Large model-based interpretable recommendation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides an interpretable recommendation method and device based on a large model, electronic equipment and a storage medium. The method comprises the steps that anchor point information generated when a recommendation engine responds to a user request is acquired; generating a Prompt text based on the anchor point information; obtaining an interpretability problem proposed for the recommendation result; and inputting the Prompt text and the interpretable question into a pre-trained large language model, wherein the large language model outputs text interpretation content for the interpretable question based on the Prompt text. According to the technical scheme, the anchor point information generated by the recommendation engine is automatically converted into the understandable explanation text by utilizing the natural language generation capability of the large language model. According to the method, the limitation of traditional manual analysis on efficiency is broken through, the processing speed of explanation generation is remarkably improved, and the requirements of industrial-grade application for real-time performance and large scale can be better met.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of data processing technology, and in particular to a method, apparatus, electronic device and storage medium for interpretable recommendations based on a large model. Background Technology

[0002] Explainable recommendation technology refers to intelligent methods that simultaneously provide the rationale behind recommendations when generating them. This technology helps users understand why specific items are recommended by revealing the recommendation logic, thereby enhancing system credibility, improving user satisfaction, and increasing recommendation transparency. In current mainstream machine learning and deep learning recommendation systems, the high complexity of the models makes the recommendation process difficult to present intuitively. For example, algorithms such as collaborative filtering, matrix factorization, and deep neural networks, while offering high recommendation accuracy, often exhibit "black box" characteristics in their decision-making mechanisms, making it impossible for users to trace the source of the recommendations. This lack of explainability can easily raise user concerns, especially when the recommended content deviates significantly from user preferences. Currently, industry attempts to generate explanations through manual analysis, which, while ensuring explanation quality, suffers from limitations such as low processing efficiency and difficulty in large-scale application. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method, apparatus, electronic device, and storage medium for interpretable recommendations based on a large model, which can effectively solve the above-mentioned technical problems.

[0004] In a first aspect, embodiments of the present invention provide an interpretable recommendation method based on a large model, which is applied to an interpretable recommendation system based on a large language model. The method includes:

[0005] Obtain anchor information generated by the recommendation engine when responding to user requests; whereby anchor information serves as a feature system for recommendation decisions, providing structured metadata support for the interpretability of recommendation results;

[0006] Generate Prompt text based on anchor point information;

[0007] Obtain interpretability questions regarding the recommendation results;

[0008] The Prompt text and the interpretability question are input into a pre-trained large language model, which outputs a textual explanation of the interpretability question based on the Prompt text.

[0009] In one possible implementation, the anchor information includes at least the following: user identification information, request identification, user profile focus, recall strategy weight, feature weights of each stage of the recommendation process, recommendation score, and ranking factor.

[0010] In one possible implementation, after obtaining the anchor information generated when the recommendation engine responds to a user request, the method further includes:

[0011] Kafka message queues are used as data pipelines to asynchronously transmit anchor information to the Flink task in the Prompt project, a distributed real-time stream processing system based on Flink technology.

[0012] In one possible implementation, generating the Prompt text based on anchor information includes:

[0013] The Flink task extracts user identification information from the anchor information;

[0014] Retrieve historical user behavior data and product feature data corresponding to user identification information from the user feature database; the user feature database stores the correspondence between user identification information and behavioral feature data, and the feature data includes historical user behavior data and product feature data.

[0015] Based on Flink's dynamic splicing technology, anchor information, user historical behavior data, and product feature data are spliced ​​and combined in real time to generate a Prompt text.

[0016] In one possible implementation, after generating the Prompt text based on the anchor information, the method further includes:

[0017] Store the Prompt text in the database with the request identifier as the primary key.

[0018] In one possible implementation, after obtaining interpretability questions raised regarding the recommendation results, the method further includes:

[0019] Obtain the business scenario;

[0020] The system retrieves the large language model corresponding to the business scenario from the large language model library, which stores the correspondence between business scenarios and large language models.

[0021] In one possible implementation, the Prompt text and interpretability question are input into a pre-trained large language model, including:

[0022] Transform interpretability questions and anchor information into chain-like thinking instructions;

[0023] The chained thinking instructions and the prompt text are input into a pre-trained large language model.

[0024] In one possible implementation, the method further includes:

[0025] Send the text explanation, Prompt text, anchor information, and interpretability questions to the front end for display.

