Large model-based intelligent enterprise decision making method

Abstract

This invention, based on a large-scale model, is a smart enterprise decision-making method belonging to the field of big data. It addresses the problems of existing smart enterprise decision-making methods, which involve limited and incomplete business considerations and require custom-built, run, and deployed decision models, resulting in complex, time-consuming, and inefficient construction processes. The method first collects datasets and decision models corresponding to various business needs. It then uses a large-scale model for data discovery to identify the most relevant datasets and multiple meta-features of these datasets. Next, it searches for decision models related to each meta-feature and uses automated machine learning strategies to optimize these models. Finally, it uses the optimal decision model to predict the business needs to be predicted, outputting various types of data related to these needs. This invention is primarily used for enterprise decision-making.

Description

Technical Field

[0001] This invention belongs to the field of big data. Background Technology

[0002] In existing technologies, solving various problems faced by businesses in their operations is often done directly by the owner or indirectly through interdepartmental cooperation. These problems typically include: market analysis and trend forecasting, inventory management and demand forecasting, customer insights and personalization, cost accounting and financial optimization, supply chain optimization, quality control and defect detection, employee performance and training, predictive maintenance of equipment, regulatory compliance assistance, sales forecasting and revenue optimization, and so on. Currently, these problems are often decided manually, either directly by the owner or indirectly through interdepartmental cooperation. While existing manual decision-making can comprehensively address these business problems, it requires significant human and material resources and lacks sufficient accuracy and efficiency.

[0003] In addition to traditional human decision-making, there are now intelligent enterprise decision-making products in the technology, but they have the following problems: (1) Limited business scope and incomplete problem-solving. For example, Tongdun Technology focuses on finance and security, with limited business scope; Youhualin can only provide intelligent supply chain planning, which also has limited business scope; (2) IBM's decision optimization model needs to be built, run and deployed by itself. The construction process is complicated, time-consuming and inefficient, which is not friendly to non-computer professionals. Moreover, IBM's decision optimization model involves data privacy and data sensitivity issues; (3) Human-computer interaction is achieved by buttons, which is not simple. Therefore, the above problems need to be solved. Summary of the Invention

[0004] The purpose of this invention is to address the problems of existing intelligent enterprise decision-making methods, which involve fewer and less comprehensive business decisions, and require the decision-making models to be built, run, and deployed independently, resulting in complex construction processes, long time consumption, and low decision-making efficiency. This invention provides an intelligent enterprise decision-making method based on a large model.

[0005] A large-model-based intelligent enterprise decision-making method, which includes the following steps:

[0006] S1. Collect datasets of corresponding attributes for various business requirement problems deployed locally, as well as various decision models for various business requirement problems; each decision model corresponds to a corresponding attribute label, and each dataset includes various types of data related to the business requirement problem at each sampling time;

[0007] S2. Using a large model for data discovery, based on a business requirement question to be predicted in each dialogue segment and the dataset of all collected attributes, determine the most relevant dataset for the business and multiple meta-features of the most relevant dataset for the business.

[0008] S3. Search for decision models related to each meta-feature, and combine them with the most relevant dataset for the business and the given performance metrics to optimize the optimal decision model using an automatic machine learning strategy.

[0009] S4. Use the optimal decision model to predict the business demand problem to be predicted, and output various types of data related to the business demand problem to be predicted.

[0010] S5. Repeat steps S2 to S4 until all types of data related to the business needs to be predicted in the dialogue are obtained, and report them as decision results.

[0011] Preferably, the decision-making model is of three types: machine learning model, statistical model and deep learning model.

[0012] Preferably, step S2, determining the most relevant dataset for the business and its multiple meta-features, is implemented in the following ways:

[0013] The large model used for data discovery is used to search for multiple datasets related to the current business needs problem to be predicted, and the multiple datasets are fused to obtain the most relevant dataset. The meta-features of the most relevant dataset are then extracted.

[0014] Preferably, step S3 and the performance metrics include one or more of the following: model execution speed, model result accuracy, and model generalization ability.

[0015] Preferably, the first implementation of step S3, automatic machine learning strategy tuning to obtain the optimal decision model, is as follows:

[0016] The decision model related to all meta-features is trained using the business demand problem to be predicted and the most relevant business dataset. The business demand problem to be predicted is used as input, and the various types of data related to the business demand problem at each sampling time in the most relevant business dataset are used as output.

