Information recommendation method, device and equipment

Abstract

Embodiments of the invention disclose an information recommendation method, apparatus and device. The method comprises the steps of receiving a user recommendation request for a target service; in response to the user recommendation request, obtaining a candidate set formed by recommendable information of the target service; the recommendation score corresponding to the recommendable information in the candidate set is queried from a pre-generated information recommendation set for a user corresponding to the user recommendation request, and the generated information recommendation set is based on user behavior sequence data in advance; calling a generative recommendation model to perform recommendation scoring on recommendable information of a plurality of different businesses including the target business to obtain a recommendation information set; and selecting target recommendation information from the candidate set based on the recommendation score corresponding to the recommended information in the queried candidate set, and sending the target recommendation information to the user.

Description

Technical Field

[0001] This document relates to the field of computer technology, and in particular to an information recommendation method, apparatus, and device. Background Technology

[0002] With the rapid development of internet technology, information recommendation has become an indispensable technology for major internet platforms. In particular, in recent years, with breakthroughs in generative artificial intelligence technology, generative recommendation systems have demonstrated powerful personalized recommendation capabilities.

[0003] Currently used information recommendation mechanisms include online and offline mechanisms. However, while protecting user privacy, online recommendation mechanisms struggle to meet time-consuming requirements and waste computational resources. Offline recommendation mechanisms, on the other hand, suffer from poor real-time performance, failing to reflect changes in user interests promptly, and have long update cycles (typically days), making them ill-suited for handling real-time personalized needs. Therefore, a superior information recommendation solution is needed to improve response speed, reduce computational resource consumption, enable cross-scenario sharing of computation results, and be applicable to high-traffic recommendation scenarios. Summary of the Invention

[0004] The purpose of the embodiments in this specification is to provide a better information recommendation scheme to improve the response speed of information recommendation, reduce the consumption of computing resources, realize the cross-scenario sharing of calculation results, and be applicable to high-traffic recommendation scenarios.

[0005] To achieve the above technical solution, the embodiments in this specification are implemented as follows: This specification provides an information recommendation method, comprising: receiving a user recommendation request for a target service; responding to the user recommendation request by obtaining a candidate set of recommendable information for the target service; querying a pre-generated information recommendation set for the user corresponding to the user recommendation request, wherein the generated information recommendation set is a set of recommendation information obtained in advance by calling a generative recommendation model to score recommendable information for multiple different services, including the target service, based on user behavior sequence data; selecting target recommendation information from the candidate set based on the queryed recommendation scores of the recommendable information, and sending the target recommendation information to the user.

[0006] This specification provides an information recommendation device, comprising: a request module for receiving user recommendation requests for a target service; a response module for responding to the user recommendation request and obtaining a candidate set of recommendable information for the target service; a score query module for querying recommendation scores corresponding to recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the user recommendation request, wherein the generated information recommendation set is a set of recommendation information obtained in advance based on user behavior sequence data and by calling a generative recommendation model to score recommendable information for multiple different services, including the target service; and an information recommendation module for selecting target recommendation information from the candidate set based on the queried recommendation scores corresponding to the recommendable information in the candidate set and sending the target recommendation information to the user.

[0007] This specification provides an information recommendation device, comprising: a processor; and a memory arranged to store computer-executable instructions, wherein the executable instructions, when executed, cause the processor to: receive a user recommendation request for a target service; in response to the user recommendation request, obtain a candidate set of recommendable information for the target service; query a pre-generated information recommendation set for the user corresponding to the user recommendation request, and query the recommendation scores corresponding to the recommendable information in the candidate set, wherein the generated information recommendation set is a set of recommendation information obtained in advance based on user behavior sequence data, by calling a generative recommendation model to score the recommendable information of multiple different services, including the target service; and select target recommendation information from the candidate set based on the recommendation scores corresponding to the recommendable information in the queried candidate set, and send the target recommendation information to the user.

[0008] This specification also provides a storage medium for storing computer-executable instructions. When executed by a processor, the executable instructions implement the following process: receiving a user recommendation request for a target service; responding to the user recommendation request, obtaining a candidate set of recommendable information for the target service; querying the recommendation score corresponding to the recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the user recommendation request, wherein the generated information recommendation set is a set of recommendation information obtained in advance based on user behavior sequence data, by calling a generative recommendation model to score the recommendable information of multiple different services, including the target service; selecting target recommendation information from the candidate set based on the recommendation score corresponding to the recommendable information in the queried candidate set, and sending the target recommendation information to the user.

[0009] This specification also provides a computer program product, including a computer program that, when executed by a processor, implements the following process: receiving a user recommendation request for a target service; responding to the user recommendation request, obtaining a candidate set of recommendable information for the target service; querying the recommendation scores corresponding to recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the user recommendation request, wherein the generated information recommendation set is a set of recommendation information pre-obtained by calling a generative recommendation model to recommend and score recommendable information for multiple different services, including the target service, based on user behavior sequence data; selecting target recommendation information from the candidate set based on the recommendation scores corresponding to the recommendable information in the queried candidate set, and sending the target recommendation information to the user. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 This is a schematic diagram of the structure of an information recommendation system as described in this specification; Figure 2 This is a schematic diagram of an information recommendation process described in this specification; Figure 3 This is a schematic diagram illustrating the process of constructing a set of recommendation information as described in this specification; Figure 4 This document presents a schematic diagram of a recommended system architecture for a fund business. Figure 5 This is a schematic diagram of an information recommendation device described in this specification; Figure 6 This is a schematic diagram of an information recommendation device described in this specification. Detailed Implementation

[0011] This specification provides an information recommendation method, apparatus, and device through its embodiments.

[0012] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0013] This specification provides a generative recommendation mechanism based on a nearline architecture. With the development of generative artificial intelligence technology, generative recommendation systems have shown great potential. However, it is difficult to directly apply generative recommendation models to currently used recommendation architectures. This is because current recommended architectures often employ online real-time computation. That is, when a user visits a page, user features need to be queried in real time, a specified algorithm needs to be called, and the recommendation result needs to be returned and displayed to the user. To ensure a good user experience, the recommendation result usually needs to be obtained within milliseconds. Furthermore, generative recommendation systems need to call large models for recommendation, resulting in long inference times, typically on the order of seconds, which does not meet the requirements of online systems. In addition to timeliness requirements, a page typically consists of multiple modules (such as a cash flow product area and a stable product area), each of which requests the recommendation model. However, generative recommendation results are only related to user behavior sequence data. That is, when the user behavior sequence data is the same, the recommendation results should also be the same and stable. Independent requests to the algorithm model in each scenario will cause duplicate calculations, resulting in a waste of computing resources. Furthermore, as generative recommendation algorithms are promoted in more and more scenarios, the demand for computing power will increase linearly. Moreover, although there are some information recommendation systems that use nearline architecture, there is no nearline architecture design specifically for generative recommendations. They only use nearline computation for feature processing, and model calls are still real-time. This specification provides an optional solution that uses a near-line calculation mechanism triggered in real time based on user behavior sequence data and a generative recommendation model to generate recommendation scores corresponding to recommendable information. Subsequently, the generated recommendation scores can be used to quickly complete the information recommendation. Thus, when applying the generative recommendation model to high-traffic recommendation scenarios, it balances the contradiction between the real-time performance and the efficiency of generative recommendation, improves the response speed of information recommendation, reduces the consumption of computing resources, and realizes cross-scenario sharing of calculation results. For specific processing details, please refer to the specific content in the following embodiments.