[0026] Secondly, embodiments of the present invention provide an interpretable recommendation device based on a large model, which is applied to an interpretable recommendation system based on a large language model. The device includes:

[0027] The first acquisition module is used to acquire anchor information generated by the recommendation engine when responding to user requests; wherein, the anchor information serves as a feature system for recommendation decisions, providing structured metadata support for the interpretability of recommendation results;

[0028] The generation module is used to generate Prompt text based on anchor point information;

[0029] The second acquisition module is used to acquire interpretability questions raised regarding the recommendation results;

[0030] The explanation module is used to input the Prompt text and the interpretability question into a pre-trained large language model. The large language model outputs a textual explanation of the interpretability question based on the Prompt text.

[0031] Thirdly, embodiments of the present invention provide an electronic device, comprising: a processor and a memory, wherein the processor is configured to execute a program stored in the memory based on large model-interpretable recommendations, to implement the above-described large model-based interpretable recommendation method.

[0032] Fourthly, embodiments of the present invention provide a storage medium storing one or more programs, which can be executed by one or more processors to implement the above-described large model-based interpretable recommendation method.

[0033] This invention provides a method, apparatus, electronic device, and storage medium for interpretable recommendation based on a large model. The method includes: acquiring anchor information generated by the recommendation engine in response to a user request; wherein the anchor information serves as a feature system for recommendation decisions, providing structured metadata support for the interpretability of the recommendation results; generating Prompt text based on the anchor information; acquiring interpretability questions raised regarding the recommendation results; and inputting the Prompt text and interpretability questions into a pre-trained large language model, which outputs textual explanations of the interpretability questions based on the Prompt text. This technical solution utilizes the natural language generation capabilities of the large language model to automatically convert the anchor information generated by the recommendation engine into understandable explanatory text. This method overcomes the efficiency limitations of traditional manual analysis, significantly improving the processing speed of explanation generation and meeting the real-time and scalability requirements of industrial applications. Through an automated explanation generation process, the system can provide personalized explanations for each recommendation result within a millisecond response time.

[0034] Furthermore, structured anchor information constructs a traceable chain of evidence for recommendation decisions by fully recording multidimensional features throughout the recommendation decision-making process. This structured feature recording mechanism effectively solves the "black box" problem caused by massive parameter interactions and feature crossovers in deep learning recommendation engines, making the originally difficult-to-explain complex model reasoning process transparent and verifiable. Based on end-to-end anchor information, the system can accurately identify the core factors affecting recommendation results and provide a reliable attribution explanation for each recommendation decision. Attached Figure Description

[0035] Figure 1 A flowchart illustrating an embodiment of an interpretable recommendation method based on a large model, provided for an embodiment of the present invention;

[0036] Figure 2 A flowchart illustrating another embodiment of the interpretable recommendation method based on a large model provided by the present invention;

[0037] Figure 3 This is a schematic diagram of an interpretable real-time processing pipeline architecture provided in an embodiment of the present invention;

[0038] Figure 4 A block diagram illustrating an embodiment of a large model-based interpretable recommendation device provided by this invention;

[0039] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0042] This invention provides an interpretable recommendation method based on a large model. This method is applied to interpretable recommendation systems based on large language models. See [link to relevant documentation]. Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of an interpretable recommendation method based on a large model, provided for an embodiment of the present invention. Figure 1 The process shown may include the following steps:

[0043] Step 101: Obtain the anchor information generated by the recommendation engine when responding to a user request;

[0044] Anchor information, serving as a feature system for recommendation decisions, provides structured metadata support for the interpretability of recommendation results. This anchor information can be understood as a set of structured features recorded by the recommendation engine when generating recommendation results. Anchor information includes at least key decision-making criteria such as user identification information, user profile focus, recall strategy weights, feature weights of each stage of the recommendation process, recommendation scores, and ranking factors. This data is stored in a standardized format, forming a traceable chain of recommendation evidence, providing original feature support for subsequent interpretation and generation.

