Information recommendation system, implementation method thereof, electronic equipment and storage medium

Abstract

The invention relates to an information recommendation system, an implementation method thereof, electronic equipment and a storage medium. The system comprises an engineering subsystem, an algorithm subsystem and a collaborative docking subsystem, wherein the engineering subsystem comprises a plurality of function modules for realizing information recommendation; the algorithm subsystem comprises a plurality of algorithm components for completing an information recommendation logic algorithm; the collaborative docking subsystem comprises a plurality of layers of docking units, and the plurality of layers of docking units are respectively packaged in corresponding function modules of the engineering subsystem; in the process that the function modules in the engineering subsystem achieve information recommendation, the docking unit calls corresponding algorithm components in the algorithm subsystem based on driving instructions of the corresponding function modules, and the algorithm components complete corresponding recommendation logic algorithms. According to the method, engineering and algorithm decoupling is realized, and the development efficiency of the system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of software development, and in particular to an information recommendation system and an implementation method thereof, an electronic device, and a storage medium. Background Art

[0002] With the rapid development of the internet, big data, and artificial intelligence technologies, recommendation systems have become widely used in various fields, including e-commerce, social networking, and content distribution. During the development of recommendation systems, the synergy between engineering implementation and algorithmic logic is a key factor influencing the iteration efficiency and stability of recommendation systems.

[0003] Currently, existing recommendation systems mainly include two types of development solutions:

[0004] The first approach is a development approach led by the engineering team. In this approach, the algorithm team typically designs the algorithm logic and experimental plan, acting as the demand side, and provides the engineering team with a requirements document. The engineering team reviews the requirements document and develops the engineering document. After reaching consensus, the engineering team takes on the development of the system's core architecture, responsible for overall system architecture, module design, interface implementation, specific algorithm implementation, system deployment, and online testing. The algorithm team conducts experimental analysis, adjusts the algorithm, and develops a new requirements document. The engineering team adjusts the corresponding engineering functional modules based on the requirements document. Under this development model, due to differences in professional background and focus between the engineering and algorithm teams, frequent cross-team requirements clarification and technical verification are required. This communication and requirements transfer process is prone to misunderstandings and information gaps, increasing coordination costs and extending the development cycle. Furthermore, because the engineering team simultaneously designs the system architecture and implements the specific algorithms, the resulting recommendation system's engineering modules are highly coupled to the algorithm. This requires continued communication and coordination between the two teams for system upgrades and algorithm adjustments, reducing the overall iteration efficiency of the recommendation system.

[0005] The second option: a separate development approach. In this development approach, the engineering team builds a general execution framework with a general interface, an execution scheduler, and various functional modules, which is responsible for the scheduling and data flow of each functional module in the system. The specific algorithm logic, data processing, and business processes are all independently developed and completed by the algorithm team within the framework. Although this development approach gives the algorithm team a high degree of freedom to a certain extent, enabling it to flexibly develop and adjust algorithms, due to the uneven engineering development capabilities of the algorithm team members and the lack of sufficient software engineering practical experience of some developers, the developed code has defects in robustness, exception handling, and module interfaces. As a result, the average failure rate of the online system is high, increasing the risk of performance degradation. In addition, different developers have different levels of mastery of code styles and engineering specifications, which will make the subsequent maintenance and expansion of the system more complicated and difficult, ultimately affecting the overall iteration efficiency of the system. Summary of the Invention

[0006] In response to the technical problems existing in the prior art, the present invention proposes an information recommendation system and its implementation method, electronic equipment and storage medium, the recommendation system and its implementation method reduce the coupling degree between engineering and algorithms and improve development efficiency.

[0007] In order to solve the above technical problems, according to one aspect of the present invention, the present invention provides an information recommendation system, comprising:

[0008] An engineering subsystem, comprising a plurality of functional modules for implementing information recommendation;

[0009] an algorithm subsystem, the algorithm subsystem including a plurality of algorithm components for implementing an information recommendation logic algorithm; and

[0010] A collaborative docking subsystem includes multi-layer docking units, which are respectively encapsulated in corresponding functional modules of the engineering subsystem; in the process of implementing information recommendation by the functional modules in the engineering subsystem, the docking unit calls the corresponding algorithm component in the algorithm subsystem based on the driving instructions of the corresponding functional module, and the algorithm component completes the corresponding recommendation logic algorithm.

[0011] Optionally, the algorithm subsystem includes a hot-swap configuration interface for receiving hot-swap configuration parameters of the algorithm component and implementing online or offline operation of the corresponding algorithm component according to the received hot-swap configuration parameters.

[0012] Optionally, each algorithm component in the algorithm subsystem includes an algorithm parameter configuration interface to provide a configuration channel for algorithm parameters.

[0013] Optionally, the information recommendation system also includes a configuration center for providing the collaborative docking subsystem with algorithm component identification parameters and algorithm parameters required for each algorithm component; correspondingly, the docking unit calls the corresponding algorithm component based on the algorithm component identification parameters and passes the corresponding algorithm parameters to the corresponding algorithm component.

[0014] Optionally, the collaborative docking subsystem includes a recall layer and a ranking layer, and the multi-layer docking unit includes a recall docking unit and a ranking docking unit; the functional module of the engineering subsystem includes a recall module and a ranking module, and the algorithm component in the algorithm subsystem includes a recall algorithm component and a ranking algorithm component;

[0015] Among them, the recall docking unit is encapsulated in the recall module, and the sorting docking unit is encapsulated in the sorting module. When the recommendation conditions are met, the recall module drives the recall docking unit to call the recall algorithm component to generate the recall conditions; the recall module recalls a preset amount of information based on the recall conditions to form a first information candidate set; the sorting module drives the sorting docking unit to call the sorting algorithm component to sort the information in the first information candidate set.

