarse ranking layer unit 6 , a fine ranking layer unit 7 and a re-ranking layer unit 8 .

[0069] In this technical solution, the entire life cycle of the recommendation system is enabled, and through powerful data analysis tools, the effectiveness of the recommendation strategy is scientifically reflected, providing data support for the continuous optimization of the recommendation strategy.

[0070] The API layer unit 4 includes a business line one module 41 , a business line two module 42 and a business line N module 43 .

[0071] The recall layer unit 5 includes a recall module 51 , a filtering module 52 and a fusion module 53 .

[0072] The coarse-clustering layer unit 6 includes a coarse-clustering model module 61 , a feature module 62 , a score weighting module 63 and a material cutoff module 64 .

[0073] The refined ranking layer unit 7 comprises a refined ranking model module 71 , a feature module 2 72 , a score weighting module 2 73 and a material cutoff module 2 74 .

[0074] The rearrangement layer unit 8 includes a weighting module 81 , a weight reduction module 82 , a forced insertion module 83 , a scattering module 84 and an operation position module 85 .

[0075] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that this is only for illustration and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A technical method for a recommendation platform with personalized recommendations for each individual, characterized by: The recommendation platform technical method comprises the following steps: Step 1: Define the experiment. Step 2: Divide the defined experiment in step 1 into algorithm and AbTest; In step 3 and step 2, AbTest controls the scheduled loading of the recommendation platform and the configuration of each layer of the recommendation system to achieve personalized recommendations for each individual.

2. The recommendation platform technology method for providing personalized recommendations for different people as claimed in claim 1, characterized in that: The recommendation platform in step three includes a pre-online unit (1), an on-line unit (2) and a post-online unit (3).

3. The recommendation platform technology method for providing personalized recommendations for different people as claimed in claim 2, characterized in that: The pre-launch unit (1) includes a performance stress testing module (11) and an experimental Mock module (12).

4. The recommendation platform technology method for providing personalized recommendations for different people as claimed in claim 2, characterized in that: The online unit (2) comprises a plug-in management module (21), a plug-in arrangement module (22), a feature management module (23), a model management module (24), a recommendation pool management module (25), an experiment management module (26) and an experiment report module (27).

5. The recommendation platform technology method for providing personalized recommendations for different people as claimed in claim 2, characterized in that: The post-online unit (3) comprises an indicator monitoring module (31), a performance monitoring module (32), a link tracking module (33) and a recommendation analysis module (34).

6. The recommendation platform technology method for providing personalized recommendations for different people as claimed in claim 1, characterized in that: The recommendation system in step three includes an API layer unit (4), a recall layer unit (5), a coarse ranking layer unit (6), a fine ranking layer unit (7) and a re-ranking layer unit (8).

7. The recommendation platform technology method for providing personalized recommendations for different people as claimed in claim 6, characterized in that: The API layer unit (4) includes a business line one module (41), a business line two module (42) and a business line N module (43).

8. The recommendation platform technology method for providing personalized recommendations for different people as claimed in claim 6, characterized in that: The recall layer unit (5) comprises a recall module (51), a filtering module (52) and a fusion module (53).

9. The recommendation platform technology method for providing personalized recommendations for different people as claimed in claim 6, characterized in that: The coarse-particle layer unit (6) comprises a coarse-particle model module (61), a feature module one (62), a score weighting module one (63) and a material cutoff module one (64).

10. The recommendation platform technology method for providing personalized recommendations for different people as claimed in claim 6, characterized in that: The refined sorting layer unit (7) comprises a refined sorting model module (71), a feature module 2 (72), a score weighting module 2 (73) and a material cutoff module 2 (74).

Images