[0014] The information recommendation method provided in one or more embodiments of this specification is applicable to the information recommendation implementation environment. (Refer to...) Figure 1 The implementation environment includes at least: client 100 and server 200, wherein: Client 100 can run on terminal devices, which can be mobile phones, personal computers, tablets, e-book readers, wearable devices, devices that interact with information based on AR (Augmented Reality) and VR (Virtual Reality), and laptop computers, etc. Client 100 can be installed on terminal devices. Client 100 can be an application, a browser, or a subroutine embedded in an application, etc.

[0015] Server 200 can run on a server, which can be one or more servers, a server cluster consisting of several servers, or a cloud server on a cloud computing platform. Server 200 can be installed on the server, and server 200 can be an application or a subroutine embedded in an application.

[0016] In this implementation environment, server 200 can use the user behavior sequence data collected by client 100 to call a generative recommendation model to score the recommendable information of multiple different services, including the target service, and generate an information recommendation set for the user. When server 200 receives a user recommendation request for the target service sent by user through client 100 installed on terminal device, it obtains a candidate set of recommendable information for the target service. Then, it queries the recommendation score corresponding to the recommendable information in the candidate set from the pre-generated information recommendation set for the user corresponding to the recommendation request. Based on the recommendation score corresponding to the recommendable information in the candidate set, it selects the target recommendation information from the candidate set and sends the target recommendation information to client 100. Client 100 can then display the target recommendation information to the user. In this way, the information recommendation is completed quickly directly through the generated recommendation score, improving the response speed of information recommendation, reducing the consumption of computing resources, and realizing the cross-scenario sharing of calculation results. When applying the generative recommendation model to high-traffic recommendation scenarios, it can also balance the contradiction between the real-time performance and the performance of generative recommendation.

[0017] like Figure 2 As shown in the embodiments of this specification, an information recommendation method is provided. The execution subject of this method can be a terminal device or a server, etc. The terminal device can be a mobile terminal device such as a mobile phone or tablet computer, a computer device such as a laptop or desktop computer, or an IoT device (specifically, a smartwatch, in-vehicle device, etc.). The server can be a single server or a server cluster composed of multiple servers. The server can be a backend server for financial business or online shopping business, or a backend server for an application, etc. This embodiment uses a server as the execution subject for detailed description. For the case where the execution subject is a terminal device, please refer to the following server case processing, which will not be repeated here. The method may specifically include the following steps: In step S202, a user recommendation request for the target service is received.

[0018] The target business can be any business, and can include a variety of businesses, such as fund business, stable wealth management business, shopping-related business (such as women's clothing selection business, digital product selection business, etc.), short video browsing or refreshing-related business (such as technology short video browsing or refreshing-related business, review short video browsing or refreshing-related business, etc.), news information browsing or refreshing-related business (such as domestic news browsing or refreshing-related business, entertainment news browsing or refreshing-related business, etc.), insurance business, etc. The specific target business can be set according to the actual situation, and the embodiments in this specification do not limit it.

[0019] In implementation, user recommendation requests can be triggered in various ways. For example, a user's terminal device may have a specific client installed (such as a financial client, shopping client, short video client, or news client). When the user launches this client, it can generate a corresponding user recommendation request. Alternatively, after launching the client, when the user needs to switch to or view a specific service, clicking the hyperlink or button for that specific service can generate a corresponding user recommendation request. For instance, after launching a financial client, if the user needs to view the fund service project, they can click the financial service button, and the client can generate a user recommendation request for the fund service. Or, after launching the client, the user can view the current page content, and when the user needs to refresh the current page, they can pull down the current page or directly click the refresh button, and the client can generate a corresponding user recommendation request. The specific details can be configured according to the actual situation.

[0020] After the client generates a user recommendation request for the target service using the above method, the client can trigger the terminal device to send the user recommendation request to the server, and the server can receive the user recommendation request for the target service.

[0021] In step S204, in response to the aforementioned user recommendation request, a candidate set consisting of recommendable information for the target service is obtained.

[0022] Among them, recommendable information can be information that can be recommended to users, and the corresponding information can include non-recommended information. Non-recommended information can be information that is unsuitable, cannot be, or is restricted from being recommended to users. For example, for minors, non-recommended information can include information about books, videos, games, etc. that have a negative connotation. The specifics can be set according to the actual situation.

[0023] In implementation, after receiving a user recommendation request for a target business through the above method, the server can query all projects or products included in the target business. For example, if the target business is a fund business, which may include 50 specific fund products (such as Fund 1, Fund 2, ... Fund 50), the server can query all fund products included in that fund business, i.e., the aforementioned 50 specific fund products (such as Fund 1, Fund 2, ... Fund 50). As another example, if the target business is an insurance business, which may include 30 specific insurance products (such as Insurance Product 1, Insurance Product 2, ... Insurance Product 30), the server can query all insurance products included in that insurance business, i.e., the aforementioned 30 specific insurance products (such as Insurance Product 1, Insurance Product 2, ... Insurance Product 30). Yet another example, if the target business is a women's clothing selection business, which may include 1500 specific women's clothing items, the server can query all women's clothing items included in that women's clothing selection business, i.e., the aforementioned 1500 specific women's clothing items, and so on. Then, all projects or products included in the target business can be analyzed to determine the recommendable information that can be recommended to users. Specifically, relevant user attribute information (such as the user's location, education, and employment status) and / or the user type targeted by the target business (such as users in a certain area, college students, middle school students, etc.) can be obtained. Based on the relevant user attribute information and / or the user type targeted by the target business, projects or products that match the above information can be selected from all projects or products included in the target business. The selected projects or products can be used as recommendable information of the target business, and a candidate set can be constructed through the recommendable information of the target business.

[0024] It should be noted that the selected projects or products can be all projects or products included in the target business, or only some products (or projects) among all products (or projects) included in the target business. In addition to selecting matching projects or products through the above methods, other methods can also be used, such as selection based on expert experience. The specific methods can be set according to the actual situation.

[0025] In step S206, the recommendation scores corresponding to the recommendable information in the candidate set are queried from the pre-generated information recommendation set for the user corresponding to the user's recommendation request. The generated information recommendation set is a set of recommendation information obtained in advance based on user behavior sequence data, by calling a generative recommendation model to recommend and score the recommendable information of multiple different businesses, including the target business.

