Data product intelligent recommendation method, system, device and medium

Abstract

The present application relates to a data product intelligent recommendation method, system, device and medium, and relates to the fields of artificial intelligence and big data technology. The method comprises: obtaining a structured feature vector; constructing a dynamic preference model, inputting the structured feature vector, and outputting a preference score; constructing a filtering recommendation model based on distributed architecture deployment, wherein the recommendation module takes the preference score, the current training task index parameter and the recommended data product set as the state space, calculates through a reward function, and outputs a first recommendation score; the collaborative filtering module performs collaborative filtering calculation according to the preference score, and outputs a second recommendation score; the first recommendation score and the second recommendation score are weighted and fused to generate a data product recommendation list; and the related parameters of the dynamic preference model and the filtering recommendation model are dynamically updated according to the feedback information of the decision maker on the data product recommendation list. The present application can accurately capture the dynamic preference of the decision maker and has efficient parallel processing capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and big data technology, in particular to a data product intelligent recommendation method, system, device and medium. BACKGROUND

[0002] In the process of emergency training management, decision makers need to process massive training data, resource information and task indicators to formulate scientific and reasonable training schemes and decisions. However, the existing information system in the field of training management still has many problems to be solved in the aspect of data product recommendation.

[0003] Current data product recommendation generally lacks pertinence, and relies on static rules or simple collaborative filtering algorithms, which is difficult to accurately match the personalized needs of different decision makers in various emergency training scenarios, and cannot adapt to the real-time adjustment of training task dynamics and decision maker preferences. At the same time, the existing recommendation technology fails to effectively integrate the professional knowledge graph in the field of emergency training, resulting in that the recommended results often deviate from the professional logic of emergency training, and a large number of invalid recommendations are generated, increasing the cost of decision makers to screen effective data products. In addition, in the face of the demand for multi-user and multi-scene parallel promotion in large-scale emergency training management scenarios, the existing system lacks an efficient parallel recommendation service architecture, and it is difficult to cope with concurrent recommendation requests, resulting in low recommendation efficiency and failing to meet the high requirements of emergency training management on decision-making timeliness. The existence of these problems makes it difficult to fully utilize data resources in the process of emergency training management, and limits the decision-making efficiency. SUMMARY

[0004] Therefore, it is necessary to provide a data product intelligent recommendation method, system, device and medium that can accurately capture the dynamic preferences of decision makers, deeply integrate field knowledge and have efficient parallel processing capability, in order to solve the above technical problems.

[0005] A data product intelligent recommendation method, the method comprising:

[0006] Obtaining the behavior data, personalized label, emergency field knowledge graph and current training task indicator parameters of the decision maker in the emergency training management scenario and preprocessing to obtain a structured feature vector;

[0007] Building a dynamic preference model, inputting the structured feature vector into the dynamic preference model for processing, and outputting the preference scores of various data products;

[0008] Building a filtering recommendation model based on distributed architecture deployment, the filtering recommendation model comprising a recommendation module and a collaborative filtering module;

[0009] The recommendation module takes the preference score, the current training task indicator parameter and the recommended data product set as a state space, calculates a first recommendation score through a reward function containing behavior feedback and task matching degree;

[0010] The collaborative filtering module performs collaborative filtering calculation according to the preference score of the data product, and outputs a second recommendation score;

[0011] The first recommendation score and the second recommendation score are weighted and fused to output a fused recommendation score; and a data product recommendation list is generated according to the fused recommendation score;

[0012] According to the feedback information of the decision maker on the data product recommendation list, the related parameters of the dynamic preference model and the filtering recommendation model are dynamically updated, and parallel recommendation services are performed based on a distributed architecture.

[0013] In another aspect, a data product intelligent recommendation system is also provided, comprising:

[0014] A data preprocessing module is configured to obtain and preprocess the behavior data of the decision maker, the personalized label, the emergency field knowledge graph and the current training task indicator parameter in the emergency training management scene to obtain a structured feature vector;

[0015] A preference score calculation module is configured to construct a dynamic preference model, input the structured feature vector into the dynamic preference model for processing, and output the preference score of each type of data product;

[0016] A filtering recommendation model construction module is configured to construct a filtering recommendation model deployed based on a distributed architecture, wherein the filtering recommendation model comprises a recommendation module and a collaborative filtering module;

[0017] A first recommendation score calculation module is configured to, in the recommendation module, take the preference score, the current training task indicator parameter and the recommended data product set as a state space, calculate a first recommendation score through a reward function containing behavior feedback and task matching degree, and output the first recommendation score;

