Method for implementing decision case module recommendation based on autonomous recommendation mechanism

Abstract

The application provides a decision case module recommendation implementation method based on an autonomous recommendation mechanism, comprising the following steps: Step 1, based on a basic database, generating a decision case template according to an event processing mechanism, classifying and managing decision case modules, and forming a case template platform; Step 2, extracting a specific decision case template from the case template platform according to an event situation, and obtaining a to-be-verified recommendation scheme; Step 3, comparing and evaluating the to-be-verified recommendation scheme through the autonomous recommendation mechanism, and obtaining an optimal recommendation scheme. The application solves the problems of low efficiency of artificial browsing and selection of decision cases in the urban governance process and the problem of difficulty in selection of decision cases due to industry barriers in the urban governance process.

Description

Technical Field

[0001] This invention relates to the fields of public infrastructure decision-making case templates and event handling, specifically to a method for recommending decision-making case modules based on a self-recommendation mechanism.

[0002] Invention title. Background Technology

[0003] With the rapid development of information technology in my country, smart cities have moved from concept to reality. The fundamental element of a smart city is the various IoT sensors within modern urban public infrastructure. These sensors are installed in various locations throughout the city, monitoring various urban characteristics in real time and reporting instantaneous monitoring data. Hundreds of millions of data points are collected on a monitoring platform, forming a big data center. Based on this foundational data, various practical application scenarios are derived. A core scenario involves defining event rules for the collected data, generating numerous event information entries from the monitoring data, and then adding various pre- and post-conditions and verification rules (complying with policy documents and laws and regulations) to these events, ultimately creating event handling case templates for city managers to reference.

[0004] Once case templates are created, city managers cannot easily select the most suitable templates for various industries from a large pool of templates. Manually browsing and selecting case templates is not only inefficient, but also unsuitable if the research on a particular industry is insufficient. Therefore, to further facilitate city managers in accurately selecting and using case templates and thereby improve the efficiency of comprehensive urban management, a self-recommendation mechanism for event case templates based on the operational status of various industries in the city has emerged. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method for recommending decision-making cases based on an autonomous recommendation mechanism, thereby solving the problems of low efficiency in manual browsing and selection of decision-making cases in urban governance, and the difficulty in selecting decision-making cases due to industry barriers.

[0006] The present invention provides a method for implementing a decision case module recommendation based on an autonomous recommendation mechanism, comprising the following steps:

[0007] Step 1: Based on the basic database and the event handling mechanism, generate decision case templates, classify and manage the decision case module, and form a case template platform;

[0008] Step 2: Extract specific decision-making case templates from the case template platform based on the event situation to obtain recommended solutions to be verified.

[0009] Step 3: The proposed recommendations to be verified are compared and evaluated using an autonomous recommendation mechanism to obtain the optimal recommendation.

[0010] Preferably, the autonomous recommendation mechanism includes a recommendation algorithm, model training, strategy design, and evaluation analysis;

[0011] The recommendation algorithm learns the recommendation scheme to be verified, obtains the recommendation scheme to be trained, and generates the corresponding recommendation model;

[0012] The model training simulates and recreates the recommendation scheme to be trained based on the recommendation model to obtain a simulated event case model;

[0013] The strategy design filters and sorts the simulated event case models to obtain recommended solutions to be evaluated.

[0014] The evaluation analysis performs a reverse evaluation of the recommended scheme to be evaluated, and obtains the optimal recommended scheme.

[0015] Preferably, the basic database performs data analysis and processing on basic data from different industries; the event handling mechanism defines and manages event handling rules for different industries and scenarios.

[0016] Preferably, the extraction of the recommended solution to be verified is a function that fits historical event decision cases and user satisfaction. This function contains variables in three dimensions:

[0017] Content: The event case content is diverse, and we need to consider how to extract features from different event case types to make better recommendations;

[0018] User: How to extract the characteristics of the user to whom the adopted event case belongs;

[0019] Scenario: The event case preferences of current users in various industries and scenarios.