[0017] The optimal decision model is selected from all trained decision models based on a given performance metric.

[0018] Preferably, the second implementation of step S3, automatic machine learning strategy tuning to obtain the optimal decision model, is as follows:

[0019] Subsampling is performed on the dataset most relevant to the business to obtain a subsample set;

[0020] The decision model related to all meta-features is trained using the business demand problem to be predicted and the subsample set. The business demand problem to be predicted is used as input, and the various types of data related to the business demand problem at each sampling time in the subsample set are used as output.

[0021] Based on a given performance metric, select the candidate decision model with the best performance from all trained decision models.

[0022] The candidate decision model is trained using the business requirement problem to be predicted and the most relevant business dataset. The trained candidate decision model is used as the optimal decision model. The business requirement problem to be predicted is used as the input, and the various types of data related to the business requirement problem at each sampling time in the most relevant business dataset are used as the output.

[0023] Preferably, the business requirements include:

[0024] ① Market analysis and trend forecasting;

[0025] ② Inventory management and demand forecasting;

[0026] ③ Customer insights and personalization;

[0027] ④ Cost accounting and financial optimization;

[0028] ⑤ Supply chain optimization;

[0029] ⑥ Quality control and defect detection;

[0030] ⑦ Employee performance and training;

[0031] ⑧ Predictive maintenance of equipment;

[0032] ⑨ Assistance with regulatory compliance;

[0033] ⑩ Sales forecasting and revenue optimization.

[0034] Preferably, ① the data in the dataset corresponding to market analysis and trend forecasting includes historical sales data, customer feedback and preference data, and fashion industry trend and forecast data;

[0035] ② The data in the datasets corresponding to inventory management and demand forecasting include sales and inventory data, customer purchase history data, market trend and seasonal pattern data;

[0036] ③ The data within the dataset corresponding to customer insights and personalization includes: customer interaction and feedback data, customer preferences and purchase history data;

[0037] ④ The data within the dataset corresponding to cost accounting and financial optimization includes: production cost data, pricing and revenue data;

[0038] ⑤ The data within the dataset corresponding to supply chain optimization includes: supplier and logistics data, and historical supply chain performance data;

[0039] ⑥ The data in the dataset corresponding to quality control and defect detection includes: manufacturing and quality inspection data;

[0040] ⑦ The data within the dataset corresponding to employee performance and training includes: employee performance data and training plan data;

[0041] ⑧ The dataset corresponding to predictive maintenance of equipment includes: equipment sensor data and maintenance history data;

[0042] ⑨ The dataset corresponding to regulatory compliance assistance includes: legal and regulatory data;

[0043] ⑩ The data in the dataset corresponding to sales forecasting and revenue optimization includes: historical sales data, market and economic data.

[0044] Preferably, step S3, searching for the decision model related to each meta-feature, is implemented as follows:

[0045] The attribute labels of each meta-feature and the decision model corresponding to the business requirement problem to be predicted are mapped to semantic vectors. The similarity between the semantic vector corresponding to each meta-feature and the semantic vector corresponding to each decision model is calculated. The decision model corresponding to the maximum similarity value is determined to be related to the meta-feature.

[0046] The intelligent enterprise decision-making method based on large models further includes step S6, which is used to collect user feedback on the decision results to guide the selection of automatic machine learning strategies in the next dialogue segment.

[0047] The beneficial effects of this invention are as follows:

[0048] This invention, based on a large-scale model, aims to help users (including but not limited to business owners) solve various problems they face in their business operations, such as market analysis and trend forecasting, inventory management and demand forecasting, customer insights and personalization, cost accounting and financial optimization, supply chain optimization, quality control and defect detection, employee performance and training, predictive maintenance of equipment, regulatory compliance assistance, sales forecasting and revenue optimization, and so on.