[0018] A second recommendation score calculation module is configured to, in the collaborative filtering module, perform collaborative filtering calculation according to the preference score of the data product, and output a second recommendation score;

[0019] A data product recommendation list generation module is configured to weight and fuse the first recommendation score and the second recommendation score to output a fused recommendation score; and generate a data product recommendation list according to the fused recommendation score;

[0020] A feedback updating module is configured to dynamically update parameters of the dynamic preference model and the filtering recommendation model according to feedback information of the decision maker on the data product recommendation list, and perform parallel recommendation services based on the distributed architecture.

[0021] In another aspect, a computer device is also provided, which includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the data product intelligent recommendation method when executing the computer program.

[0022] In another aspect, a computer readable storage medium is also provided, which stores a computer program, and the processor implements the steps of the data product intelligent recommendation method when executing the computer program.

[0023] Compared with the prior art, the data product intelligent recommendation method, system, device and medium provided by the present application have the following beneficial effects:

[0024] 1. The dynamic preference model is constructed by fusing the emergency field knowledge graph, the decision maker behavior data and the training task indicators, and the dual computing logic of reinforcement learning and collaborative filtering is combined to make the recommendation results not only fit the personalized preferences of the decision maker, but also meet the professional scene requirements of the emergency training, greatly reducing the proportion of invalid recommendations, accurately capturing the dynamic preferences of the decision maker, and significantly improving the recommendation accuracy.

[0025] 2. Based on the feedback information of the decision maker on the recommendation list, the related parameters of the dynamic preference model and the filtering recommendation model are updated in real time, which can quickly respond to changes in training tasks and adjustments of decision maker preferences, and has stronger dynamic adaptability.

[0026] 3. Through automatic data preprocessing, preference modeling and accurate recommendation, the decision maker is replaced to complete the tedious data screening and scheme comparison process, the time cost of finding effective data products is reduced, and the decision efficiency is greatly improved.

[0027] 4. The filtering recommendation model is deployed based on the distributed architecture, supports parallel recommendation services for multiple users and multiple scenes, can efficiently handle concurrent recommendation requests in large-scale emergency training management, and meets the needs of simultaneous promotion of multiple training tasks in multiple places.

[0028] 5. Through precise preference matching and professional logic fusion, the problem of scattered data resource islands is broken, resource waste caused by data product mismatch is avoided, the utilization rate of emergency training related data resources is improved, and the management cost is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required in the embodiments will be briefly introduced. It should be understood that the following drawings only show some embodiments of the present application, and those skilled in the art can obtain other related drawings according to these drawings without creative labor.

[0030] Figure 1 A flowchart of a data product intelligent recommendation method in an embodiment;

[0031] Figure 2 A structural block diagram of a data product intelligent recommendation system in an embodiment;

[0032] Figure 3 An internal structure diagram of a computer device in an embodiment.

[0033] The purposes, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0035] It should be noted that the description of "first", "second" and the like in the present application is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise specifically limited.

[0036] It can be understood that the technical solutions of various embodiments of the present application can be combined with each other, but it must be based on the fact that a person skilled in the art can realize it, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, and is not within the scope of protection required by the present application.

[0037] The embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application.

[0038] Embodiment 1

[0039] As shown in the data product intelligent recommendation method provided in the present embodiment, the following steps are included: Figure 1

[0040] ​Step 201, obtaining the behavior data of the decision maker in the emergency training management scene, the personalized label, the emergency field knowledge graph and the current training task index parameter and preprocessing to obtain a structured feature vector.

[0041] Step 202, constructing a dynamic preference model, inputting the structured feature vector into the dynamic preference model for processing, and outputting the preference score of each type of data product.

[0042] Step 203, constructing a filtering recommendation model based on a distributed architecture deployment, the filtering recommendation model including a recommendation module and a collaborative filtering module.

[0043] Step 204, the recommendation module takes the preference score, the current training task index parameter and the recommended data product set as a state space, and calculates through a reward function including behavior feedback and task matching degree to output a first recommendation score.

[0044] Step 205, the collaborative filtering module performs collaborative filtering calculation according to the preference score of the data product to output a second recommendation score.

[0045] Step 206, weighting and fusing the first recommendation score and the second recommendation score to output a fused recommendation score; generating a data product recommendation list according to the fused recommendation score.