[0026] User behavior sequence data can be a sequence of actions performed by a user during the execution of a certain business, arranged chronologically or sequentially (such as clicking, swiping, inputting, etc.). Furthermore, user behavior sequence data can also include sequence data of actions performed by the user during other business or operations prior to the current business. For example, if a user browsed or executed advanced financial management services before browsing or executing the "Stable Financial Management" service, then the user behavior sequence data would include not only the sequence of actions performed during the execution of the "Stable Financial Management" service (clicking the "Stable Financial Management" icon or button - clicking a specific financial product - browsing...), but also the sequence of actions taken during the execution of the "Stable Financial Management" service. In addition to browsing data, clicking the "Buy Now" button, entering the purchase amount, entering the purchase amount, and clicking the "Confirm" button (the specific settings can be customized according to actual circumstances), the data can also include behavioral sequence data during the browsing or execution of advanced financial management services. In this case, user behavioral sequence data can include, for example, browsing advanced financial management services (clicking the "Advanced Financial Management Service" icon or button, clicking a specific financial product, and browsing data). It should be noted that user behavioral sequence data can also only include user behavioral sequence data during the current browsing or execution of stable financial management services; the specific settings can be customized according to actual circumstances, and this specification does not limit this. The generative recommendation model views recommendation information as a sequence-to-sequence transformation process, predicting subsequent tokens based on the user's historical behavioral sequence data. Combining the advantages of generative AI (such as large language models) and commonly used information recommendation mechanisms, it can generate personalized and highly interpretable recommendation information.

[0027] In implementation, considering that online recommendation mechanisms cannot meet the time requirements of recommendation and will waste computing resources, while offline recommendation mechanisms have poor real-time performance, cannot reflect changes in user interests in a timely manner, have long update cycles, and are difficult to handle real-time personalized needs, this specification's embodiments introduce a near-line pre-computation mechanism to avoid real-time computing overhead. This mechanism separates the time-consuming generative recommendation model calculation from the request path, performing only the result query of the recommendation score when making online user recommendation requests, thus meeting the time requirements of recommendation. Simultaneously, it enables the reuse of calculation results across multiple scenarios, significantly reducing computing resource consumption. The near-line pre-computation mechanism is a data storage and access mechanism between online and offline recommendation mechanisms. Under the near-line pre-computation mechanism, the execution process of the calculation is exactly the same as the online recommendation mechanism, but it does not require immediate provision of results after calculation; instead, the results can be stored, making it asynchronous. Specifically, once user behavior data (such as clicks, purchases, browsing, etc.) is captured, a near-line pre-calculation mechanism can be triggered. This mechanism organizes the captured user behavior data into a user behavior sequence for that user. Then, it can obtain recommended information from multiple preset different services (this could be obtaining all items or products included in multiple preset different services, or using all of the aforementioned items or products as recommended information, or even using only some of the aforementioned items or products as recommended information, etc.). For example, for wealth management, the preset multiple different services could include fund services, stable wealth management services, advanced wealth management services, stock services, etc., and the recommended information from these multiple different services could include recommended information from fund services, stable wealth management services, advanced wealth management services, stock services, etc. The near-line pre-computation mechanism can invoke a generative recommendation model. User behavior sequence data and pre-defined recommendable information from multiple different services can be input into the generative recommendation model. The user behavior sequence data guides the generative recommendation model to score the recommendable information from these services. Based on the score, the recommendable information from these services can be sorted according to its matching degree with the user. The recommendable information with the highest matching degree is placed first and given a higher recommendation score. This sorting result and recommendation score are obtained in this way. Finally, a set of recommendation information is obtained, consisting of recommendable information from multiple different services and corresponding recommendations. This set of recommendation information and the user's relevant information can be stored in a designated database (such as Redis) to ensure data persistence.

[0028] After obtaining the candidate set of recommendable information for the target business through the above processing, the database can be accessed, and the set of recommended information corresponding to the user can be obtained from the database. Then, the recommendation score corresponding to each recommendable information in the candidate set can be queried from the set of recommended information.

[0029] In step S208, based on the recommendation scores corresponding to the recommendable information in the queried candidate set, target recommendation information is selected from the candidate set and sent to the user.

[0030] In implementation, after retrieving the recommendation score for each recommendable piece of information from the candidate set, the recommendation scores of different recommendable pieces of information in the candidate set can be compared. Recommendable information with a recommendation score greater than a preset threshold can be selected as the target recommendation information. Alternatively, the recommendable information with the highest recommendation score can be selected as the target recommendation information. Furthermore, the different recommendable pieces of information in the candidate set can be sorted according to their recommendation scores, with higher-scoring recommendable information appearing earlier in the ranking. A preset number of recommended pieces of information at the top of the ranking can be selected as the target recommendation information. For example, the top 5 recommended pieces of information can be selected as the target recommendation information, and so on, depending on the actual situation. The target recommendation information can be sent to the user, and the user's terminal device can display the target recommendation information. If there are multiple target recommendations, multiple different target recommendations can be displayed simultaneously on the corresponding client, or multiple different target recommendations can be displayed cyclically on the corresponding client at certain time intervals (such as 5 seconds or 10 seconds), etc., depending on the actual situation.

[0031] This specification provides an information recommendation method. It receives user recommendation requests for a target service, responds to these requests by obtaining a candidate set of recommendable information for the target service, and then queries the recommendation scores of the recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the recommendation request. The generated information recommendation set is a set of recommended information obtained in advance by calling a generative recommendation model to score recommendable information from multiple different services, including the target service, based on user behavior sequence data. Finally, based on the recommendation scores of the recommendable information in the queried candidate set, target recommended information can be selected from the candidate set and sent to the user. This method utilizes a real-time near-line calculation mechanism triggered by user behavior sequence data and a generative recommendation model to generate recommendation scores for recommendable information. Subsequently, the generated recommendation scores can be used to quickly complete the information recommendation, thereby improving the response speed of information recommendation, reducing computational resource consumption, and enabling cross-scenario sharing of calculation results. Furthermore, when applying the generative recommendation model to high-traffic recommendation scenarios, it can balance the trade-off between real-time performance and performance.

[0032] In practical applications, the above information recommendation set can be constructed in the following way, as detailed in steps A2 to A8 below.

[0033] In step A2, user behavior sequence data generated by the user is obtained.

[0034] In implementation, the client on the user's terminal device can detect the user's behavioral data on the terminal device. After the terminal device captures the user's behavioral data, it can record the captured behavioral data and send it to the server. The server can then obtain the user's behavioral sequence data. In addition, the user's behavioral sequence data can be stored in a behavioral sequence database to ensure data persistence.

[0035] In step A4, recommendable information for multiple different services is obtained, including the target service.

[0036] In implementation, to manage recommendable information from multiple different business segments, a corresponding database can be set up to store this information. Alternatively, a separate database can be set up for each business segment, storing all items or products associated with that segment. The specific configuration can be tailored to the specific needs of the business. Based on this, a nearline computing module can be implemented. This module can include a nearline pre-calculation mechanism, which can be triggered based on acquired user behavior sequence data. This mechanism can retrieve recommendable information from the aforementioned database, or retrieve all items or products from multiple different business segments. Recommendable information can then be determined from the retrieved list of all items or products from the various business segments. The specific methods for determining recommendable information are detailed in the aforementioned sections and will not be repeated here.

[0037] In step A6, based on the user behavior sequence data generated by the user, a generative recommendation model is invoked to perform recommendation scoring on the recommendable information of multiple different services, and a recommendation score is obtained for each recommendable information in multiple different services.