[0016] Optionally, the recall docking unit includes:

[0017] a recall parameter subunit configured to determine a corresponding recall category and a recall algorithm component inheriting the recall category based on the driving instruction of the recall module; and

[0018] The calling subunit is configured to call the recall algorithm component that inherits the recall category and pass the corresponding algorithm parameters to the recall algorithm component.

[0019] Optionally, the recall categories include multiple ones, and each recall algorithm component inherits one recall category.

[0020] Optionally, the functional module of the engineering subsystem also includes a scenario module, which is configured to determine a recommended scenario based on front-end data and send the recommended scenario data to the recall module; the recall module includes the recommended scenario data in the driving instruction, and the recall parameter subunit determines the corresponding recall category and the recall algorithm component that inherits the recall category based on the recommended scenario data in the driving instruction; the recall algorithm component that inherits the recall category generates a recall condition that meets the recommended scenario.

[0021] Optionally, there are one or more recall categories corresponding to a recommendation scenario, and correspondingly, there are one or more recall conditions that meet the recommendation scenario.

[0022] Optionally, the sorting module of the engineering subsystem includes a coarse sorting unit and a fine sorting unit, and correspondingly, the sorting layer of the collaborative docking subsystem includes a coarse sorting sublayer and a fine sorting sublayer, and the sorting docking unit includes a coarse sorting docking unit encapsulated in the coarse sorting unit and a fine sorting docking unit encapsulated in the fine sorting unit; the sorting algorithm component in the algorithm subsystem includes a coarse sorting algorithm component and a fine sorting algorithm component, wherein the coarse sorting docking unit calls the coarse sorting algorithm component based on the sorting instruction of the coarse sorting unit to filter part of the information from the first information candidate set to obtain the second information candidate set; the fine sorting docking unit calls the fine sorting algorithm component based on the sorting instruction of the fine sorting unit to perform refined scoring on the information in the second information candidate set, and filters out part of the information from the second information candidate set based on the scoring.

[0023] Optionally, the functional module of the engineering subsystem also includes a rearrangement module, and correspondingly, the algorithm component of the algorithm subsystem includes a rearrangement algorithm component, the collaborative docking subsystem also includes a rearrangement layer, and the multi-layer docking unit also includes a rearrangement docking unit encapsulated in the rearrangement module; when the fine sorting docking unit calls the fine sorting algorithm component based on the sorting instruction of the fine sorting unit to perform refined scoring on the information in the second information candidate set, and filters out part of the information from the second information candidate set based on the score to obtain a third information candidate set, the rearrangement docking unit calls the rearrangement algorithm component based on the driving instruction of the rearrangement module to rearrange the information in the third information candidate set to obtain recommended information for output.

[0024] Optionally, the re-ranking algorithm component includes one or more filtering components, one or more weighting components, and one or more fatigue control components.

[0025] Optionally, the refined ranking algorithm component further includes a multi-objective fusion optimization component configured to achieve multi-objective fusion optimization when performing fine scoring on the information in the second information candidate set.

[0026] According to another aspect of the present invention, the present invention further provides a method for implementing an information recommendation system, comprising the following steps:

[0027] Creating an engineering subsystem, wherein the engineering subsystem includes a plurality of functional modules for implementing information recommendation;

[0028] Creating an algorithm subsystem, the algorithm subsystem including a plurality of algorithm components for implementing an information recommendation logic algorithm; and

[0029] Create a collaborative docking subsystem, wherein the collaborative docking subsystem includes multi-layer docking units, and the multi-layer docking units are respectively encapsulated in the corresponding functional modules of the engineering subsystem. In the process of implementing information recommendation by the functional modules in the engineering subsystem, the docking unit calls the corresponding algorithm component in the algorithm subsystem based on the driving instructions of the corresponding functional module, and the algorithm component completes the corresponding recommendation logic algorithm.

[0030] According to another aspect of the present invention, the present invention also provides an electronic device, which includes a processor and a memory, wherein a computer program instruction set is stored on the memory, and when the processor executes the computer program instruction set on the memory, the aforementioned information recommendation system is implemented or the implementation method of the aforementioned information recommendation system is executed.

[0031] According to another aspect of the present invention, the present invention also provides a computer-readable storage medium, wherein a computer program instruction set is stored on the computer-readable storage medium, and when the computer program instruction set is executed by a processor, it implements the aforementioned information recommendation system or executes the implementation method of the aforementioned information recommendation system.

[0032] According to another aspect of the present invention, the present invention further provides a computer program product, which includes a computer program instruction set, and when the computer program instruction set is executed by a processor, it implements the aforementioned information recommendation system or executes the implementation method of the aforementioned information recommendation system.

[0033] The present invention achieves the decoupling of engineering and algorithms, allowing the engineering team and the algorithm team to focus on their respective development tasks, improving development efficiency, and achieving a stable recommendation system that can be efficiently iterated and continuously optimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Below, the preferred embodiments of the present invention will be further described in detail with reference to the accompanying drawings, in which:

[0035] Figure 1 Principle block diagram of an information recommendation system according to an embodiment of the present invention;

[0036] Figure 2 This is a block diagram of the system structure principle for the recall process according to one embodiment of the present invention;

[0037] Figure 3 is a principle block diagram of an information recommendation system according to another embodiment of the present invention;

[0038] Figure 4 is a principle block diagram of an information recommendation system according to another embodiment of the present invention;

[0039] Figure 5is a principle block diagram of an information recommendation system according to another embodiment of the present invention;

[0040] Figure 6 This is a principle block diagram of a job recommendation system in a recruitment platform according to one embodiment of the present invention;

[0041] Figure 7 is a principle block diagram of an algorithm subsystem according to one embodiment of the present invention;

[0042] Figure 8 is a flow chart of a method for implementing an information recommendation system according to an embodiment of the present invention; and

[0043] Figure 9 FIG. 1 is a schematic diagram of the hardware structure principle of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0045] In the detailed description that follows, reference may be made to the various drawings that form part of this application and illustrate specific embodiments of the present application. In the drawings, similar reference numerals describe substantially similar components in different figures. Each specific embodiment of the present application is described below in sufficient detail to enable a person of ordinary skill in the art to implement the technical solutions of the present application. It should be understood that other embodiments may be utilized or that structural, logical, or electrical changes may be made to the embodiments of the present application.