[0020] Preferably, the model training first involves processing operational data from historical decision event cases, updating the recommendation model based on the collected data, and simultaneously importing complete user data from a cluster using a file queue containing real-time features recorded on an online server. This data is then combined with the recommendation scheme to be trained to construct simulated sample event cases. The parameters of the recommendation model are then updated based on these simulated sample event cases, and the model is run online to obtain the simulated event case model.

[0021] Preferably, the parameters of the recommendation model are stored in a high-performance server cluster, containing a massive amount of original features and vector features.

[0022] Preferably, the recommendation algorithms shown include Logistic Regression, Factorization Machine, and GBDT (Gradient Boosting Decision Tree);

[0023] The Logistic Regression algorithm constructs a classification model to automatically classify the recommendation scheme to be validated and obtain classification feature data.

[0024] The Factorization Machine algorithm combines classification feature data to obtain decision cases to be recommended.

[0025] The GBDT (Gradient Boosting Decision Tree) algorithm fits the decision cases to be recommended to obtain predicted values ​​close to the true values ​​based on actual needs, and the decision cases to be recommended with high predicted values ​​are defined as recommendation schemes to be trained.

[0026] Preferably, the strategy design is based on the recall module filtering and sorting the simulated event case models to generate a recommendation list, and realizes online real-time updates.

[0027] Preferably, the recall module includes scene deduplication, diversity control, and weighting methods.

[0028] Preferably, the evaluation analysis assesses the rationality and correctness of the recommended scheme through reverse deduction, outputs an application analysis report on the recommended scheme according to the established evaluation system, scores the recommended scheme, and determines the optimal recommended scheme based on the preset recommendation ranking quota.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. This invention can effectively improve the efficiency of producing case studies for event decision-making in urban governance.

[0031] 2. This invention can effectively break down industry barriers to the use of case studies in urban governance.

[0032] 3. This invention can effectively shorten the decision-making time for events in urban governance and improve the efficiency of event handling, thus empowering smart cities. Attached Figure Description

[0033] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0034] Figure 1This is a schematic diagram of the logical flow structure of a decision case module recommendation implementation method based on an autonomous recommendation mechanism according to the present invention.

[0035] Figure 2 This is a schematic diagram of the operational flow structure of the autonomous recommendation mechanism event case of the present invention; Detailed Implementation

[0036] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0037] like Figure 1 As shown, this invention provides a method for implementing a decision-making case module recommendation based on an autonomous recommendation mechanism, comprising the following steps:

[0038] Step 1: Based on the basic database and the event handling mechanism, generate decision case templates, classify and manage the decision case module, and form a case template platform;

[0039] The basic database performs data analysis and processing on fundamental data from different industries. This fundamental data includes, but is not limited to: massive amounts of fundamental data and related technical parameters of a vast number of devices across various industries.

[0040] The incident handling mechanism defines and manages incident handling rules for different industries and scenarios. These rules include, but are not limited to, the legal regulations, codes of conduct, and other rules governing incident handling defined for each industry and scenario.

[0041] Specifically, a massive amount of basic data and related technical parameters from various industries' equipment are aggregated, calculated, and managed on an IoT platform to form the foundational database needed for generating decision case templates. Event rules are defined and managed for various industries, including major sectors like public security, fire protection, transportation, and urban management, as well as specific sub-sectors such as shopping malls, airports, hotels, schools, and government agencies. Warning rules for scenarios such as typhoons, river levels, and smoke concentrations are defined, along with event handling rules. These are then incorporated into the event handling mechanism for categorized and standardized management. Based on these two conditions, a large number of decision case templates for various events are intelligently generated, and these templates are managed to form a case template platform.

[0042] Step 2: Extract specific decision-making case templates from the case template platform based on the event situation to obtain recommended solutions to be verified.

[0043] Specifically, the extraction of the recommendation scheme to be validated is a function that fits historical event decision cases and user satisfaction. This function contains variables in three dimensions:

[0044] Content: The event case content is diverse, and we need to consider how to extract features from different event case types to make recommendations.