[0049] This invention presents a large-scale model-based intelligent enterprise decision-making method that comprehensively addresses business problems. The comprehensiveness of this method depends on the dataset collection stage and is related to the comprehensiveness of various business needs addressed locally. It also supports intelligent decision-making, providing more relevant and accurate search results to improve decision-making efficiency and quality. Through automatic machine learning strategy optimization, it enhances the accuracy and efficiency of business-related models, providing more accurate insights and a better user experience. Furthermore, the acquisition, optimization, training, and prediction of the decision-making model upon which this invention relies are all self-constructed. The construction process is simple, time-efficient, and requires no user intervention in building, running, or deploying the model, making it highly user-friendly for computer users. It also boasts high decision-making efficiency, accepting natural language input from users and outputting natural language, thus achieving natural language-based human-computer interaction.

[0050] This invention provides a comprehensive, accurate, and efficient decision-making tool that analyzes market trends, customer feedback and preferences, and industry trends to help users (including but not limited to business owners) better operate their businesses. Furthermore, this invention possesses four key characteristics: interpretability—the decision-making process is verifiable; robustness—it corrects unreasonable business requirements; privacy—locally deployed models protect information security; and generalization—massive pre-trained knowledge enhances decision-making capabilities. Attached Figure Description

[0051] Figure 1 This is a schematic diagram illustrating the principle of the intelligent enterprise decision-making method based on a large model as described in this invention.

[0052] Figure 2 This is a schematic diagram illustrating the industrial value of the intelligent enterprise decision-making method based on a large model as described in this invention. Detailed Implementation

[0053] 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, and 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.

[0054] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0055] See Figure 1 This first embodiment describes the intelligent enterprise decision-making method based on a large model, which includes:

[0056] S1. Collect datasets and decision models: Collect datasets of corresponding attributes for various business requirement problems deployed locally, as well as various decision models for various business requirement problems; each decision model corresponds to a corresponding attribute label, and each dataset includes various types of data related to the business requirement problem at each sampling time;

[0057] Examples of business requirement questions include: ① Market analysis and trend forecasting; ② Inventory management and demand forecasting; ③ Customer insights and personalization; ④ Cost accounting and financial optimization; ⑤ Supply chain optimization; ⑥ Quality control and defect detection; ⑦ Employee performance and training; ⑧ Predictive maintenance of equipment; ⑨ Regulatory compliance assistance; ⑩ Sales forecasting and revenue optimization. In specific applications, business requirement questions should be collected based on local deployment, and are not limited to the above 10 types.

[0058] ① The data related to business needs within the dataset corresponding to market analysis and trend forecasting includes historical sales data, customer feedback and preference data, and fashion industry trend and forecast data;

[0059] ② Data related to business needs within the datasets corresponding to inventory management and demand forecasting include sales and inventory data, customer purchase history data, market trend and seasonal pattern data;

[0060] ③ The data related to business needs within the dataset corresponding to customer insights and personalization includes: customer interaction and feedback data, customer preferences and purchase history data;

[0061] ④ The data related to business needs within the dataset corresponding to cost accounting and financial optimization includes: production cost data, pricing and revenue data;

[0062] ⑤ The data related to business needs within the dataset corresponding to supply chain optimization includes: supplier and logistics data, and historical supply chain performance data;

[0063] ⑥ The data related to business needs within the dataset corresponding to quality control and defect detection includes: manufacturing and quality inspection data;

[0064] ⑦ The data related to business needs within the dataset corresponding to employee performance and training includes: employee performance data and training plan data;

[0065] ⑧ The data related to business needs within the dataset corresponding to predictive maintenance of equipment includes: equipment sensor data and maintenance history data;

[0066] ⑨ The data related to business needs within the dataset corresponding to regulatory compliance assistance includes: legal and regulatory data;

[0067] ⑩ The data related to business needs within the datasets corresponding to sales forecasting and revenue optimization include: historical sales data, market and economic data.

[0068] S2. The large-scale model for data discovery, based on a business requirement question to be predicted in each dialogue segment and the collected dataset of all attributes, determines the most relevant dataset and multiple meta-features of that dataset. This step is to leverage large-scale model data discovery technology to support intelligent decision-making for enterprises. This step uses large-scale model technology to extract meta-features from the dataset, thereby achieving accurate retrieval of business-related datasets. Large-scale model technology is an existing technology; meta-features are used to describe the dataset. A dialogue segment includes multiple questions, and in specific applications, each question is processed sequentially.

[0069] S3. Search for decision models related to each meta-feature, and combine them with the most relevant dataset for the business and the given performance metrics to optimize the optimal decision model using an automatic machine learning strategy; furthermore, the performance metrics include one or more of the following: model execution speed, model result accuracy, and model generalization ability.