[0046] Information recommendation methods typically include two major steps. The first step is to retrieve information, with the goal of quickly filtering out a "potentially relevant" subset (high recall rate) from a massive candidate set. Applied recall algorithms include collaborative filtering (ItemCF / UserCF), content matching (TF-IDF / Embedding), graph algorithms (Node2Vec), and rule-based strategies (hot / regional recall). The second step is to rank the recalled information, with the goal of fine-tuning the ranking of the recalled results to select a small number of "most relevant" high-precision results. Applied ranking algorithms include traditional algorithms (such as LR / GBDT), deep learning (such as DeepFM / DIN), and multi-task learning (such as CTR+CVR). When developing a system to implement the aforementioned information recommendation method, both the engineering and algorithm components work together to complete the aforementioned process.

[0047] The present invention provides an information recommendation system and its implementation method. The information recommendation system includes an engineering subsystem, an algorithm subsystem, and a collaborative docking subsystem between the two. The engineering subsystem includes multiple functional modules for implementing information recommendation and is developed by an engineering team. The algorithm subsystem includes multiple algorithm components that implement the information recommendation logic algorithm and is completed by the algorithm team. The collaborative docking subsystem includes multi-layer docking units, which are respectively encapsulated in the corresponding functional modules of the engineering subsystem. During the information recommendation process, the algorithm components are called as needed. Therefore, the engineering development team only needs to focus on the implementation of functions and the acquisition of data. The algorithm team is responsible for the recommended algorithm logic and no longer needs to pay attention to the functional implementation on the engineering side, thereby achieving decoupling of engineering and algorithms.

[0048] See also Figure 1 , Figure 1According to a block diagram of the information recommendation system according to one embodiment of the present invention, the information recommendation system includes an engineering subsystem 1, an algorithm subsystem 2, and a collaborative docking subsystem 3. In this embodiment, the engineering subsystem 1 includes an execution scheduler 10, a recall module 11, a ranking module 12, and other functional modules, such as other business modules and basic functional modules. The algorithm subsystem 2 includes a configuration center 20, one or more recall algorithm components 21, and one or more ranking algorithm components 22. The collaborative docking subsystem 3 includes a recall layer and a ranking layer. The recall docking unit 31 in the recall layer is encapsulated in the recall module 11, and the ranking docking unit 32 is encapsulated in the ranking module 12. The execution scheduler 10 schedules each functional module in an orderly manner according to the module execution scheduling graph. Multiple functional modules execute in an orderly and concurrent manner to achieve their corresponding functions. For example, when the information recommendation conditions are met, the execution scheduler 10 issues a command to the recall module 11. The recall module 11 executes the recall process to recall a certain amount of information. It then sends a command to the ranking module 12. The ranking module 12 executes the sorting process and sorts the recalled information according to a preset strategy.

[0049] Among them, when the recall module 11 executes the recall process, the recall docking unit 31 calls the recall algorithm component 21 to generate the recall conditions, and then the recall module 11 recalls a certain number of information that meets the recall conditions in the database according to the recall conditions. Among them, the recall module 11 recalls a preset number of information based on the recall conditions to form a first information candidate set. When the sorting module 12 executes the sorting process, the sorting docking unit 32 calls the relevant sorting algorithm component 22 to sort the information in the first information candidate set accordingly. The configuration center 20 provides the collaborative docking subsystem 3 with the algorithm component identification parameters and the algorithm parameters required for each algorithm component. The docking units, such as the recall docking unit 31 and the sorting docking unit 32, call the corresponding algorithm component according to the algorithm component identification parameters, and pass the corresponding algorithm parameters to the corresponding algorithm component.

[0050] See also Figure 2 , Figure 2It is a block diagram of the system structure principle for the recall process according to an embodiment of the present invention. In this embodiment, the recall module 11 includes a trigger unit 111 and three recall engines, namely a vector recall engine 112, an inverted index recall engine 112 and a KV recall engine 113. Correspondingly, the recall docking unit 31 encapsulated in the recall module 11 includes a recall parameter subunit 311 and a call subunit 312. This embodiment abstracts the recall mode into three categories: vector recall, inverted index recall and KV index recall. All recall conditions belong to one of these three categories. The recall parameter subunit 311 can obtain two types of parameters from the configuration center 20, one is the recall category parameter, and the other is the algorithm component parameter. The algorithm component parameter includes the identification parameter of the algorithm component, such as name, ID, etc., which is used to identify the numerous algorithm components in the algorithm subsystem. The algorithm component parameter also includes the algorithm parameters used by the algorithm component, such as model configuration parameters, feature dimensions, etc. when the model is used in the algorithm. Correspondingly, the recall algorithm component 21 also includes multiple components, such as the DSSM recall condition component, the U2J (User To Job) recall condition component, the J2J (Job To Job) recall condition component, the functional recall condition component, etc., to generate recall conditions for different scenarios. Among them, the full name of the DSSM model is Deep Structured Semantic Model (deep semantic model), which was developed by Microsoft Research. It uses a deep neural network to represent text (sentences, search terms (Query), entities, etc.) as vectors, and calculates text similarity through vectors. The U2J recall condition component constructs recall conditions based on the user vector to position vector in the graph model. The J2J recall condition component constructs recall conditions based on the position vector to position vector of the user behavior (such as delivery) of the graph model.