[0038] In implementation, the near-line pre-computation mechanism can invoke a generative recommendation model. User-generated user behavior sequence data and recommendable information from multiple different services can be input into the generative recommendation model. The user behavior sequence data guides the generative recommendation model to score the recommendable information from multiple different services. Through the recommendation score, the recommendable information from multiple different services can be sorted according to its matching degree with the user. The recommendable information with the highest matching degree with the user can be ranked first and given a higher recommendation score. The above sorting results and recommendation scores can be obtained in the above way.

[0039] In step A8, the recommendation score corresponding to each recommendable information in multiple different services and the corresponding recommendable information in multiple different services are stored in the user's information recommendation set.

[0040] In practical applications, there are various ways to process step A6 above. Here is another optional processing method, which can be found in steps A62 to A66 below.

[0041] In step A62, the user behavior sequence data generated by the user is filtered to obtain target user behavior data in which the user behavior sequence data generated by the user contains preset recommendation semantics.

[0042] The preset recommendation semantics can be set according to the actual situation. For example, the user's click / purchase behavior on the financial management page can be used as the preset recommendation semantics, or the user's click / purchase behavior on the women's clothing page can be used as the preset recommendation semantics, etc.

[0043] In implementation, such as Figure 3 As shown, the nearline computing module can pre-set filtering rules to clean and filter user behavior sequence data generated by users, retaining only valid recommendation behavior data (i.e., behavior data corresponding to the preset recommendation semantics) and cleaning and filtering out invalid recommendation behavior data and noise data. Specifically, it can retain valid recommendation behavior data such as user clicks / purchases on the financial management page, and not retain invalid recommendation behavior data and noise data such as user behavior data on the payment page, user behavior data on the video page, or message page. Finally, it can obtain target user behavior data containing the preset recommendation semantics in the user behavior sequence data generated by users.

[0044] In step A64, the user's historical behavior sequence data is obtained, and the historical behavior sequence data is concatenated with the target user behavior data to obtain the concatenated user behavior sequence data.

[0045] In implementation, such as Figure 3 As shown, the nearline calculation module can query the specified database to see if the user has historical behavior sequence data through the nearline pre-calculation mechanism. If it does, the user's historical behavior sequence data can be obtained. Then, the target user behavior data can be concatenated to the end of the historical behavior sequence data, or to the beginning of the historical behavior sequence data, or to any position in the historical behavior sequence data, thereby forming a complete user behavior sequence data and obtaining the concatenated user behavior sequence data.

[0046] In step A68, the generative recommendation model is invoked, and the concatenated user behavior sequence data and recommendable information from multiple different services are input into the generative recommendation model. The concatenated user behavior sequence data guides the generative recommendation model to perform recommendation scoring on the recommendable information from multiple different services, thereby obtaining the recommendation score corresponding to each recommendable information from multiple different services.

[0047] In implementation, such as Figure 3As shown, the near-line pre-computation mechanism can call a generative recommendation model and input spliced ​​user behavior sequence data and recommendable information from multiple different services into the generative recommendation model. The spliced ​​user behavior sequence data guides the generative recommendation model to score the recommendable information from multiple different services. Through the recommendation score, the recommendable information from multiple different services can be sorted according to the degree of matching with the user. The recommendable information with the highest degree of matching with the user can be ranked first and given a higher recommendation score. The above sorting results and recommendation scores can be obtained in the above way.

[0048] In practical applications, the specific processing method of step A62 above can be varied. Here is another optional processing method, which can be found in the following content: encapsulate the user behavior sequence data generated by the user into a data stream, and filter the encapsulated data stream based on the stream processing rules to obtain the target user behavior data in the user behavior sequence data generated by the user that contains the preset recommendation semantics.

[0049] In implementation, to better process the specified data, a stream processing engine module can be pre-configured. After acquiring user-generated user behavior sequence data, the user behavior sequence data can be transmitted to the stream processing engine module, which can then trigger a near-line pre-computation mechanism. The stream processing engine module encapsulates the user-generated user behavior sequence data into a data stream and filters the encapsulated data stream based on stream processing rules to obtain target user behavior data containing preset recommendation semantics within the user-generated user behavior sequence data. The specific processing steps for filtering the data stream described above are detailed above and will not be repeated here.

[0050] Accordingly, there are various ways to process step A64 above. Here is another optional processing method, which can be found in steps A642 to A646 below.

[0051] In step A642, the target user behavior data is packaged into a behavior message and the behavior message is passed to the message queue. The behavior message includes information about the executing entity, the time when the user behavior occurred, information about the user behavior, and information about the target entity corresponding to the user behavior.

[0052] The information of the executing entity can be the current user's information, such as the current user's identifier (e.g., nickname or code) or the identifier of the terminal device currently being used by the user (e.g., IP address or MAC address). User behavior information can include various actions, such as purchasing, redeeming, reporting, or clicking confirmation, which can be set according to the actual situation. The information of the target entity corresponding to the user behavior can be information about a specific business project or product, such as a specific fund product or insurance product, which can be set according to the actual situation.

[0053] In implementation, to ensure real-time data processing and an orderly data processing procedure, target user behavior data can be packaged into behavior messages. These behavior messages can include, for example, "User A - purchased - Fund B at 9:15 AM". The behavior messages can then be passed to a message queue.

[0054] In step A644, the action message to be executed is pulled from the message queue, and the user's historical action sequence data is obtained.

[0055] In implementation, the nearline computing module can be configured with a corresponding server (such as a Java server). This server monitors the message queue and continuously receives behavioral messages from it. These messages can be received via a pull mechanism, where the server actively pulls messages based on its processing capacity. When there is limited target user behavior data and the generative recommendation model's scoring speed is fast, resulting in a user behavior sequence data generation rate lower than the model's throughput, each generated user behavior sequence data can be promptly retrieved and consumed, updating recommendation information and achieving high real-time performance. Conversely, when there is excessive target user behavior data exceeding the generative recommendation model's throughput (e.g., during peak traffic periods in large promotional events, where numerous users generate large amounts of user behavior sequence data), the server aggregates and consumes data according to a preset time window. Within a certain timeframe (e.g., 10 or 30 seconds), it aggregates multiple user behavior sequence data sets and updates recommendation information only once, achieving near real-time performance. The specific processing steps for acquiring historical user behavior sequence data can be found in the aforementioned content and will not be elaborated upon here.

[0056] In step A646, the historical behavior sequence data and the target user behavior data corresponding to the retrieved behavior messages are concatenated to obtain the concatenated user behavior sequence data.

[0057] The specific process of concatenating historical behavior sequence data with the target user behavior data corresponding to the retrieved behavior messages can be found in the aforementioned content and will not be repeated here.

[0058] In practical applications, the behavior messages retrieved in step A644 above include multiple different behavior messages. Therefore, the specific processing method of step A646 above can vary. Here is another optional processing method, which can be found in the processing of steps A6462 and A6464 below.

[0059] In step A6462, the target user behavior data corresponding to multiple different behavior messages are aggregated to obtain aggregated target user behavior data.