[0046] Step 207, dynamically updating the related parameters of the dynamic preference model and the filtering recommendation model according to the feedback information of the decision maker on the data product recommendation list, and performing parallel recommendation service based on the distributed architecture.

[0047] In the specific implementation process of step 201, multi-dimensional data of the decision maker in the emergency training management process is collected, including behavior data and static data.

[0048] The behavior data includes operation logs, historical data retrieval records and instruction issuing preferences, wherein the operation logs include the type, time and object of each operation, the historical data retrieval records include the data type, time and frequency of each retrieval, and the instruction issuing preferences include the style, frequency and object of each type of instruction.

[0049] The static data includes a personalized label, an emergency field knowledge graph and a current training task index parameter, wherein the personalized label is label information input by the decision maker; the emergency field knowledge graph , is an emergency field entity, is the relationship between entities; the current training task index parameter includes training type, scale, target and environment parameters.

[0050] During preprocessing, unstructured data (such as log text) is converted into structured feature vectors using log template extraction and TF-IDF methods. For example, feature extraction is performed on operation logs to obtain behavioral feature vectors. The personalized tags are encoded to obtain the tag feature vector. Knowledge graph feature vectors are extracted from emergency domain knowledge graphs using graph embedding algorithms. The task feature vector is obtained by performing feature processing on the current training task index parameters. .

[0051] This step, through multi-dimensional data collection and structured processing, provides comprehensive and standardized input data for subsequent model building, breaking the limitations of a single data type and ensuring that the data accurately reflects the needs of decision-makers and the characteristics of the scenario.

[0052] In the specific implementation of step 202, the structured feature vector includes behavioral feature vectors, label feature vectors, knowledge graph feature vectors, and task feature vectors. These four types of vectors are fused to obtain a fused feature vector. The calculation expression for the fused feature vector is as follows:

[0053] ;

[0054] In the formula, Represents the fused feature vector; Represents a behavioral feature vector; Represents the label feature vector; Represents the feature vector of a knowledge graph; Represents the task feature vector; , , , This indicates the fusion weight.

[0055] Dynamic preference model Implemented using deep neural networks, such as Attention-LSTM, the feature vectors are fused. Input the dynamic preference model, and map the relationship between the models. Output decision-makers' preference score vectors for various data products Preference score vector Each element corresponds to a quantitative value representing the degree of preference for a particular data product.

[0056] This step, through multi-feature fusion and deep neural network modeling, can accurately capture the decision-maker's personalized preferences, while incorporating emergency domain knowledge and task requirements, making the preference scores more aligned with the professional logic of emergency training and management scenarios.

[0057] In the implementation of step 203, the distributed architecture can be implemented by using the MapReduce framework, through which the recommendation tasks of multiple users are divided into independent sub-tasks, distributed to different computing nodes for parallel execution, thereby providing a basis for subsequent parallel recommendation of multiple users and multiple scenarios.

[0058] Among the two modules of the filtering recommendation model, the recommendation module focuses on the adaptability calculation of the individual dynamic preference of the decision maker and the current task, and the collaborative filtering module supplements the recommendation basis by means of the common preference of the group decision makers, thereby combining the individualization and the group to form a dual recommendation logic of individualization and group, which ensures the pertinence of the recommendation and improves the reliability of the recommendation result.

[0059] In the implementation of step 204, the recommendation module uses a reinforcement learning algorithm to model the recommendation process as a Markov decision process (MDP). The state space wherein, is the preference score vector output by step 202, is the current task feature vector, is the set of recommended data products.

[0060] The action space is the set of all optional data products in the emergency training management scene (such as training progress visualization dashboards, risk warning analysis reports, etc.); and the reward function expression is:

[0061] ;

[0062] In the expression, represents the reward function; represents the state space; represents the data product; represents the weight of the reward function; represents the behavioral feedback of the decision maker to the data product , including the quantified scores corresponding to the use duration, click frequency, and acceptance or rejection of instructions; represents the matching degree of the data product and the task feature vector .

[0063] The reinforcement learning agent selects a data product through a reinforcement strategy , and the goal is to maximize the cumulative reward, and finally outputs the first recommendation score , which quantifies the degree of adaptation of the data product to the dynamic preference of the decision maker and the current task through the first recommendation score .

[0064] The maximization of the cumulative reward expression is:

[0065] ;

[0066] wherein, represents a reinforcement policy corresponding cumulative reward expectation; represents a mathematical expectation operation; represents a time step; represents a discount factor; represents a time step , in a state selects an action obtained instant reward; represents a time step corresponding state space; represents a time step selected data product.