[0051] When the trigger unit 111 determines that information recommendation is required, a driving instruction is issued to the recall parameter subunit 311 and the calling subunit 312. The recall parameter subunit 311 determines the recall category of this recall based on the information in the driving instruction, and determines the corresponding algorithm component parameters, and sends the algorithm component parameters to the calling subunit 312. The calling subunit 312 calls the corresponding algorithm component based on the algorithm component identifier in the received algorithm component parameters, and passes the corresponding algorithm parameters to the algorithm component. For example, the algorithm parameters of the recall algorithm component 21 correspond to a specific recall strategy, which is stored in the configuration center (a configuration file in the database). In one embodiment, the algorithm parameters include two categories: general parameters and experimental parameters. The experimental parameters include multiple groups of parameters of different groups, such as Group A experimental parameters and Group B experimental parameters. The recall parameter subunit 311 parses the driving instruction to determine whether the current algorithm parameters should be general parameters or experimental parameters. If they are experimental parameters, should they be Group A experimental parameters or Group B experimental parameters? After determination, the algorithm parameters are passed to the algorithm component.

[0052] The recall algorithm component 21 creates a recall channel based on the call of the calling sub-unit 312, sets a channel name for the recall channel, constructs recall conditions based on the algorithm parameters determined by the recall parameter sub-unit 311, and sends them to the recall engine corresponding to the corresponding category, and writes the corresponding recall channel and the constructed recall conditions into the log.

[0053] See also Figure 3 , Figure 3 This is a block diagram of the principles of an information recommendation system according to another embodiment of the present invention. The engineering subsystem 1 in the information recommendation system is connected to the front-end system and receives data sent by the front-end system, such as the user's current behavior data. The functional modules of the engineering subsystem 1 include a scenario module 13. The scenario module 13 determines the current recommendation scenario based on the front-end data and sends the recommendation scenario data to the recall module 11. The recall module 11 includes the recommendation scenario data in the driving instruction. The recall docking unit 31 determines the corresponding recall category based on the recommendation scenario data in the driving instruction. Figure 2 The recall parameter subunit 311 determines the corresponding recall category based on the recommendation scenario data in the driving instruction, and determines the recall algorithm component that inherits the recall category. The calling subunit 312 calls the corresponding recall algorithm component based on the identifier of the recall algorithm component, and the recall algorithm component generates a recall condition that meets the recommendation scenario. In a further embodiment, there can be one or more recall categories corresponding to a recommendation scenario, so multiple recall channels can be created to recall information from different angles. For example, when applying product recommendations on e-commerce platforms, collaborative filtering can be used to recall mobile phone models that are similar to those browsed by users in the past; semantic matching can be used to recall products with titles containing "mobile phone" and matching descriptions; and popular recall can also be used to recall mobile phones with high recent sales. Each method corresponds to a recall channel, and each recall channel provides a recall condition to the recall engine, that is, corresponds to a recommendation scenario, and realizes multi-angle information recall in parallel, thereby improving the comprehensiveness and accuracy of the recommendation.

[0054] See also Figure 4 , Figure 41 is a block diagram of the information recommendation system according to another embodiment of the present invention. In this embodiment, the sorting module 12 of the engineering subsystem 1 includes a coarse sorting unit 121 and a fine sorting unit 122. Correspondingly, the sorting layer of the collaborative docking subsystem 3 includes a coarse sorting sublayer and a fine sorting sublayer. The sorting docking unit 32 includes a coarse sorting docking unit 321 encapsulated in the coarse sorting unit 121 and a fine sorting docking unit 322 encapsulated in the fine sorting unit 122. The sorting algorithm component 22 in the algorithm subsystem 2 includes a coarse sorting algorithm component 221 and a fine sorting algorithm component 222. In the embodiment, the coarse sorting docking unit 321 calls the coarse sorting algorithm component 221 based on the sorting instruction of the coarse sorting unit 121 and the coarse sorting strategy configured by the configuration center to sort the information in the first information candidate set to obtain the second information candidate set; the fine sorting docking unit 322 calls the fine sorting algorithm component 222 based on the sorting instruction of the fine sorting unit 122 and the fine sorting strategy configured by the configuration center to perform refined scoring on the information in the second information candidate set, and sorts the information in the second information candidate subset based on the score to obtain a preset number of recommended information for output. In one embodiment, the coarse sorting algorithm component 221 includes multiple algorithm components with different preset rules, such as rules based on user attributes, rules based on information attributes, rules based on user behavior, business policy rules, etc., to filter out a portion of information from the recalled job information. For example, for the job recommendation system of the recruitment platform, the recalled information is job information. The rough sorting docking unit 321 selects a rough sorting algorithm component 221 with different rules based on the rough sorting strategy to filter out some job information from the first information candidate set. For example, some job information is filtered out based on geographic location, salary, etc., or some job information is filtered out based on user preferences, or some job information is filtered out based on job freshness, popularity, etc., or based on some business strategies, such as whether the position is a paid position, whether the display quantity limit for the same company or similar positions is reached, and other rules are used for filtering.

[0055] Different refined ranking strategies correspond to different refined ranking models and corresponding sorting strategies. For example, for a single-indicator refined ranking strategy, the refined ranking model may be a gradient boosting tree (GBDT / XGBoost / LightGBM), logistic regression (LR), or a deep learning model such as Wide&Deep, DeepFM, DIN (Deep Interest Network) and other machine learning models. The corresponding sorting strategy is to sort the scores of each information in the second information candidate set based on the model, and select a certain number (such as N) of information as recommended information for output.