[0060] In implementation, as mentioned above, when the target user behavior data is excessive and exceeds the throughput of the generative recommendation model (e.g., during peak traffic periods in large-scale promotional events, a large number of users generate a large amount of user behavior sequence data), behavior messages can be retrieved according to a preset time window. Multiple different behavior messages can be retrieved within a certain time period (e.g., 10 seconds or 30 seconds). The retrieved multiple different behavior messages can be aggregated and consumed, thereby aggregating multiple user behavior sequence data within a certain time period to obtain aggregated target user behavior data.

[0061] In step A6464, the historical behavior sequence data is concatenated with the aggregated target user behavior data to obtain the concatenated user behavior sequence data.

[0062] The specific processing steps for concatenating historical behavior sequence data with aggregated target user behavior data can be found in the aforementioned related content, and will not be repeated here.

[0063] In practical applications, stream processing rules include stream processing rules built based on the Flink stream processing engine.

[0064] The core of the Flink stream processing engine is a distributed data stream engine written in Java and Scala. Flink executes arbitrary streaming data programs in a data-parallel and pipelined manner. Its pipelined runtime system can execute both batch and stream processing programs. Furthermore, the Flink runtime itself supports the execution of iterative algorithms. During program execution, Flink stream processing engines map data streams. Each Flink data stream begins and ends with one or more message queues. Flink can perform any number of transformations on the data streams, which can be orchestrated into a directed acyclic dataflow graph, allowing applications to branch and merge data streams.

[0065] In practical applications, the specific processing method of step A2 above can vary. Here is another optional processing method, which may include the following: receiving the behavior log uploaded by the user's terminal device. The behavior log is a log generated by the terminal device by collecting user behavior sequence data.

[0066] In implementation, a logging module can be set up. This module can trigger the client on the user's terminal device to upload relevant log data. The client on the user's terminal device can detect the user's behavioral data. Once the terminal device captures the user's behavioral data, it can record the captured behavioral data as a log and send the log to the server. The server can determine the user's behavioral sequence data based on the received log. In addition, the logging module can also store the user's behavioral sequence data in a behavioral sequence database to ensure data persistence.

[0067] Based on the above log processing method, the following processing can also be performed: store the behavior log in the log database, and store the user behavior sequence data generated by the user in the behavior log in the behavior sequence database.

[0068] Based on the processing of the behavior sequence library described above, there are various ways to process step A64. Here is another optional processing method, which may include the following: obtaining the user's historical behavior sequence data from the behavior sequence library.

[0069] In addition, the spliced ​​user behavior sequence data can be stored in a behavior sequence library.

[0070] In practical applications, multiple different businesses include those related to financial management.

[0071] Financial management-related businesses can include fund management, conservative wealth management, advanced wealth management, time deposits, and stock trading. Target businesses can be fund management, conservative wealth management, etc., and the specifics can be determined based on actual circumstances.

[0072] This specification provides an information recommendation method. It receives user recommendation requests for a target service, responds to these requests by obtaining a candidate set of recommendable information for the target service, and then queries the recommendation scores of the recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the recommendation request. The generated information recommendation set is a set of recommended information obtained in advance by calling a generative recommendation model to score recommendable information from multiple different services, including the target service, based on user behavior sequence data. Finally, based on the recommendation scores of the recommendable information in the queried candidate set, target recommended information can be selected from the candidate set and sent to the user. This method utilizes a real-time near-line calculation mechanism triggered by user behavior sequence data and a generative recommendation model to generate recommendation scores for recommendable information. Subsequently, the generated recommendation scores can be used to quickly complete the information recommendation, thereby improving the response speed of information recommendation, reducing computational resource consumption, and enabling cross-scenario sharing of calculation results. Furthermore, when applying the generative recommendation model to high-traffic recommendation scenarios, it can balance the trade-off between real-time performance and performance.

[0073] Furthermore, a near-line pre-computation mechanism based on real-time triggering of user behavior sequence data is proposed, along with a generative recommendation model based on this mechanism. This achieves near real-time recommendation updates, reduces update latency, and lowers computational pressure. In addition, a message queue mechanism using a pull method is used to control the rate of generative recommendation information, achieving dynamic matching between recommendation real-time performance and processing capacity, improving service availability, reducing system downtime risk, and maximizing the utilization of computing resources. Moreover, a cross-scenario recommendation information sharing mechanism can be implemented, enabling one-time computation for multiple uses, significantly reducing computational resource consumption. Furthermore, the architecture design of separating computation and services allows for independent expansion and optimization of each system component, improving system reliability and maintainability.

[0074] The following describes in detail an information recommendation method provided by the embodiments of this specification, with specific application scenarios. The target business is fund business, and multiple different businesses include fund business, stable wealth management business, advanced wealth management business, fixed deposit business, stock business, and other wealth management businesses. The recommendable information for multiple different businesses can be information on wealth management products, and the recommendable information for the target business can be information on fund products.

[0075] like Figure 4As shown in the embodiments of this specification, an information recommendation method is provided. The execution subject of this method can be a terminal device or a server, etc. The terminal device can be a mobile terminal device such as a mobile phone or tablet computer, a computer device such as a laptop or desktop computer, or an IoT device (specifically, a smartwatch, in-vehicle device, etc.). The server can be a single server or a server cluster composed of multiple servers. The server can be a backend server for financial services or online shopping services, or a backend server for an application, etc. This embodiment uses a server as the execution subject for detailed description. For the case where the execution subject is a terminal device, please refer to the following section on server-side processing, which will not be repeated here. Based on Figure 4 The architecture diagram shows that the method may specifically include the following steps: Step B02: Receive the behavior log uploaded by the user's terminal device through the log module. The behavior log is generated by the terminal device by collecting user behavior sequence data.

[0076] Step B04: Store the behavior logs in the log database through the log module, and store the user behavior sequence data generated by the user in the behavior logs in the behavior sequence database.

[0077] Step B06: The user behavior sequence data generated by the user is transmitted to the stream processing engine module through the log module. The stream processing engine module encapsulates the user behavior sequence data into a data stream and filters the encapsulated data stream using stream processing rules built based on the Flink stream processing engine to obtain target user behavior data containing preset recommendation semantics in the user behavior sequence data generated by the user.

[0078] Step B08: The target user behavior data is packaged into behavior messages by the stream processing engine module and the behavior messages are sent to the message queue. The behavior messages include user information, the time when the user behavior occurred, information about the user behavior, and information about the financial product.

[0079] Step B10: The nearline computing module is triggered by the stream processing engine module. The nearline computing module pulls the behavior messages to be executed from the message queue, obtains the user's historical behavior sequence data from the behavior sequence library, and queries the information of currently available financial products in multiple different financial services.

[0080] Step B12: The target user behavior data corresponding to the retrieved behavior message is concatenated to the historical behavior sequence data at the end by the nearline computing module to obtain the concatenated user behavior sequence data, and the concatenated user behavior sequence data is stored in the behavior sequence library.

[0081] The retrieved behavior messages may include multiple different behavior messages. In this case, the above processing can be as follows: aggregate the target user behavior data corresponding to the multiple different retrieved behavior messages to obtain aggregated target user behavior data; and append the aggregated target user behavior data to the historical behavior sequence data as the end to obtain the appended user behavior sequence data.