[0067] In the implementation process of step 205, the collaborative filtering module first calculates a similarity value according to the preference score of the current decision maker to the data product and the preference score of other decision makers to the data product, and the expression is:

[0068]

[0069] wherein, represents a similarity value; represents a current decision maker; represents other decision makers; represents a preference score vector of the current decision maker to the data product; represents a preference score vector of other decision makers to the data product; represents a preference score of the current decision maker to the data product; represents a preference score of other decision makers to the data product.

[0070] Then, the second recommendation score of the current decision maker is calculated based on the similarity value, and the expression is:

[0071]

[0072] wherein, represents a second recommendation score; represents a set of decision makers similar to represents a score of other decision makers to the data product .

[0073] ​​​In the implementation process of steps 203 to 205, the fusion of reinforcement learning and collaborative filtering algorithm is realized by filtering the two modules of the recommendation model design, which can not only capture the dynamic preference and task adaptability of the decision maker through reinforcement learning, but also learn from the selection of other decision makers through collaborative filtering, making the recommendation logic more comprehensive and effectively improving the accuracy and reliability of the recommendation results. At the same time, the deployment of the distributed architecture lays the foundation for subsequent parallel recommendation services and solves the recommendation efficiency problem in large-scale scenarios.

[0074] In the implementation process of step 206, the first recommendation score and the second recommendation score are weighted and fused, and the calculation expression is:

[0075] ;

[0076] In the formula, represents the fusion recommendation score; represents the score fusion coefficient, which is used to balance the dynamic adaptability of reinforcement learning and the group reference value of collaborative filtering.

[0077] Then, according to the fusion recommendation score , all the selectable data products are sorted in descending order, and the top data products (N is a preset recommendation quantity threshold) are selected to generate a data product recommendation list and recommended to the current decision maker.

[0078] This step integrates the recommendation advantages of the two modules through a reasonable weighted fusion strategy, so that the recommendation results not only meet the individual needs of the decision maker, but also have group recognition. At the same time, through sorting and screening, the quality of the recommendation list is ensured.

[0079] In the implementation process of step 207, the feedback information includes the decision maker's acceptance or rejection instructions, data product usage time, manual score, etc. These feedback information is converted into parameter adjustment basis through a preset quantitative rule.

[0080] The related parameters updated dynamically include: the fusion weight of the dynamic preference model , , , , the reward function weight of reinforcement learning , and the score fusion coefficient of the filtering recommendation model . After updating the parameters, based on the distributed architecture, independent computing resources are allocated to multiple users, and each computing node independently completes the data preprocessing, model inference, and data product recommendation list generation process. Through the distributed synchronization mechanism, the consistency of the model parameter update is guaranteed, supporting multiple users and multiple scenarios of parallel recommendation requests per second, and meeting the demand of large-scale training management scenarios. ​

[0081] The model is self-optimized through feedback-driven parameter updating, so that the recommendation system can adapt to the dynamic changes of decision maker preferences and training tasks in real time, and meanwhile, with the parallel processing capability of the distributed architecture, the concurrent recommendation requirements of large-scale emergency training management scenarios are met.

[0082] In an embodiment, taking the fire rescue training management scenario as an example, the implementation process of the present application is described in detail:

[0083] First, data collection and preprocessing are performed. The operation logs of fire commanders (decision makers) (such as viewing training videos, adjusting training parameters), historical access records (such as simulation training data in the past 3 months), and command preferences (such as preferring detailed instructions or summary instructions) are collected. The commanders input personalized tags: {“rescue”, “command and decision”, “resource allocation”}. The fire rescue domain knowledge graph (including rescue types, rescue equipment, tactical methods, etc.) is loaded. The current training task indicators are obtained: {“training type = high-rise building rescue”, “number of trainees = 50”, “training goal = improve initial control ability”}. The above data is preprocessed to generate behavior feature vectors , label feature vectors , knowledge feature vectors , and task feature vectors .

[0084] Then, a dynamic preference model is constructed. The fusion feature vector is calculated according to the formula , and the dynamic preference model is trained through a deep neural network , and the preference score vector of the commander for each type of data product is output .

[0085] Among them, in the filtering recommendation model calculation process, the recommendation module takes the , the current task feature vector , and the recommended data product set as the state space , selects a data product such as “high-rise building rescue resource allocation prediction model” according to the state space , calculates the reward value according to the use duration of the commander and the task matching degree , and outputs it as the first recommendation score.