[0056] The fine ranking strategy also includes a multi-index optimization strategy. The corresponding fine ranking model can be multiple models that implement different single indicators, or a multi-task learning model, such as ESMM (Entire Space Multi-task Model), MMoE (Multi-gate Mixture-of-Experts), etc. The corresponding sorting strategy includes: first, the scores of multiple single-index models are fused to obtain a final score. For a multi-task learning model used to obtain a multi-index score, a fusion part can be added to the output end of the model during model construction to directly output the fused final score when applied; or, the scores output by multiple task branches of the multi-task learning model are fused to obtain a final score. Then, sorting is performed based on the final score, and a certain number of information (such as N) ranked at the top are selected as the recommended information for output.

[0057] Therefore, the refined ranking algorithm components include a refined ranking model component, a ranking component, and a multi-objective fusion component. The refined ranking docking unit 322 determines the corresponding refined ranking strategy based on the configuration data provided by the configuration center. Based on the refined ranking strategy, it determines the applied refined ranking model and ranking component, as well as whether the multi-objective fusion component is required. The number of recommended information output, N, can be configured by the configuration center. The refined ranking docking unit 322 can obtain the number of recommended information output, N, based on the configuration data provided by the configuration center.

[0058] The fine ranking unit 122 obtains the model data, such as the feature data as the model input, based on the fine ranking model determined by the fine ranking docking unit 322, and passes it to the fine ranking model; the fine ranking model component configures the model operation-related parameters to the fine ranking model, and the fine ranking model processes the input feature data and outputs a score.

[0059] See also Figure 5 , Figure 5 It is a principle block diagram of an information recommendation system according to another embodiment of the present invention. In this embodiment, the functional module of the engineering subsystem 1 also includes a rearrangement module 14. Correspondingly, the algorithm component of the algorithm subsystem includes a rearrangement algorithm component 23. The collaborative docking subsystem 3 also includes a rearrangement layer. The multi-layer docking unit also includes a rearrangement docking unit 23 encapsulated in the rearrangement module 14.

[0060] When the fine sorting docking unit 322 calls the fine sorting algorithm component 222 based on the sorting instruction of the fine sorting unit 122 to perform refined scoring on the information in the second information candidate set, and sorts the information in the second information candidate set based on the score to obtain the third information candidate set, the rearrangement docking unit 33 calls the rearrangement algorithm component 23 based on the driving instruction of the rearrangement module 14 to rearrange the information in the third information candidate set to obtain recommended information for output.

[0061] The reordering docking unit 33 determines the corresponding reordering strategy through the configuration data provided by the configuration center, and determines the corresponding reordering algorithm component 23 based on the reordering strategy. In one embodiment, the reordering docking unit 33 identifies the determined reordering algorithm component 23 according to the reordering strategy, and passes the corresponding reordering parameters to the determined reordering algorithm component 23. The reordering parameters are, for example, parameter data corresponding to a set of specific reordering rules, such as parameters corresponding to filtering rules, parameters corresponding to weighting rules, parameters corresponding to fatigue control rules, etc. The reordering algorithm component 23 executes the corresponding reordering algorithm according to the received parameters, and reorders the information in the third information candidate set to obtain recommended information for output. In addition, there can also be multiple reordering algorithm components 23, each of which implements a reordering strategy.

[0062] See also Figure 6 , Figure 6 This is a principle block diagram of a job recommendation system in a recruitment platform according to an embodiment of the present invention. In this embodiment, the module in the rectangular box in the figure is an engineering module, which belongs to the engineering subsystem, the unit in the rounded box with a gray background belongs to the collaborative docking subsystem, and the component in the semi-circular box belongs to the algorithm subsystem.

[0063] The front-end system issues information recommendation requests based on user behavior. The request processing module receives and parses these requests, then sends the parsed data to the scenario module. The scenario module analyzes the user's current scenario and sends this data to the recall module. These scenarios include logging in, viewing job listings, and chatting with recruiters. Different scenarios correspond to different recommendation strategies, including recall and ranking strategies.

[0064] In this embodiment, the recall module includes a recall trigger module and a recall engine. The recall docking unit encapsulated in the recall module is connected to the configuration center. It determines the recall category and corresponding recall algorithm component based on the recall policy configured by the configuration center and the current scenario. It then invokes the recall algorithm component to generate recall conditions and sends them to the recall engine. Based on the recall conditions, the recall engine retrieves a preset number of job information from the index database to form a first set of job candidates, which it then sends to the sorting module.

[0065] The ranking module is used for coarse and / or fine ranking, and it encapsulates a ranking docking unit. The ranking docking unit is connected to the configuration center and, based on the ranking strategy configured by the configuration center, such as coarse or fine ranking strategy, calls the corresponding ranking algorithm component to score and sort the job information in the first job candidate set, thereby generating a second job candidate set and sending it to the re-ranking module.

[0066] The rearrangement module encapsulates a rearrangement docking unit, which is connected to the configuration center. According to the rearrangement strategy configured by the configuration center, the corresponding rearrangement algorithm component is called to rearrange the position information in the second position candidate set, such as filtering out some positions based on filtering rules, or adjusting the weights of certain indicators of the position information to adjust the current ranking, etc., or determining the position information with high rankings in the current position information but not adopted by users many times based on the fatigue control strategy, and removing these position information, etc. The position information after rearrangement constitutes the result set and is sent to the result processing module. After the result processing module performs corresponding processing, such as displaying the quantity, displaying the shift, etc., it is sent to the front-end system through the result return module and displayed to the user in the front-end system.

[0067] Although the above embodiments take a job recommendation system as an example, it can be known that the recommended information may also be resume information.

[0068] Through the system provided by the present invention, when implementing an information recommendation system, while taking into account the coordinated cooperation between engineering and algorithms, engineering developers can focus on the engineering system, and algorithm developers can focus on various logical algorithms in the recommendation process, thereby decoupling the work of engineering development and algorithm development, which is conducive to improving the performance and iterative optimization of the information recommendation system.