[0082] Step B14: The generative recommendation model is invoked through the nearline computing module. The spliced ​​user behavior sequence data and information on financial products from multiple different financial businesses are input into the generative recommendation model. The spliced ​​user behavior sequence data guides the generative recommendation model to perform recommendation scoring on financial products from multiple different financial businesses, thereby obtaining the recommendation score for each financial product in the multiple different financial businesses.

[0083] Step B16: The near-line calculation module stores the recommendation score corresponding to each financial product in multiple different financial services and the information of the corresponding financial products in multiple different financial services into the user's information recommendation set.

[0084] Step B18: Receive user recommendation requests for fund business.

[0085] Step B20: In response to the aforementioned user recommendation request, recall the fund products included in the fund business and form a candidate set based on the information of the fund products included in the fund business.

[0086] Step B22: From the information recommendation set of the user corresponding to the user recommendation request generated above, query the recommendation score of the fund product in the candidate set.

[0087] Step B24: Based on the recommendation scores of the fund products in the queried candidate set, select the target fund product from the candidate set and send and display the information of the target fund product to the user.

[0088] This specification provides an information recommendation method. It receives user recommendation requests for a target service, responds to these requests by obtaining a candidate set of recommendable information for the target service, and then queries the recommendation scores of the recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the recommendation request. The generated information recommendation set is a set of recommended information obtained in advance by calling a generative recommendation model to score recommendable information from multiple different services, including the target service, based on user behavior sequence data. Finally, based on the recommendation scores of the recommendable information in the queried candidate set, target recommended information can be selected from the candidate set and sent to the user. This method utilizes a real-time near-line calculation mechanism triggered by user behavior sequence data and a generative recommendation model to generate recommendation scores for recommendable information. Subsequently, the generated recommendation scores can be used to quickly complete the information recommendation, thereby improving the response speed of information recommendation, reducing computational resource consumption, and enabling cross-scenario sharing of calculation results. Furthermore, when applying the generative recommendation model to high-traffic recommendation scenarios, it can balance the trade-off between real-time performance and performance.

[0089] Furthermore, a near-line pre-computation mechanism based on real-time triggering of user behavior sequence data is proposed, along with a generative recommendation model based on this mechanism. This achieves near real-time recommendation updates, reduces update latency, and lowers computational pressure. In addition, a message queue mechanism using a pull method is used to control the rate of generative recommendation information, achieving dynamic matching between recommendation real-time performance and processing capacity, improving service availability, reducing system downtime risk, and maximizing the utilization of computing resources. Moreover, a cross-scenario recommendation information sharing mechanism can be implemented, enabling one-time computation for multiple uses, significantly reducing computational resource consumption. Furthermore, the architecture design of separating computation and services allows for independent expansion and optimization of each system component, improving system reliability and maintainability.

[0090] The above describes the information recommendation method provided in the embodiments of this specification. Based on the same idea, the embodiments of this specification also provide an information recommendation device, such as... Figure 5 As shown.

[0091] The information recommendation device includes: a request module 501, a response module 502, a score query module 503, and an information recommendation module 504, wherein: Request module 501 receives user recommendation requests for the target business; The response module 502, in response to the user recommendation request, obtains a candidate set consisting of recommendable information of the target service; The score query module 503 queries the recommended scores of the recommendable information in the candidate set from the information recommendation set for the user corresponding to the user recommendation request, which is generated in advance. The information recommendation set is a set of recommended information obtained in advance based on user behavior sequence data and by calling a generative recommendation model to recommend and score the recommendable information of multiple different services, including the target service. The information recommendation module 504 selects target recommendation information from the candidate set based on the recommendation score corresponding to the recommendable information in the queried candidate set, and sends the target recommendation information to the user.

[0092] In the embodiments described in this specification, the device further includes: The behavior data acquisition module acquires user behavior sequence data generated by the user; The information acquisition module acquires recommendable information from multiple different services, including the target service. The scoring module, based on the user behavior sequence data generated by the user, calls the generative recommendation model to perform recommendation scoring on the recommendable information of multiple different services, and obtains the recommendation score corresponding to each recommendable information in multiple different services; The information set construction module stores the recommendation score corresponding to each recommendable information in multiple different services and the corresponding recommendable information in multiple different services into the user's information recommendation set.

[0093] In the embodiments of this specification, the scoring module includes: The filtering unit filters the user behavior sequence data generated by the user to obtain target user behavior data in the user behavior sequence data that contains preset recommendation semantics. The splicing unit acquires the user's historical behavior sequence data, splices the historical behavior sequence data with the target user behavior data, and obtains spliced ​​user behavior sequence data. The scoring unit calls the generative recommendation model and inputs the spliced ​​user behavior sequence data and recommendable information from multiple different services into the generative recommendation model. The spliced ​​user behavior sequence data guides the generative recommendation model to perform recommendation scoring on the recommendable information from multiple different services, and obtains the recommendation score corresponding to each recommendable information from multiple different services.

[0094] In the embodiments of this specification, the filtering unit encapsulates the user behavior sequence data generated by the user into a data stream, and performs filtering processing on the encapsulated data stream based on stream processing rules to obtain target user behavior data in the user behavior sequence data generated by the user that contains preset recommendation semantics. The splicing unit packages the target user behavior data into behavior messages and transmits the behavior messages to a message queue. The behavior messages include information about the executing entity, the time of occurrence of the user behavior, information about the user behavior, and information about the target entity corresponding to the user behavior. The unit pulls the behavior messages to be executed from the message queue and obtains the user's historical behavior sequence data. The unit then splices the historical behavior sequence data with the target user behavior data corresponding to the pulled behavior messages to obtain spliced ​​user behavior sequence data.

[0095] In this embodiment of the specification, the retrieved behavior messages include multiple different behavior messages. The splicing unit aggregates the target user behavior data corresponding to the multiple different behavior messages to obtain aggregated target user behavior data; and splices the historical behavior sequence data with the aggregated target user behavior data to obtain spliced ​​user behavior sequence data.

[0096] In the embodiments described in this specification, the stream processing rules include stream processing rules built based on the Flink stream processing engine.

[0097] In this embodiment of the specification, the behavior data acquisition module receives behavior logs uploaded by the user's terminal device. The behavior logs are logs generated by the terminal device by collecting user behavior sequence data generated by the user. The device further includes: The first storage module stores the behavior logs in a log database and stores the user behavior sequence data generated by the user in the behavior logs in a behavior sequence database. The splicing unit obtains the user's historical behavior sequence data from the behavior sequence library; The device further includes: The second storage module stores the spliced ​​user behavior sequence data into the behavior sequence library.

[0098] In the embodiments described in this specification, the various businesses include those related to financial management.