[0045] In this embodiment, taking the video domain as an example, the user identification information included in the anchor information can be unique information indicating the user, such as user ID (Identification) or username; the request identifier can be understood as a unique event ID dynamically generated for each user request, used to identify different request instances; the core dimensions of the user profile focus can focus on key indicator data that can effectively characterize user behavior features, such as video content preferences and platform usage frequency; the recall strategy weight is a key parameter in the recommendation engine used to quantify the contribution of different recall strategies. In this embodiment, the recall strategy weight includes the weights corresponding to each recall strategy in the recommendation engine; the feature weights of each stage of the recommendation process reflect the quantitative evaluation of user preferences, making the importance of features verifiable. Since the recommendation process mainly includes stages such as recall, coarse ranking, fine ranking, re-ranking, and rules, the feature weights of each stage of the recommendation process include the feature weights corresponding to the output features of each stage; the recommendation score and ranking factor are core indicators used to quantify user preferences and determine the order of recommended video display.

[0046] In actual business scenarios, the anchor information varies across different fields (such as e-commerce, content platforms, social networks, etc.), and no restrictions are imposed here.

[0047] Step 102: Generate Prompt text based on anchor information;

[0048] Prompt text constructed using structured anchor information can make implicit feature interactions explicit, providing clear reasoning basis for large language models and enabling them to understand task requirements more accurately. This transformation is equivalent to establishing a "semantic coordinate system" for large language models, reducing output bias caused by input ambiguity.

[0049] Step 103: Obtain interpretability questions for the recommendation results;

[0050] Users can ask specific, explainable questions in real time (such as "Why are these videos recommended to me?" or "Why are videos with ID 872455629331401 not recommended to me?") so that the larger model can output corresponding text explanations based on the specific questions.

[0051] These types of questions often involve personalized needs, and the question type directly affects the response strategy of the large language model: for pre-set standardized questions, the large language model usually adopts a structured answer; while when faced with real-time and personalized questions, the large language model will dynamically adjust the explanation dimensions. In the algorithm recommendation logic, the large language model can also automatically control the level of detail of the text explanation content according to the complexity of the question.

[0052] Step 104: Input the Prompt text and the interpretability question into the pre-trained large language model. The large language model outputs a textual explanation of the interpretability question based on the Prompt text.

[0053] Large language models output accurate text explanations based on the following mechanisms:

[0054] 1. Feature association analysis: Identify key influencing factors in the Prompt text (such as recall strategy weight of 10%, recommendation score of 0.0308, and ranking factor of 0);

[0055] 2. Logical chain construction: Generate textual explanations that conform to human cognition (e.g., video ID 872455629331401 may have been removed during the rule filtering stage if it was not ultimately exposed. Checking its recommendation score (0.0308) and recall strategy weight (10%), it was found that its recommendation score was low and its ranking factor was 0, which resulted in it not meeting the exposure conditions.)

[0056] 3. Personalized Expression: Adjust the explanation style according to user profile. For example, use popular internet slang for young users, switch to terminology for professionals, and automatically enlarge the font size and use colloquial expressions when dealing with the elderly. This adaptive capability makes the interactive experience more human.

[0057] By leveraging the natural language generation capabilities of large language models, anchor information generated by the recommendation engine is automatically transformed into understandable explanatory text. This method overcomes the efficiency limitations of traditional manual analysis, significantly improving the processing speed of explanation generation and meeting the real-time and scalable requirements of industrial applications. Through an automated explanation generation process, the system can provide personalized explanations for each recommendation result within millisecond response times.

[0058] Furthermore, structured anchor information constructs a traceable chain of evidence for recommendation decisions by fully recording multidimensional features throughout the recommendation decision-making process. This structured feature recording mechanism effectively solves the "black box" problem caused by massive parameter interactions and feature crossovers in deep learning recommendation engines, making the originally difficult-to-explain complex model reasoning process transparent and verifiable. Based on end-to-end anchor information, the system can accurately identify the core factors affecting recommendation results and provide a reliable attribution explanation for each recommendation decision.

[0059] exist Figure 1 Based on this, see Figure 2 , Figure 2 A flowchart illustrating another embodiment of the large model-based interpretable recommendation method provided by this invention. Figure 2 The process shown may include the following steps:

[0060] Step 201: Obtain anchor information generated by the recommendation engine when responding to user requests; wherein, the anchor information serves as a feature system for recommendation decisions, providing structured metadata support for the interpretability of recommendation results;

[0061] Step 202: Using Kafka message queue as a data pipeline, the anchor information is transmitted asynchronously to the Flink task of the distributed real-time stream processing Prompt project implemented based on Flink technology.