[0045] User: How to extract the characteristics of the user to whom the adopted event case belongs?

[0046] Scenario: The event case preferences of current users in various industries and scenarios.

[0047] Step 3: The recommended solutions to be verified are compared and evaluated through the self-recommendation mechanism to obtain the optimal recommended solution;

[0048] The autonomous recommendation mechanism includes recommendation algorithms, model training, strategy design, and evaluation analysis;

[0049] The recommendation algorithm learns the recommendation scheme to be validated, obtains the recommendation scheme to be trained, and generates the corresponding recommendation model;

[0050] Specifically, recommendation algorithms include three main algorithms: Logistic Regression, Factorization Machine, and GBDT (Gradient Boosting Decision Tree).

[0051] The Logistic Regression algorithm automates the classification of recommendation schemes by constructing classification models, thereby obtaining classification feature data. Specifically, Logistic Regression is a machine learning method for solving binary (0 or 1) classification problems, used to estimate the probability of something. This algorithm is used in the classification module of a recommendation system, primarily to solve multi-class problems. It constructs various classification models required by the system and supervises the system to learn automatically, thus automating the classification process.

[0052] The Factorization Machine algorithm combines categorical feature data to obtain decision cases to be recommended. Specifically, the Factorization Machine algorithm adds second-order (or higher-order) feature interactions to a standard linear model, using matrix factorization to map an n*n weight matrix into an n*k space. This algorithm is used to combine atomic decision cases of the system according to features, generating a new decision case, i.e., a decision case to be recommended, for use in recommendations.

[0053] The GBDT (Gradient Boosting Decision Tree) algorithm fits the proposed decision cases to obtain predicted values ​​close to the true values ​​based on actual needs. Decision cases with high predicted values ​​are defined as the recommendation schemes to be trained. Specifically, GBDT is an ensemble algorithm based on decision trees. Gradient Boosting is an algorithm within the boosting ensemble method, which iterates the new learner using gradient descent. GBDT uses the CART decision tree. The core function of this algorithm in this system is to continuously train the predicted values ​​closest to the true values ​​using the concept of negative gradient approximation residuals, that is, to recommend the decision case with the highest similarity to the user.

[0054] The three algorithms mentioned above are independent in implementation but interconnected in business logic. The Logistic Regression algorithm serves the classification module, obtaining classification feature data. The Factorization Machine algorithm combines the classification feature data according to features to generate recommendation decision cases. The GBDT algorithm fits the user's requirements to the system to obtain predicted values ​​close to the true values, and finally selects a recommendation scheme to recommend to the user.

[0055] Model training involves simulating and recreating the recommended solutions to be trained based on the recommendation model to obtain simulated event case models.

[0056] Specifically, model training begins with processing historical decision event data, including clicks, impressions, and shares. The recommendation model is updated based on this processed data. Simultaneously, real-time features recorded on an online server are imported into a file queue to concatenate complete user data within the cluster. This data, combined with the recommendation scheme to be trained, constructs simulated sample event cases. The parameters of the recommendation model are then updated based on these simulated sample event cases, and the model is run online to obtain the simulated event case model. Notably, the parameters of the recommendation model are stored on a high-performance server cluster, containing massive amounts of raw features and vector features.

[0057] The strategy design involves screening and ranking simulated event case models to obtain recommended solutions for evaluation.

[0058] Specifically, the strategy design uses a recall module to filter and rank simulated event case models, generate a recommendation list, and update it online in real time. The recall module includes methods such as scenario deduplication, diversity control, and weighting. A crucial strategy used in the recommendation strategy design is the recall strategy. The purpose of this strategy is to select a small subset of eligible event decision cases from a massive number of event decision cases.

[0059] The above implementation steps require deep learning computing. Specifically, data extraction, including event cases, event rules, laws and regulations, policy documents, industry data, and scenario data, is performed using both streaming and batch processing. A large-scale, high-performance storage system supports the reading and writing of massive amounts of events, and the entire event decision-making case recommendation system is calculated based on this data.