[0099] This specification provides an information recommendation device that receives user recommendation requests for a target service, and in response to the user recommendation request, obtains a candidate set of recommendable information for the target service. Then, it queries the recommendation scores corresponding to the recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the user recommendation request. The generated information recommendation set is a set of recommendation information obtained in advance based on user behavior sequence data, by calling a generative recommendation model to score the recommendable information of multiple different services, including the target service. Finally, based on the recommendation scores corresponding to the recommendable information in the queried candidate set, the target recommendation information can be selected from the candidate set and sent to the user. In this way, the near-line calculation mechanism is triggered in real time based on user behavior sequence data, and the recommendation scores corresponding to the recommendable information are generated based on the generative recommendation model. Subsequently, the information recommendation can be completed quickly directly using the generated recommendation scores, thereby improving the response speed of information recommendation, reducing the consumption of computing resources, and realizing the cross-scenario sharing of calculation results. When applying the generative recommendation model to high-traffic recommendation scenarios, it can also balance the contradiction between the real-time performance and the performance of generative recommendation.

[0100] Furthermore, a near-line pre-computation mechanism based on real-time triggering of user behavior sequence data is proposed, along with a generative recommendation model based on this mechanism. This achieves near real-time recommendation updates, reduces update latency, and lowers computational pressure. In addition, a message queue mechanism using a pull method is used to control the rate of generative recommendation information, achieving dynamic matching between recommendation real-time performance and processing capacity, improving service availability, reducing system downtime risk, and maximizing the utilization of computing resources. Moreover, a cross-scenario recommendation information sharing mechanism can be implemented, enabling one-time computation for multiple uses, significantly reducing computational resource consumption. Furthermore, the architecture design of separating computation and services allows for independent expansion and optimization of each system component, improving system reliability and maintainability.

[0101] The above are the information recommendation devices provided in the embodiments of this specification. Based on the same idea, the embodiments of this specification also provide an information recommendation device, such as... Figure 6 As shown.

[0102] The information recommendation device can provide terminal devices or servers, etc., for the above embodiments.

[0103] Information recommendation devices can vary considerably depending on their configuration and performance. They may include one or more processors 601 and memory 602, with memory 602 storing one or more application programs or data. Memory 602 can be temporary or persistent storage. The application programs stored in memory 602 may include one or more modules (not shown), each module including a series of computer-executable instructions for the information recommendation device. Furthermore, processor 601 may be configured to communicate with memory 602 and execute the series of computer-executable instructions stored in memory 602 on the information recommendation device. The information recommendation device may also include one or more power supplies 603, one or more wired or wireless network interfaces 604, one or more input / output interfaces 605, and one or more keyboards 606.

[0104] Specifically, in this embodiment, the information recommendation device includes a memory and one or more programs, wherein one or more programs are stored in the memory, and one or more programs may include one or more modules, and each module may include a series of computer-executable instructions for the information recommendation device, and is configured to be executed by one or more processors. The one or more programs include computer-executable instructions for performing the following: Receive user recommendation requests tailored to the target business; In response to the user recommendation request, obtain a candidate set consisting of recommendable information for the target service; From the pre-generated information recommendation set for the user corresponding to the user recommendation request, query the recommendation score corresponding to the recommendable information in the candidate set. The generated information recommendation set is a set of recommendation information obtained in advance based on user behavior sequence data, by calling a generative recommendation model to recommend and score the recommendable information of multiple different services, including the target service. Based on the recommendation scores corresponding to the recommendable information in the candidate set, target recommendation information is selected from the candidate set and sent to the user.

[0105] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the information recommendation device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0106] This specification provides an information recommendation device that receives user recommendation requests for a target service, responds to the user recommendation request by obtaining a candidate set of recommendable information for the target service, and then queries the recommendation scores corresponding to the recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the user recommendation request. The generated information recommendation set is a set of recommendation information obtained in advance by calling a generative recommendation model to score the recommendable information of multiple different services, including the target service, based on user behavior sequence data. Finally, based on the recommendation scores corresponding to the recommendable information in the queried candidate set, target recommendation information can be selected from the candidate set and sent to the user. In this way, the near-line calculation mechanism is triggered in real time based on user behavior sequence data, and the recommendation scores corresponding to the recommendable information are generated based on the generative recommendation model. Subsequently, the information recommendation can be completed quickly directly using the generated recommendation scores, thereby improving the response speed of information recommendation, reducing the consumption of computing resources, and realizing the cross-scenario sharing of calculation results. When applying the generative recommendation model to high-traffic recommendation scenarios, it can also balance the contradiction between the real-time performance and performance of generative recommendation.

[0107] Furthermore, based on the above Figures 2 to 4 This specification also provides a storage medium for storing computer-executable instruction information in one or more embodiments. In one specific embodiment, the storage medium may be a USB flash drive, optical disc, hard disk, etc. When the computer-executable instruction information stored in the storage medium is executed by a processor, it can realize the following process: Receive user recommendation requests tailored to the target business; In response to the user recommendation request, obtain a candidate set consisting of recommendable information for the target service; From the pre-generated information recommendation set for the user corresponding to the user recommendation request, query the recommendation score corresponding to the recommendable information in the candidate set. The generated information recommendation set is a set of recommendation information obtained in advance based on user behavior sequence data, by calling a generative recommendation model to recommend and score the recommendable information of multiple different services, including the target service. Based on the recommendation scores corresponding to the recommendable information in the candidate set, target recommendation information is selected from the candidate set and sent to the user.

[0108] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the above-described storage medium embodiment is basically similar to the method embodiment, so the description is relatively simple; relevant parts can be referred to the description of the method embodiment.

[0109] This specification provides a storage medium that receives user recommendation requests for a target service, and in response, obtains a candidate set of recommendable information for the target service. Then, it queries the recommendation scores corresponding to the recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the recommendation request. The generated information recommendation set is a set of recommendation information obtained in advance by calling a generative recommendation model to score recommendable information from multiple different services, including the target service, based on user behavior sequence data. Finally, based on the recommendation scores corresponding to the recommendable information in the queried candidate set, target recommendation information can be selected from the candidate set and sent to the user. This real-time near-line calculation mechanism based on user behavior sequence data and the generation of recommendation scores corresponding to recommendable information based on the generative recommendation model allow for rapid information recommendation directly using the generated recommendation scores. This improves the response speed of information recommendation, reduces computational resource consumption, and enables cross-scenario sharing of calculation results. Furthermore, when applying the generative recommendation model to high-traffic recommendation scenarios, it can balance the trade-off between real-time performance and performance.

[0110] Furthermore, based on the above Figures 2 to 4 This specification also provides one or more embodiments of a computer program product, including a computer program, which, when executed by a processor, can perform the following processes: Receive user recommendation requests tailored to the target business; In response to the user recommendation request, obtain a candidate set consisting of recommendable information for the target service; From the pre-generated information recommendation set for the user corresponding to the user recommendation request, query the recommendation score corresponding to the recommendable information in the candidate set. The generated information recommendation set is a set of recommendation information obtained in advance based on user behavior sequence data, by calling a generative recommendation model to recommend and score the recommendable information of multiple different services, including the target service. Based on the recommendation scores corresponding to the recommendable information in the candidate set, target recommendation information is selected from the candidate set and sent to the user.

[0111] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the above-described embodiment of a computer program product is relatively simple in description because it is fundamentally similar to the method embodiment; relevant parts can be referred to the description of the method embodiment.