[0062] The recommendation engine asynchronously transmits anchor information via message queues such as Kafka. This design offers two advantages: firstly, it supports a high throughput of millions of messages per second (typical latency less than 50 milliseconds), ensuring real-time performance in high-concurrency scenarios; secondly, it decouples the recommendation engine from Flink tasks, enabling anchor information to be quickly and reliably transmitted to downstream stream processing, providing a data foundation for subsequent real-time computation tasks such as feature concatenation and prompt generation. This loosely coupled architecture improves system scalability while ensuring the stability of the data processing pipeline.

[0063] Step 203: The Flink task extracts user identification information from the anchor information;

[0064] The Flink task parses the anchor information to extract user identification information, which facilitates the accurate retrieval of subsequent user historical behavior data and product feature data.

[0065] Step 204: Obtain user historical behavior data and product feature data corresponding to user identification information from the user feature database;

[0066] The user feature database stores the correspondence between user identification information and behavioral feature data. The feature data includes user historical behavior data and product feature data. Taking the video domain as an example, the user feature database is illustrated in Table 1:

[0067] Table 1

[0068]

[0069]

[0070] It should be noted that the above-mentioned correspondence between user identification information and behavioral feature data is only an example. The specific correspondence between user identification information and behavioral feature data can be set according to actual needs, and is not limited here.

[0071] Step 205: Based on Flink dynamic splicing technology, anchor point information, user historical behavior data and product feature data are spliced ​​and combined in real time to generate Prompt text;

[0072] Using Flink dynamic splicing technology, anchor information, user historical behavior data, and product feature data are combined into a structured Prompt. For example: "User 1's viewing record shows that he watched the suspense drama 'The Three-Body Problem' for 2 hours."

[0073] When converting anchor information into Prompt text that a large language model can understand, the following aspects need to be considered:

[0074] 1. Feature structuring: Transforming discrete feature values ​​into natural language descriptions;

[0075] 2. Contextual Supplement: Add auxiliary explanations for recommended scenarios;

[0076] 3. Explicit instructions: Define the requirements for the model output format.

[0077] Step 206: Store the Prompt text in the database with the request identifier as the primary key;

[0078] In practical applications, the Prompt text can be bound to a unique request identifier and stored in the database as a composite primary key. At the same time, an in-memory index (such as Redis or a B+ tree structure) can be built in real time using a streaming computing framework (such as Flink / Spark Streaming), ultimately enabling millisecond-level association queries between the Prompt text and real-time requests.

[0079] Step 207: Obtain interpretability questions regarding the recommendation results;

[0080] Step 208, Obtain the business scenario;

[0081] In practical business implementation, for differentiated scenarios such as e-commerce, content platforms, social networks, and video, the technical implementation paths of each business form differ significantly, necessitating the adoption of adaptable large language model solutions. Specifically, to ensure a high degree of alignment between the model's output content and business requirements, this solution is designed to dynamically invoke the corresponding vendor's base large model based on scenario characteristics through an intelligent routing mechanism, thereby achieving precise text generation services.

[0082] Step 209: Call the large language model corresponding to the business scenario from the large language model library;

[0083] The large language model library stores the correspondence between business scenarios and large language models; for ease of understanding, it is shown in Table 2:

[0084] Table 2

[0085]

[0086] It should be noted that the above-listed business scenarios and their correspondence with the large language model are only examples. The specific correspondence between user identification information and behavioral feature data can be set according to actual needs, and is not limited here.

[0087] Step 210: Transform the interpretability question and anchor information into chain-like thinking instructions;

[0088] The purpose of transforming interpretability issues and anchor information into chain-like thinking instructions that large language models can understand is to improve the reasoning ability and interpretability of large language models in complex tasks.

[0089] Step 211: Input the chained thinking instruction and the Prompt text into the pre-trained large language model. The large language model outputs textual explanations of the interpretability issues based on the Prompt text.

[0090] Step 212: Send the text explanation, Prompt text, anchor information, and interpretability questions to the front end for display.

[0091] The text explanation is returned along with the original data (Prompt text, anchor information, and interpretability questions). The front-end can flexibly display structured information (such as feature weights) and natural language descriptions, thereby improving user transparency. Developers can choose between full data binding or a simplified return mode (returning the text explanation to the front-end) based on specific needs.

[0092] It is important to note that the text explanations output by the large language model can be filtered (sensitive information is removed) before being returned to the front end for display. This mechanism ensures transparency of explanations while meeting data compliance requirements.