[0069] See also Figure 7 , Figure 7It is a principle block diagram of the algorithm subsystem according to an embodiment of the present invention. The various algorithms used to implement information recommendation in the present invention are provided by the algorithm subsystem 2 in the form of components, which not only has clear logic, but also facilitates the development, modification, upgrade, and online and offline of the algorithm. In the present invention, the implementation methods of the algorithm components include two categories: code implementation and configuration implementation. Among them, some complex algorithm logics are implemented through customized code blocks. For example, algorithm components for generating recall conditions, algorithm components for implementing complex re-arrangement logic, etc. For some algorithms with fixed logic patterns, they are implemented by configuration based on the parameters passed by the developer by the configuration center. Therefore, for this type of algorithm component, it has a corresponding algorithm parameter configuration interface, and the developer configures the corresponding algorithm parameters for the algorithm component through the algorithm parameter configuration interface when needed. For example, the recall algorithm component includes a recall channel component and a recall condition construction component. When creating a recall channel, the recall channel component needs to obtain the algorithm parameters corresponding to the recall strategy applied by the recall channel. Algorithm parameters include general parameters and experimental parameters. The algorithm parameter configuration interface is used to specify the algorithm parameters corresponding to the recall channel as general parameters or experimental parameters, so that the recall channel can be applied in different scenarios. For example, when it needs to be applied in a common online recommendation scenario, the algorithm parameter configuration interface is used to specify the algorithm parameters corresponding to the recall channel as using general parameters. When an AB experiment is required, the algorithm parameter configuration interface is used to specify the algorithm parameters corresponding to the recall channel as using either Group A or Group B experimental parameters in the experimental parameters. When implementing the recall algorithm, the recall channel obtains specific parameter data from the configuration center based on the specified algorithm parameters. This avoids repeated development of algorithm logic, improves development efficiency, and reduces the failure rate of the algorithm.

[0070] In addition, when adjusting algorithm parameters, developers can configure all algorithm parameters of the algorithm components through the algorithm parameter configuration interface, thereby greatly improving the efficiency of algorithm parameter adjustment.

[0071] Furthermore, in iterative experiments for algorithm optimization, developers can use the algorithm parameter configuration interface to enter the algorithm components to be experimented and the corresponding algorithm parameter configuration data as experimental parameters, and configure them into the experimental group parameters according to the agreed format. This does not require writing any code and is not restricted by whether the system is online, thereby improving the iterative efficiency of the algorithm.

[0072] The algorithm subsystem 2 of the present invention includes a hot-swap configuration interface for receiving hot-swap configuration parameters and implementing the online or offline operation of algorithm components based on the received hot-swap configuration parameters. For example, the present invention provides a hot-swap configuration interface in the form of a configuration item. Developers configure the package name of the algorithm component to be online or offline in the configuration item. After the configuration center in the algorithm subsystem 2 receives the package name of the algorithm component to be online or offline through the hot-swap configuration interface, it will online or offline the corresponding algorithm component.

[0073] The algorithm subsystem of the present invention only needs to focus on the recommendation strategy and its logical algorithm, without having to focus on engineering implementation. This also avoids duplicate development, achieves engineering-algorithm decoupling, reduces the communication costs between the engineering team and the algorithm team, and avoids the various defects caused by the algorithm team's heavy involvement in engineering implementation, thereby effectively improving the performance of the information recommendation system. At the same time, the algorithm configuration implementation method adopted by the present invention means that the tuning of most algorithms does not require any code development, and the experimental launch only requires the configuration of experimental parameters. This not only reduces the development workload, but also greatly reduces the workload of algorithm tuning, experimental methods, etc., making the subsequent maintenance and expansion of the system simple and easy to operate, and improving the overall iterative efficiency of the system.

[0074] Figure 8 This is a flow chart of a method for implementing an information recommendation system according to an embodiment of the present invention. Figure 1 Taking the information recommendation system shown in FIG. 1 as an example, the method includes the following steps:

[0075] In step S11, the engineering developer creates an engineering subsystem. The engineering subsystem 1 includes multiple functional modules for implementing information recommendation, such as an execution scheduler 10, a recall module 11, a sorting module 12, and other functional modules, such as other business modules and basic functional modules.

[0076] Step S12: Create a collaborative docking subsystem. The collaborative docking subsystem 3 includes multiple layers of docking units, each of which is encapsulated in a corresponding functional module of the engineering subsystem. In this embodiment, the collaborative docking subsystem 3 includes a recall layer and a ranking layer. The recall docking unit 31 in the recall layer is encapsulated in the recall module 11, and the ranking docking unit 32 is encapsulated in the ranking module 12.

[0077] In step S2, the algorithm developer creates an algorithm subsystem. The algorithm subsystem includes multiple algorithm components that implement the information recommendation logic algorithm. In this embodiment, the algorithm subsystem 2 includes a configuration center 20, one or more recall algorithm components 21, and one or more sorting algorithm components 22.

[0078] During the information recommendation process implemented by the functional modules in the engineering subsystem, the docking unit calls the corresponding algorithm component in the algorithm subsystem based on the driving instructions of the corresponding functional module, and the algorithm component completes the corresponding recommendation logic algorithm. For example, when the recall module 11 executes the recall process, the recall docking unit 31 calls the recall algorithm component 21 to generate the recall conditions. The recall module 11 then recalls a certain number of information that meets the recall conditions from the database based on the recall conditions. The recall module 11 recalls a preset number of information based on the recall conditions to form a first information candidate set. When the sorting module 12 executes the sorting process, the sorting docking unit 32 calls the relevant sorting algorithm component 22 to sort the information in the first information candidate set accordingly. The configuration center 20 provides the collaborative docking subsystem 3 with the algorithm component identification parameters and the algorithm parameters required by each algorithm component. The docking units, such as the recall docking unit 31 and the sorting docking unit 32, call the corresponding algorithm component based on the algorithm component identification parameters and pass the corresponding algorithm parameters to the corresponding algorithm component.