[0112] This specification provides a computer program product that receives user recommendation requests for a target service, responds to the user recommendation request by obtaining a candidate set of recommendable information for the target service, and then queries the recommendation scores corresponding to the recommendable information in the candidate set from a pre-generated information recommendation set for the user corresponding to the user recommendation request. The generated information recommendation set is a set of recommendation information obtained in advance by calling a generative recommendation model to score the recommendable information of multiple different services, including the target service, based on user behavior sequence data. Finally, based on the recommendation scores corresponding to the recommendable information in the queried candidate set, target recommendation information can be selected from the candidate set and sent to the user. In this way, the near-line calculation mechanism is triggered in real time based on user behavior sequence data, and the recommendation scores corresponding to the recommendable information are generated based on the generative recommendation model. Subsequently, the information recommendation can be completed quickly directly using the generated recommendation scores, thereby improving the response speed of information recommendation, reducing the consumption of computing resources, and realizing the cross-scenario sharing of calculation results. When applying the generative recommendation model to high-traffic recommendation scenarios, it can also balance the contradiction between the real-time performance and the performance of generative recommendation.

[0113] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0114] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0115] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0116] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0117] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0118] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] Embodiments in this specification are described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable parallel device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable parallel device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0120] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable fraud device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0122] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0123] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0124] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0125] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0126] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] One or more embodiments of this specification can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. One or more embodiments of this specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0128] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0129] The above description is merely an embodiment of this specification and is not intended to limit this document. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.

Claims

1. An information recommendation method, comprising: receiving a user recommendation request for a target service; in response to the user recommendation request, obtaining a candidate set of recommendable information of the target service; from a pre-generated information recommendation set corresponding to a user of the user recommendation request, querying a recommendation score corresponding to the recommendable information in the candidate set, the generated information recommendation set being obtained by calling a generative recommendation model to score the recommendable information of a plurality of different services including the target service based on user behavior sequence data; based on the queried recommendation score corresponding to the recommendable information in the candidate set, selecting target recommendation information from the candidate set, and sending the target recommendation information to the user.

2. The method of claim 1, further comprising: obtaining user behavior sequence data generated by the user; obtaining recommendable information of a plurality of different services, the plurality of different services including the target service; based on the user behavior sequence data generated by the user, calling a generative recommendation model to score the recommendable information of a plurality of different services, to obtain a recommendation score corresponding to each recommendable information in the plurality of different services; storing the recommendation score corresponding to each recommendable information in the plurality of different services and the corresponding recommendable information in the information recommendation set of the user.

3. The method of claim 2, wherein the calling of the generative recommendation model to score the recommendable information of a plurality of different services based on the user behavior sequence data generated by the user to obtain a recommendation score corresponding to each recommendable information in the plurality of different services comprises: filtering the user behavior sequence data generated by the user to obtain target user behavior data in the user behavior sequence data generated by the user that has a preset recommendation semantic; obtaining historical behavior sequence data of the user, and splicing the historical behavior sequence data with the target user behavior data to obtain spliced user behavior sequence data; calling a generative recommendation model, and inputting the spliced user behavior sequence data and the recommendable information of a plurality of different services into the generative recommendation model, to guide the generative recommendation model to score the recommendable information of a plurality of different services based on the spliced user behavior sequence data, to obtain a recommendation score corresponding to each recommendable information in the plurality of different services.

4. The method of claim 3, wherein the filtering of the user behavior sequence data generated by the user to obtain target user behavior data in the user behavior sequence data generated by the user that has a preset recommendation semantic comprises: encapsulating the user behavior sequence data generated by the user as a data stream, and filtering the encapsulated data stream based on a stream processing rule to obtain target user behavior data in the user behavior sequence data generated by the user that has a preset recommendation semantic. The obtaining the historical behavior sequence data of the user, the splicing processing of the historical behavior sequence data and the target user behavior data, and the obtaining of spliced user behavior sequence data, comprise: packaging the target user behavior data as behavior messages, and delivering the behavior messages to a message queue, wherein the behavior messages comprise information of an execution subject, a time of occurrence of a user behavior, information of the user behavior, and information of a target subject corresponding to the user behavior; pulling the behavior messages to be executed from the message queue, and obtaining the historical behavior sequence data of the user; splicing processing of the historical behavior sequence data and the target user behavior data corresponding to the pulled behavior messages, to obtain spliced user behavior sequence data.

5. The method of claim 4, wherein the pulled behavior messages comprise a plurality of different behavior messages, and the splicing processing of the historical behavior sequence data and the target user behavior data corresponding to the pulled behavior messages, to obtain spliced user behavior sequence data, comprises: aggregating the target user behavior data corresponding to the pulled plurality of different behavior messages, to obtain aggregated target user behavior data; splicing processing of the historical behavior sequence data and the aggregated target user behavior data, to obtain spliced user behavior sequence data.

6. The method of claim 4, wherein the stream processing rule comprises a stream processing rule constructed based on a Flink stream processing engine.

7. The method of claim 5, wherein the obtaining the user behavior sequence data generated by the user comprises: receiving behavior logs uploaded by a terminal device of the user, wherein the behavior logs are logs generated by the terminal device by collecting the user behavior sequence data generated by the user; The method further comprises: storing the behavior logs to a log database, and storing the user behavior sequence data generated by the user in the behavior logs to a behavior sequence library; The obtaining the historical behavior sequence data of the user comprises: obtaining the historical behavior sequence data of the user from the behavior sequence library; The method further comprises: storing the spliced user behavior sequence data to the behavior sequence library.

8. The method of any one of claims 1-7, wherein the plurality of different services comprise services related to financial management.

9. An information recommendation device, the device comprising: a request module that receives a user recommendation request for a target service; a response module that, in response to the user recommendation request, obtains a candidate set of recommendable information of the target service; a score query module that queries, from a pre-generated information recommendation set for a user corresponding to the user recommendation request, a recommendation score corresponding to the recommendable information in the candidate set, wherein the information recommendation set is generated by pre- calling a generative recommendation model to recommend and score the recommendable information of a plurality of different services including the target service based on user behavior sequence data.

10. An information recommendation method, the method comprising: receiving a user recommendation request for a target service; in response to the user recommendation request, obtaining a candidate set of recommendable information of the target service; from a pre-generated information recommendation set for a user corresponding to the user recommendation request, querying a recommendation score corresponding to the recommendable information in the candidate set, wherein the information recommendation set is generated by pre-calling a generative recommendation model to recommend and score the recommendable information of a plurality of different services including the target service based on user behavior sequence data. The information recommendation module selects target recommended information from the candidate set based on the recommended scores of the recommendable information in the candidate set, and sends the target recommended information to the user.

10. An information recommendation device, the information recommendation device comprising: a processor; and a memory arranged to store computer-executable instructions that, when executed, cause the processor to: receive a user recommendation request for a target service; in response to the user recommendation request, acquire a candidate set of recommendable information of the target service; query the recommended scores of the recommendable information in the candidate set from a pre-generated information recommendation set for a user corresponding to the user recommendation request, the generated information recommendation set being a recommended information set obtained by calling a generative recommendation model to recommend and score the recommendable information of a plurality of different services including the target service based on user behavior sequence data in advance; select target recommended information from the candidate set based on the recommended scores of the recommendable information in the candidate set, and send the target recommended information to the user.

Images