[0093] In practical applications, users can also rate their satisfaction with the text explanations and feed those ratings back to the recommendation engine to optimize the Prompt text, enabling the large language model to output more accurate text explanations.

[0094] To more clearly demonstrate the implementation logic of the large model-based interpretable recommendation method provided in this embodiment, Figure 3 A visual design for an interpretable real-time processing pipeline is provided. Implementation of this architecture requires the pre-deployment of the infrastructure components listed in Table 3.

[0095]

[0096]

[0097] It should be noted that the infrastructure components listed above are only examples, and specific infrastructure components can be set according to actual needs, without limitation here.

[0098] This invention provides an interpretable recommendation device based on a large model. This device is applied to an interpretable recommendation system based on a large language model. See [link to related documentation]. Figure 4 This is a block diagram illustrating an embodiment of a large-model-based interpretable recommendation device provided by an embodiment of the present invention. Figure 4 As shown, the device includes:

[0099] The first acquisition module 401 is used to acquire anchor information generated by the recommendation engine when responding to user requests; wherein, the anchor information serves as a feature system for recommendation decisions, providing structured metadata support for the interpretability of recommendation results;

[0100] Generation module 402 is used to generate Prompt text based on anchor point information;

[0101] The second acquisition module 403 is used to acquire interpretability questions raised regarding the recommendation results;

[0102] The explanation module 404 is used to input the Prompt text and the interpretability question into a pre-trained large language model. The large language model outputs a text explanation of the interpretability question based on the Prompt text.

[0103] This technical solution leverages the natural language generation capabilities of large language models to automatically transform anchor information generated by the recommendation engine into understandable explanatory text. This method overcomes the efficiency limitations of traditional manual analysis, significantly improving the processing speed of explanation generation and meeting the real-time and scalable requirements of industrial applications. Through an automated explanation generation process, the system can provide personalized explanations for each recommendation result within millisecond response times.

[0104] Furthermore, structured anchor information constructs a traceable chain of evidence for recommendation decisions by fully recording multidimensional features throughout the recommendation decision-making process. This structured feature recording mechanism effectively solves the "black box" problem caused by massive parameter interactions and feature crossovers in deep learning recommendation engines, making the originally difficult-to-explain complex model reasoning process transparent and verifiable. Based on end-to-end anchor information, the system can accurately identify the core factors affecting recommendation results and provide a reliable attribution explanation for each recommendation decision.

[0105] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Figure 5The illustrated electronic device 1200 includes at least one processor 1201, a memory 1202, at least one network interface 1204, and other user interfaces 1203. The various components in the electronic device 1200 are coupled together via a bus system 1205. It is understood that the bus system 1205 is used to implement communication between these components. In addition to a data bus, the bus system 1205 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 5 The general labeled all buses as Bus System 1205.

[0106] The user interface 1203 may include a display, keyboard, or clicking device (e.g., mouse, trackball, touchpad, or touchscreen).

[0107] It is understood that the memory 1202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memory 1202 described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0108] In some implementations, memory 1202 stores elements, executable units or data structures, or subsets thereof, or extended sets thereof: operating system 12021 and application program 12022.

[0109] The operating system 12021 includes various system programs, such as the framework layer, core library layer, and driver layer, used to implement various basic business functions and handle hardware-based tasks. The application program 12022 includes various applications, such as a media player and a browser, used to implement various application functions. The program implementing the method of this embodiment can be included in the application program 12022.

[0110] In this embodiment of the invention, by calling the program or instructions stored in the memory 1202, specifically the program or instructions stored in the application program 12022, the processor 1201 executes the method steps provided in each method embodiment, including, for example:

[0111] Obtain anchor information generated by the recommendation engine when responding to user requests; whereby anchor information serves as a feature system for recommendation decisions, providing structured metadata support for the interpretability of recommendation results;

[0112] Generate Prompt text based on anchor point information;

[0113] Obtain interpretability questions regarding the recommendation results;

[0114] The Prompt text and the interpretability question are input into a pre-trained large language model, which outputs a textual explanation of the interpretability question based on the Prompt text.

[0115] In one possible implementation, the anchor information includes at least the following: user identification information, request identification, user profile focus, recall strategy weight, feature weights of each stage of the recommendation process, recommendation score, and ranking factor.