[0079] When the rough sorting and fine sorting of information are realized in the recommendation process according to the recommendation strategy, in step S11, the engineering subsystem 1 created by the engineering developer is as follows: Figure 4 As shown, the sorting module 12 includes a rough sorting unit 121 and a fine sorting unit 122. In step S12, the sorting docking unit 32 in the collaborative docking subsystem created includes a rough sorting docking unit 321 encapsulated in the rough sorting unit 121 and a fine sorting docking unit 322 encapsulated in the fine sorting unit 122. In step S2, the sorting algorithm component 22 in the algorithm subsystem created by the algorithm developer includes a rough sorting algorithm component 221 and a fine sorting algorithm component 222. Figure 4 The description will not be repeated here.

[0080] When the information is rearranged in the recommendation process according to the recommendation strategy, in step S11, the engineering subsystem 1 created by the engineering developer is as follows: Figure 5 As shown, it also includes a rearrangement module 14. In step S12, the collaborative docking subsystem created also includes a rearrangement docking unit 23 encapsulated in the rearrangement module 14. In step S2, the algorithm subsystem created by the algorithm developer also includes a rearrangement algorithm component 23. For details, please refer to Figure 5 The description will not be repeated here.

[0081] When creating an algorithm subsystem, algorithm developers implement the corresponding algorithm logic as components. They also create algorithm parameter configuration interfaces for some algorithm components to enable parameter configuration. To enable hot-swapping of algorithm components, a corresponding hot-swap configuration interface is also created. This interface allows you to configure the package name of the algorithm component to be brought online or offline, thus enabling the online and offline operation of the algorithm component.

[0082] In another aspect, the present invention further provides an electronic device, see Figure 9 , Figure 9 6 is a schematic diagram of the hardware structure principle of an electronic device according to an embodiment of the present invention. The electronic device can be implemented as a server or various other terminal devices, such as a desktop personal computer, a tablet computer, a laptop computer, a mobile phone, etc., and includes a processor 601 and a memory 602. The memory 602 stores a program instruction set. When the processor 601 executes the program instruction set on the memory 602, any of the aforementioned information recommendation systems or information recommendation system implementation methods is implemented.

[0083] Specifically, the processor 601 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiment of the present invention.

[0084] The memory 602 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 602 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 602 may include removable or non-removable (or fixed) media. Where appropriate, the memory 602 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 602 is a non-volatile solid-state memory.

[0085] The memory may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it implements any of the aforementioned information recommendation systems or information recommendation system implementation methods.

[0086] In one example, the electronic device may further include a communication interface 603 and a bus 604. The processor 601, the memory 602, and the communication interface 603 are connected via the bus 604 and communicate with each other.

[0087] The communication interface 603 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiment of the present invention.

[0088] Bus 604 includes hardware, software or both, and the components of online data flow metering equipment are coupled to each other. For example, but not limitation, bus can include accelerated graphics port (AGP) or other graphics bus, enhanced industry standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industry standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnect (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 604 can include one or more buses. Although the embodiment of the present invention describes and shows a specific bus, the present invention considers any suitable bus or interconnection.

[0089] The present invention also provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement any one of the information recommendation systems or information recommendation system implementation methods in the aforementioned embodiments. The computer-readable storage medium can be any medium that can tangibly contain or store computer-executable instructions for use by or in combination with an instruction execution system, device, and apparatus. The storage medium can be a transient computer-readable storage medium or a non-transient computer-readable storage medium. Non-transient computer-readable storage media may include, but are not limited to, magnetic storage devices, optical storage devices, and / or semiconductor storage devices. Examples of such storage devices include, for example, magnetic disks, optical disks based on CD, DVD, or Blu-ray technology, and persistent solid-state memories such as flash memory, solid-state drives, and the like.

[0090] The present invention also provides a computer program product comprising a set of computer program instructions that, when executed by a processor, implement any of the information recommendation systems or methods for implementing information recommendation systems described in the aforementioned embodiments. The computer program product includes, but is not limited to, an application installation package published on a website or in an app store, an application plug-in, or a mini-program that can be run within certain applications.

[0091] The above embodiments are only used to illustrate the present invention, and are not intended to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the scope of the present invention. Therefore, all equivalent technical solutions should also fall within the scope of the present invention.

Claims

1. An information recommendation system, characterized in that: include: An engineering subsystem, comprising a plurality of functional modules for implementing information recommendation; an algorithm subsystem, the algorithm subsystem including a plurality of algorithm components for implementing an information recommendation logic algorithm; and A collaborative docking subsystem includes multi-layer docking units, which are respectively encapsulated in corresponding functional modules of the engineering subsystem; in the process of implementing information recommendation by the functional modules in the engineering subsystem, the docking unit calls the corresponding algorithm component in the algorithm subsystem based on the driving instructions of the corresponding functional module, and the algorithm component completes the corresponding recommendation logic algorithm.

2. The information recommendation system according to claim 1, characterized in that The algorithm subsystem includes a hot-swap configuration interface for receiving hot-swap configuration parameters of an algorithm component and implementing online or offline operation of a corresponding algorithm component according to the received hot-swap configuration parameters.

3. The information recommendation system according to claim 1, characterized in that Each algorithm component in the algorithm subsystem includes an algorithm parameter configuration interface for providing a configuration channel for algorithm parameters.