[0060] The evaluation analysis performs a reverse evaluation of the recommended solution to obtain the optimal recommended solution.

[0061] Specifically, the evaluation and analysis assesses the rationality and correctness of the recommended solutions through reverse engineering. Based on a predetermined evaluation system, it outputs an application analysis report detailing the applicability and recommendation level of each solution. Furthermore, it scores the recommended solutions and determines the optimal recommendation based on a preset ranking limit. For example, if the preset ranking limit is 3, the top 3 scoring solutions will be recommended to managers for decision-making.

[0062] In practical applications, determining whether the final recommendation effect of the optimal recommendation scheme meets user requirements requires complex evaluation and analysis. This involves reverse engineering to determine the rationality and correctness of the recommendation scheme, algorithm, strategy, and computation. Firstly, factors that may affect recommendation performance include changes in the candidate content set, the addition and improvement of the recall module, the addition of recommendation features, improvements to the recommendation system architecture, optimization of algorithm parameters, and changes in rules and strategies. Secondly, the evaluation requires a comprehensive evaluation system: synthesizing as many comprehensive indicators as possible into a single evaluation metric, a robust experimental platform, and user-friendly experimental analysis tools. Finally, the evaluation should pay attention to the following points: focusing on both short-term and long-term indicators, considering the impact of synergistic effects, and performing thorough statistical isolation when necessary.

[0063] In summary, the prerequisites for this invention include massive amounts of sensor monitoring data, numerous event rules across various industries (such as rules for handling emergencies, rules for monitoring and early warning of various equipment), and a large number of multi-industry event case templates intelligently generated from the above two elements. Secondly, the autonomous recommendation mechanism contains a core recommendation algorithm. This algorithm performs collision calculations with various existing event rules within the system, categorizing them by industry and scenario. It also searches and compares event templates already adopted by various industries (using behavioral analysis algorithms and big data retrieval) to meet the system's recommendation scheme. The algorithm is then run, trained according to the recommendation model, and subjected to deep computation using a pre-designed recommendation strategy, followed by evaluation using an evaluation system. After this entire process of scheme, algorithm, strategy, training, and evaluation is completed, multiple case templates are obtained. For each case template, an industry- and scenario-specific application analysis report is provided, detailing the applicability and recommendation level of the case. Finally, these case templates are scored, and the top-scoring cases are recommended to city managers for decision-making. In this way, a relatively complete self-recommendation mechanism is formed, which includes the above recommendation schemes, algorithms, recommendation models, strategies, logical calculations and evaluation systems. All of the above elements work together to search for event cases that meet the user's actual needs in the system and finally recommend them to the user for reference.

[0064] In practical applications, such as Figure 2 As shown, by selecting an industry and application scenario to determine the demand direction, and adjusting the complexity (generally set to 50-100), decision case templates are generated through a case template platform. Then, the decision case templates are evaluated using the self-recommendation mechanism-based decision case module recommendation implementation method described above. Based on the approval of the review and analysis report, the optimal recommended solution is obtained and implemented; otherwise, decision case templates that fail the review and analysis report are discarded. Furthermore, after the optimal recommended solution is implemented, it can be assessed whether a new round of review and evaluation based on the self-recommendation mechanism is needed, based on actual usage. Thus, through two-way self-recommendation evaluation learning based on historical data and actual usage, and by continuously updating data online to promote repeated learning of the self-recommendation mechanism, the final evaluation values ​​of the decision case module recommendation implementation method based on the self-recommendation mechanism have greater reference and application value.

[0065] Furthermore, in practical applications, when a responsible department is unclear about the business processes under its jurisdiction or the incident handling process, it can use the incident case self-recommendation mechanism to refer to the mature incident case handling processes of other cities or departments, and absorb the excellent and mature experience in handling similar incident cases to empower its own department.