[0116] In one possible implementation, after obtaining the anchor information generated when the recommendation engine responds to a user request, the method further includes:

[0117] Kafka message queues are used as data pipelines to asynchronously transmit anchor information to the Flink task in the Prompt project, a distributed real-time stream processing system based on Flink technology.

[0118] In one possible implementation, generating the Prompt text based on anchor information includes:

[0119] The Flink task extracts user identification information from the anchor information;

[0120] Retrieve historical user behavior data and product feature data corresponding to user identification information from the user feature database; the user feature database stores the correspondence between user identification information and behavioral feature data, and the feature data includes historical user behavior data and product feature data.

[0121] Based on Flink's dynamic splicing technology, anchor information, user historical behavior data, and product feature data are spliced ​​and combined in real time to generate a Prompt text.

[0122] In one possible implementation, after generating the Prompt text based on the anchor information, the method further includes:

[0123] Store the Prompt text in the database with the request identifier as the primary key.

[0124] In one possible implementation, after obtaining interpretability questions raised regarding the recommendation results, the method further includes:

[0125] Obtain the business scenario;

[0126] The system retrieves the large language model corresponding to the business scenario from the large language model library, which stores the correspondence between business scenarios and large language models.

[0127] In one possible implementation, the Prompt text and interpretability question are input into a pre-trained large language model, including:

[0128] Transform interpretability questions and anchor information into chain-like thinking instructions;

[0129] The chained thinking instructions and the prompt text are input into a pre-trained large language model.

[0130] In one possible implementation, the method further includes:

[0131] Send the text explanation, Prompt text, anchor information, and interpretability questions to the front end for display.

[0132] The methods disclosed in the above embodiments of the present invention can be applied to processor 1201, or implemented by processor 1201. Processor 1201 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 1201 or by instructions in the form of software. The processor 1201 may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software units in the decoding processor. The software units may be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 1202. Processor 1201 reads the information in memory 1202 and completes the steps of the above method in conjunction with its hardware.

[0133] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0134] For software implementation, the techniques described herein can be implemented by units that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0135] The electronic device provided in this embodiment may be as follows: Figure 5 The electronic device shown can perform the following: Figure 1-2 Based on a large model, all steps of the recommendation method can be explained, thereby achieving... Figure 1-2 The technical performance of the large-model-based interpretable recommendation method is shown below. Please refer to [link / reference] for details. Figure 1-2 The relevant descriptions are presented concisely and will not be elaborated upon here.

[0136] This invention also provides a storage medium (computer-readable storage medium). This storage medium stores one or more programs. The storage medium may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as read-only memory, flash memory, hard disk, or solid-state drive; the memory may also include combinations of the above types of memory.

[0137] When one or more programs in the storage medium can be executed by one or more processors, the above-mentioned large model-based interpretable recommendation method can be implemented.

[0138] The processor is used to execute a program stored in memory based on large model-based interpretable recommendations to implement the following steps of the large model-based interpretable recommendation method:

[0139] Obtain anchor information generated by the recommendation engine when responding to user requests; whereby anchor information serves as a feature system for recommendation decisions, providing structured metadata support for the interpretability of recommendation results;

[0140] Generate Prompt text based on anchor point information;

[0141] Obtain interpretability questions regarding the recommendation results;

[0142] The Prompt text and the interpretability question are input into a pre-trained large language model, which outputs a textual explanation of the interpretability question based on the Prompt text.

[0143] In one possible implementation, the anchor information includes at least the following: user identification information, request identification, user profile focus, recall strategy weight, feature weights of each stage of the recommendation process, recommendation score, and ranking factor.

[0144] In one possible implementation, after obtaining the anchor information generated when the recommendation engine responds to a user request, the method further includes:

[0145] Kafka message queues are used as data pipelines to asynchronously transmit anchor information to the Flink task in the Prompt project, a distributed real-time stream processing system based on Flink technology.

[0146] In one possible implementation, generating the Prompt text based on anchor information includes:

[0147] The Flink task extracts user identification information from the anchor information;

[0148] Retrieve historical user behavior data and product feature data corresponding to user identification information from the user feature database; the user feature database stores the correspondence between user identification information and behavioral feature data, and the feature data includes historical user behavior data and product feature data.

[0149] Based on Flink's dynamic splicing technology, anchor information, user historical behavior data, and product feature data are spliced ​​and combined in real time to generate a Prompt text.