4. The information recommendation system according to claim 1, characterized in that It also includes a configuration center for providing the collaborative docking subsystem with algorithm component identification parameters and the algorithm parameters required by each algorithm component; correspondingly, the docking unit calls the corresponding algorithm component based on the algorithm component identification parameters and passes the corresponding algorithm parameters to the corresponding algorithm component.

5. The information recommendation system according to any one of claims 1 to 4, characterized in that: The collaborative docking subsystem includes a recall layer and a ranking layer, and the multi-layer docking unit includes a recall docking unit and a ranking docking unit; the functional modules of the engineering subsystem include a recall module and a ranking module, and the algorithm components in the algorithm subsystem include a recall algorithm component and a ranking algorithm component; Among them, the recall docking unit is encapsulated in the recall module, and the sorting docking unit is encapsulated in the sorting module. When the recommendation conditions are met, the recall module drives the recall docking unit to call the recall algorithm component to generate the recall conditions; the recall module recalls a preset amount of information based on the recall conditions to form a first information candidate set; the sorting module drives the sorting docking unit to call the sorting algorithm component to sort the information in the first information candidate set.

6. The information recommendation system according to claim 5, characterized in that The recall docking unit includes: a recall parameter subunit configured to determine a corresponding recall category and a recall algorithm component inheriting the recall category based on the driving instruction of the recall module; and The calling subunit is configured to call the recall algorithm component that inherits the recall category and pass the corresponding algorithm parameters to the recall algorithm component.

7. The information recommendation system according to claim 6, characterized in that The recall categories include multiple ones, and each recall algorithm component inherits one recall category.

8. The information recommendation system according to claim 6, characterized in that The functional modules of the engineering subsystem further include a scenario module configured to determine a recommended scenario based on the front-end data and send the recommended scenario data to the recall module; The recall module includes the recommended scene data in the driving instruction, and the recall parameter subunit determines the corresponding recall category and the recall algorithm component that inherits the recall category based on the recommended scene data in the driving instruction; the recall algorithm component that inherits the recall category generates a recall condition that meets the recommended scene.

9. The information recommendation system according to claim 8, characterized in that There are one or more recall categories corresponding to a recommendation scenario, and correspondingly, there are one or more recall conditions that meet the recommendation scenario.

10. The information recommendation system according to claim 5, characterized in that The sorting module of the engineering subsystem includes a coarse sorting unit and a fine sorting unit. Correspondingly, the sorting layer of the collaborative docking subsystem includes a coarse sorting sublayer and a fine sorting sublayer. The sorting docking unit includes a coarse sorting docking unit encapsulated in the coarse sorting unit and a fine sorting docking unit encapsulated in the fine sorting unit. The sorting algorithm component in the algorithm subsystem includes a coarse sorting algorithm component and a fine sorting algorithm component, wherein the coarse sorting docking unit calls the coarse sorting algorithm component based on the sorting instruction of the coarse sorting unit to filter part of the information from the first information candidate set to obtain the second information candidate set; the fine sorting docking unit calls the fine sorting algorithm component based on the sorting instruction of the fine sorting unit to perform refined scoring on the information in the second information candidate set, and filters out part of the information from the second information candidate set based on the scoring.

11. The information recommendation system according to claim 10, characterized in that: The functional module of the engineering subsystem further includes a rearrangement module. Correspondingly, the algorithm component of the algorithm subsystem includes a rearrangement algorithm component. The collaborative docking subsystem further includes a rearrangement layer. The multi-layer docking unit further includes a rearrangement docking unit encapsulated in the rearrangement module. When the fine sorting docking unit calls the fine sorting algorithm component based on the sorting instruction of the fine sorting unit to perform refined scoring on the information in the second information candidate set, and filters out part of the information from the second information candidate set based on the scoring to obtain the third information candidate set, the rearrangement docking unit calls the rearrangement algorithm component based on the driving instruction of the rearrangement module to rearrange the information in the third information candidate set to obtain recommended information for output.

12. The information recommendation system according to claim 11, characterized in that The re-ranking algorithm component includes one or more filtering components, one or more weighting components, and one or more fatigue control components.

13. The information recommendation system according to claim 11, characterized in that: The refined ranking algorithm component also includes a multi-objective fusion optimization component, which is configured to achieve multi-objective fusion optimization when performing refined scoring on the information in the second information candidate set.

14. A method for implementing an information recommendation system, characterized in that: include: Creating an engineering subsystem, wherein the engineering subsystem includes a plurality of functional modules for implementing information recommendation; Creating an algorithm subsystem, wherein the algorithm subsystem includes a plurality of algorithm components for completing an information recommendation logic algorithm; Create a collaborative docking subsystem, wherein the collaborative docking subsystem includes multi-layer docking units, and the multi-layer docking units are respectively encapsulated in the corresponding functional modules of the engineering subsystem. In the process of implementing information recommendation by the functional modules in the engineering subsystem, the docking unit calls the corresponding algorithm component in the algorithm subsystem based on the driving instructions of the corresponding functional module, and the algorithm component completes the corresponding recommendation logic algorithm.

15. An electronic device comprising a processor and a memory, wherein the memory stores a computer program instruction set, characterized in that: When the processor executes the computer program instruction set on the memory, the information recommendation system according to any one of claims 1 to 13 is implemented or the implementation method of the information recommendation system according to claim 14 is implemented.

16. A computer-readable storage medium, wherein: The computer-readable storage medium stores a computer program instruction set, wherein when the computer program instruction set is executed by a processor, the information recommendation system according to any one of claims 1 to 13 or the method for implementing the information recommendation system according to claim 14 is implemented.

17. A computer program product comprising a set of computer program instructions, characterized in that When the computer program instruction set is executed by a processor, it implements the information recommendation system according to any one of claims 1 to 13 or the implementation method of the information recommendation system according to claim 14.

Images