[0066] On the other hand, when responsible departments are not very familiar with industry standards, laws and regulations, and policy documents, resulting in knowledge gaps that affect the selection of case studies and decision-making on major events, a self-recommendation mechanism for case studies can be used. This mechanism allows them to select several specific case studies with higher scores, and the analysis report summarizing the self-recommended case studies can be cross-checked with industry standards, laws and regulations, and policy documents. This saves time on case study compilation, review, and trial operation, improving the efficiency of urban management operations.

[0067] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0068] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A decision case module recommendation implementation method based on an autonomous recommendation mechanism, characterized in that, The method comprises the following steps: Step 1: Based on the basic database, according to the event processing mechanism, a decision case template is generated, the decision case module is classified and managed, and a case template platform is formed; Step 2: According to the event, a specific decision case template is extracted from the case template platform to obtain a to-be-verified recommendation scheme; Step 3: The to-be-verified recommendation scheme is compared and evaluated through the autonomous recommendation mechanism to obtain an optimal recommendation scheme; The autonomous recommendation mechanism includes a recommendation algorithm, model training, strategy design, and evaluation analysis; The recommendation algorithm learns the to-be-verified recommendation scheme to obtain a to-be-trained recommendation scheme and generates a corresponding recommendation model; The model training simulates and restores the to-be-trained recommendation scheme based on the recommendation model to obtain a simulated event case model; The strategy design filters and sorts the simulated event case model to obtain a to-be-evaluated recommendation scheme; The evaluation analysis reversely evaluates the to-be-evaluated recommendation scheme to obtain an optimal recommendation scheme; The basic database analyzes and processes the basic data of different industries; the event processing mechanism defines and manages the event processing rules of different industries and different scenes; The extraction of the to-be-verified recommendation scheme is a function of fitting historical event decision cases and user satisfaction, which includes three-dimensional variables: Content: The content of the event case is diversified, and how to extract the characteristics of different event case types for recommendation needs to be considered; User: How to extract the characteristics of the user of the adopted event case; Scene: In various industries and various scenes, the event case preferences of the current user are needed.

2. The method as claimed in claim 1, wherein the autonomous recommendation mechanism based decision case module recommendation implementation method is characterized in that, The model training is based on the operation data processing of historical decision event cases, and the recommendation model is updated according to the collected data processing. At the same time, the online server records the real-time feature import file queue import cluster splicing user complete data, combines the to-be-trained recommendation scheme, and constructs a simulated sample event case. Then, the parameters of the recommendation model are updated according to the simulated sample event case, the online model is run, and the simulated event case model is obtained.

3. The method of claim 2, wherein the method further comprises: The parameters of the recommendation model are stored in a high-performance server cluster, which includes a large amount of original features and vector features.

4. The method as claimed in claim 1, wherein the autonomous recommendation mechanism based decision case module recommendation implementation method is characterized in that, The recommendation algorithm includes Logistic Regression, Factorization Machine, and GBDT; The Logistic Regression algorithm realizes the automatic classification of the to-be-verified recommendation scheme by constructing a classification model to obtain classification feature data; The Factorization Machine algorithm combines the classification feature data to obtain a to-be-recommended decision case; The GBDT algorithm fits the to-be-recommended decision case based on actual demand to obtain a predicted value close to the true value, and defines the to-be-recommended decision case with a high predicted value as a to-be-trained recommendation scheme.

5. The method as claimed in claim 1, wherein the autonomous recommendation mechanism based decision case module recommendation implementation method further comprises: The strategy design filters and sorts the simulated event case model based on the recall module to generate a recommendation list and realize online real-time updating.

6. The method of claim 5, wherein the method further comprises: The recall module includes scene deduplication, diversity control, and weighted methods.

7. The method as claimed in claim 1, wherein the autonomous recommendation mechanism based decision case module recommendation implementation method further comprises: The evaluation analysis evaluates the rationality and correctness of the to-be-evaluated recommendation scheme by reverse deduction, outputs an application analysis report of the to-be-evaluated recommendation scheme according to the established evaluation system, scores the to-be-evaluated recommendation scheme, and determines the optimal recommendation scheme according to a preset recommendation ranking quota.

Images