[0150] In one possible implementation, after generating the Prompt text based on the anchor information, the method further includes:

[0151] Store the Prompt text in the database with the request identifier as the primary key.

[0152] In one possible implementation, after obtaining interpretability questions raised regarding the recommendation results, the method further includes:

[0153] Obtain the business scenario;

[0154] The system retrieves the large language model corresponding to the business scenario from the large language model library, which stores the correspondence between business scenarios and large language models.

[0155] In one possible implementation, the Prompt text and interpretability question are input into a pre-trained large language model, including:

[0156] Transform interpretability questions and anchor information into chain-like thinking instructions;

[0157] The chained thinking instructions and the prompt text are input into a pre-trained large language model.

[0158] In one possible implementation, the method further includes:

[0159] Send the text explanation, Prompt text, anchor information, and interpretability questions to the front end for display.

[0160] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0161] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0162] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A large-model-based interpretable recommendation method, characterized in that, The method is applied to an interpretable recommendation system based on a large language model, and the method includes: Obtain anchor information generated by the recommendation engine when responding to user requests; wherein, the anchor information serves as a feature system for recommendation decisions, providing structured metadata support for the interpretability of recommendation results; Generate Prompt text based on the anchor point information; Obtain interpretability questions raised regarding the recommendation results; The Prompt text and the interpretability question are input into a pre-trained large language model, which outputs a textual explanation of the interpretability question based on the Prompt text.

2. The method according to claim 1, characterized in that, The anchor information includes at least the following: user identification information, request identification, user profile focus, recall strategy weight, feature weights of each stage of the recommendation process, recommendation score, and ranking factor.

3. The method according to claim 2, characterized in that, After obtaining the anchor information generated when the recommendation engine responds to a user request, the method further includes: Kafka message queues are used as data pipelines to asynchronously transmit the anchor information to the Flink task in the Prompt project, a distributed real-time stream processing project based on Flink technology.

4. The method according to claim 3, characterized in that, The generation of Prompt text based on the anchor information includes: The Flink task extracts user identification information from the anchor point information; Retrieve user historical behavior data and product feature data corresponding to the user identification information from the user feature database; wherein, the user feature database stores the correspondence between user identification information and behavior feature data, and the feature data includes user historical behavior data and product feature data; Based on Flink's dynamic splicing technology, the anchor point information, the user's historical behavior data, and the product feature data are spliced ​​and combined in real time to generate a Prompt text.

5. The method according to claim 2, characterized in that, After generating the Prompt text based on the anchor information, the method further includes: The Prompt text is stored in the database using the request identifier as the primary key.

6. The method according to claim 1, characterized in that, After obtaining the interpretability questions raised regarding the recommendation results, the method further includes: Obtain the business scenario; The large language model corresponding to the business scenario is called from the large language model library; wherein the large language model library stores the correspondence between business scenarios and large language models.

7. The method according to claim 1, characterized in that, The step of inputting the Prompt text and the interpretability question into a pre-trained large language model includes: Transform the interpretability issues and anchor information into chain-like thinking instructions; The chained thinking instructions and the Prompt text are input into a pre-trained large language model.

8. The method according to claim 1, characterized in that, The method further includes: The text explanation, the Prompt text, the anchor information, and the interpretability question are sent to the front end for display.

9. A large-model-based interpretable recommendation device, characterized in that, The device is applied to an interpretable recommendation system based on a large language model, and the device includes: The first acquisition module is used to acquire anchor information generated by the recommendation engine when responding to user requests; wherein, the anchor information serves as a feature system for recommendation decisions, providing structured metadata support for the interpretability of recommendation results; The generation module is used to generate Prompt text based on the anchor information; The second acquisition module is used to acquire interpretability questions raised in response to the recommendation results; An explanation module is used to input the Prompt text and the interpretability question into a pre-trained large language model, which outputs a textual explanation of the interpretability question based on the Prompt text.

10. An electronic device, characterized in that, include: A processor and a memory, the processor being configured to execute a program stored in the memory based on large model-interpretable recommendations to implement the large model-based interpretable recommendation method according to any one of claims 1 to 8.

11. A storage medium, characterized in that, The storage medium stores one or more programs, which can be executed by one or more processors to implement the large model-based interpretable recommendation method as described in any one of claims 1 to 8.

Images