[0086] The collaborative filtering module calculates the similarity between the commander and other rescue commanders, discovers that similar users often use “risk warning analysis report”, and generates a collaborative filtering score ; the fusion recommendation score is calculated according to the fusion formula: .

[0087] Then, the recommendation and feedback optimization are performed. The data product recommendation list is generated according to the fusion recommendation score ranking and is pushed to the commander; if the commander accepts and frequently uses the recommended product, the related parameters of the dynamic preference model and the filtering recommendation model are dynamically updated, and the weight of the related features is improved.

[0088] Finally, the distributed architecture is used to provide the recommendation service for multiple commanders in parallel in the fire rescue scene.

[0089] It should be understood that although each step in the embodiment Figure 1 is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps in the embodiment may include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with other steps or at least part of the sub-steps or stages of other steps.

[0090] Embodiment 2

[0091] Based on the data product intelligent recommendation method in Embodiment 1, the embodiment discloses a data product intelligent recommendation system, as shown in Figure 2 The data product intelligent recommendation system includes a data preprocessing module 401, a preference score calculation module 402, a filtering recommendation model construction module 403, a first recommendation score calculation module 404, a second recommendation score calculation module 405, a data product recommendation list generation module 406, and a feedback update module 407, wherein:

[0092] The data preprocessing module 401 is configured to obtain the behavior data of the decision maker, the personalized label, the emergency field knowledge graph, and the current training task index parameter in the emergency training management scene, and to preprocess the same to obtain a structured feature vector.

[0093] The preference score calculation module 402 is configured to construct a dynamic preference model, input the structured feature vector into the dynamic preference model for processing, and output the preference scores of various data products.

[0094] The filtering recommendation model construction module 403 is configured to construct a filtering recommendation model deployed based on a distributed architecture, and the filtering recommendation model includes a recommendation module and a collaborative filtering module.

[0095] The first recommendation score calculation module 404 is used in the recommendation module to calculate and output the first recommendation score using the preference score, the current training task indicator parameters, and the set of recommended data products as the state space, through a reward function that includes behavioral feedback and task matching degree.

[0096] The second recommendation score calculation module 405 is used in the collaborative filtering module to perform collaborative filtering calculations based on the preference scores of the data products and output the second recommendation score.

[0097] The data product recommendation list generation module 406 is used to perform weighted fusion of the first recommendation score and the second recommendation score, and output the fused recommendation score; and generate a data product recommendation list based on the fused recommendation score.

[0098] The feedback update module 407 is used to dynamically update the relevant parameters of the dynamic preference model and the filtering recommendation model based on the decision-maker's feedback on the data product recommendation list, and to provide parallel recommendation services based on a distributed architecture.

[0099] In this embodiment, the specific working process and working principle of the data preprocessing module 401, preference score calculation module 402, filtering recommendation model construction module 403, first recommendation score calculation module 404, second recommendation score calculation module 405, data product recommendation list generation module 406, and feedback update module 407 are the same as those in Embodiment 1, and therefore will not be described again in this embodiment. Each unit module can be implemented entirely or partially through software, hardware, or a combination thereof. Each unit module can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above unit modules.

[0100] Example 3

[0101] like Figure 3 The diagram illustrates a computer device disclosed in this embodiment, including a transmitter, a receiver, a memory, and a processor. The transmitter transmits instructions and data, the receiver receives instructions and data, the memory stores computer execution instructions, and the processor executes the computer execution instructions stored in the memory to implement the method in Embodiment 1 described above.

[0102] It is important to note that the aforementioned memory can be either standalone or integrated with the processor. When the memory is set up independently, the computer device also includes a bus for connecting the memory and the processor.

[0103] Example 4

[0104] The embodiment discloses a computer readable storage medium, and computer readable storage medium has computer execution instructions stored therein. When a processor executes the computer execution instructions, the method in the above embodiment 1 is realized.

[0105] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the processes of the above-mentioned embodiments can be included. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0106] Each technical feature of the above embodiments can be combined arbitrarily. In order to make the description simple, each technical feature of the above embodiments is not described in all possible combinations, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present disclosure.

[0107] The above embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application.

Claims

1. A method for intelligent recommendation of data products, characterized in that, The method comprises: acquiring behavior data of a decision maker in an emergency training management scenario, personalized labels, an emergency domain knowledge graph, and current training task index parameters, and preprocessing to obtain a structured feature vector; constructing a dynamic preference model, inputting the structured feature vector into the dynamic preference model for processing, and outputting preference scores of various data products; constructing a filtering recommendation model deployed based on a distributed architecture, the filtering recommendation model comprising a recommendation module and a collaborative filtering module; the recommendation module taking the preference scores, the current training task index parameters, and a set of recommended data products as a state space, calculating through a reward function comprising behavior feedback and task matching degree, and outputting a first recommendation score; the collaborative filtering module performing collaborative filtering calculation according to the preference scores of data products, and outputting a second recommendation score; weighting and fusing the first recommendation score and the second recommendation score to output a fused recommendation score, and generating a data product recommendation list according to the fused recommendation score; dynamically updating related parameters of the dynamic preference model and the filtering recommendation model based on feedback information of the decision maker on the data product recommendation list, and performing parallel recommendation services based on the distributed architecture.

2. The method of claim 1, wherein, The structured feature vector comprises a behavior feature vector, a label feature vector, a knowledge graph feature vector, and a task feature vector; performing fusion processing on the behavior feature vector, the label feature vector, the knowledge graph feature vector, and the task feature vector to obtain a fused feature vector; inputting the fused feature vector into the dynamic preference model for processing to output preference scores of various data products by the decision maker.

3. The method of claim 2, wherein, The calculation expression of the fused feature vector is: ； In the formula, denotes a fusion feature vector; denotes a behavior feature vector; denotes a label feature vector; denotes a knowledge graph feature vector; denotes a task feature vector; , , , denotes a fusion weight.

4. The method of claim 1, wherein, In the recommendation module, the reward function expression is: ； wherein, represents a reward function; represents a state space; represents a data product; represents a reward function weight; represents a decision maker's behavioral feedback on a data product ; represents a matching degree of a data product to a task feature vector .

5. The method of claim 4, wherein, The collaborative filtering module performs collaborative filtering calculation according to the preference scores of data products, and outputs a second recommendation score, comprising: In the collaborative filtering module, a similarity value is calculated according to the preference score of the current decision maker on the data product and the preference scores of other decision makers on the data product; the second recommendation score of the current decision maker is calculated based on the similarity value.

6. The method of claim 5, wherein, The similarity value is calculated according to the preference score of the current decision maker on the data product and the preference scores of other decision makers on the data product, and the expression is: ； wherein, represents a similarity value; represents a current decision maker; represents other decision makers; represents a preference score vector of the current decision maker for the data product; represents a preference score vector of the other decision makers for the data product; represents a preference score of the current decision maker for the data product; represents a preference score of the other decision makers for the data product.

7. The method of claim 6, wherein, The second recommendation score of the current decision maker is calculated based on the similarity value, and the expression is: ； wherein, represents a second recommendation score; represents a set of decision makers similar to represents other decision makers rating of a data product .​ 8. A data product intelligent recommendation system, characterized in that, The system comprises: a data preprocessing module configured to acquire behavior data of a decision maker in an emergency training management scenario, personalized labels, an emergency domain knowledge graph, and current training task index parameters, and preprocess to obtain a structured feature vector; a preference score calculation module configured to construct a dynamic preference model, input the structured feature vector into the dynamic preference model for processing, and output preference scores of various data products; a filtering recommendation model construction module configured to construct a filtering recommendation model deployed based on a distributed architecture, the filtering recommendation model comprising a recommendation module and a collaborative filtering module; The first recommendation score calculation module is configured to, in the recommendation module, take the preference score, the current training task index parameter and the recommended data product set as a state space, and calculate a first recommendation score by using a reward function containing behavior feedback and task matching degree. The second recommendation score calculation module is configured to, in the collaborative filtering module, perform collaborative filtering calculation according to the preference score of the data product, and output a second recommendation score. The data product recommendation list generation module is configured to perform weighted fusion on the first recommendation score and the second recommendation score, output a fused recommendation score, and generate a data product recommendation list according to the fused recommendation score. The feedback update module is configured to dynamically update related parameters of the dynamic preference model and the filtering recommendation model according to feedback information of the decision maker on the data product recommendation list, and perform parallel recommendation services based on a distributed architecture. 9.A computer device, comprising a memory and a processor, and characterized in that, The memory stores a computer program, and the processor implements the steps of the data product intelligent recommendation method in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer program is stored on the memory and is executed by the processor to implement the steps of the data product intelligent recommendation method in any one of claims 1 to 7.

Images