[0070] This step is to optimize the business-related model using automated machine learning techniques when the search space formed by the algorithm and its corresponding hyperparameters is very large, based on the user's needs and the business-most relevant dataset and its meta-features obtained in steps S1 and S2 above.

[0071] S4. Use the optimal decision model to predict the business demand problem to be predicted, and output various types of data related to the business demand problem to be predicted.

[0072] S5. Repeat steps S2 to S4 until all types of data related to the business needs to be predicted in the dialogue are obtained, and report them as decision results. Outputting decision results in natural language improves decision-making efficiency and accuracy, and solves a wide variety of real-world business problems.

[0073] Specifically, there are three types of decision-making models: machine learning models, statistical models, and deep learning models.

[0074] Furthermore, step S2, determining the most relevant dataset and its multiple meta-features, involves: using a large model for data discovery to search for multiple datasets related to the current business requirement to be predicted; fusing these datasets to obtain the most relevant dataset; and extracting its meta-features. This process provides more accurate decision-making results by integrating knowledge from different datasets.

[0075] This step involves fusing datasets. By integrating knowledge from different datasets, the information from each dataset can be fully utilized, thereby improving the accuracy and reliability of decision-making. Based on the fused dataset, an effective decision-making model is selected to ensure the efficiency and stability of the entire process. These models include machine learning models, statistical models, and deep learning models. Effective evaluation metrics are used to ensure that the selected decision-making model provides more accurate and reliable decision results.

[0076] The intelligent enterprise decision-making method based on large models further includes step S6, which is used to collect user feedback on the decision results to guide the selection of automatic machine learning strategies in the next dialogue segment.

[0077] Furthermore, step S3, the first implementation of automatically optimizing the machine learning strategy to obtain the optimal decision model, is as follows:

[0078] The decision model related to all meta-features is trained using the business demand problem to be predicted and the most relevant business dataset. The business demand problem to be predicted is used as input, and the various types of data related to the business demand problem at each sampling time in the most relevant business dataset are used as output.

[0079] The optimal decision model is selected from all trained decision models based on a given performance metric.

[0080] Furthermore, step S3, the second implementation of automatically optimizing the machine learning strategy to obtain the optimal decision model, is as follows:

[0081] Subsampling is performed on the dataset most relevant to the business to obtain a subsample set;

[0082] The decision model related to all meta-features is trained using the business demand problem to be predicted and the subsample set. The business demand problem to be predicted is used as input, and the various types of data related to the business demand problem at each sampling time in the subsample set are used as output.

[0083] Based on a given performance metric, select the candidate decision model with the best performance from all trained decision models.

[0084] The candidate decision model is trained using the business requirement problem to be predicted and the most relevant business dataset. The trained candidate decision model is used as the optimal decision model. The business requirement problem to be predicted is used as the input, and the various types of data related to the business requirement problem at each sampling time in the most relevant business dataset are used as the output.

[0085] Furthermore, step S3, searching for the decision model related to each meta-feature, is implemented as follows:

[0086] The attribute labels of each meta-feature and the decision model corresponding to the business requirement problem to be predicted are mapped to semantic vectors. The similarity between the semantic vector corresponding to each meta-feature and the semantic vector corresponding to each decision model is calculated. The decision model corresponding to the maximum similarity value is determined to be related to the meta-feature.

[0087] Implementation Method 2: A computer-readable storage device storing a computer program that, when executed, implements a method for intelligent enterprise decision-making based on a large model.

[0088] Implementation Method 3: A large-model-based intelligent enterprise decision-making system, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, wherein the processor executes the computer program to implement a large-model-based intelligent enterprise decision-making method.

[0089] This invention provides the industry with a comprehensive, accurate, and efficient decision-making tool, helping users solve business problems better, more personally, and with greater privacy; and it enables continuous optimization of product quality through user feedback to provide better customer service, improve customer satisfaction, and enhance customer loyalty. For details of its industrial value, please refer to [link to relevant documentation]. Figure 2 .

[0090] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A method for intelligent enterprise decision making based on large models, characterized in that, The method includes the following steps: S1. Collect datasets of corresponding attributes for various business requirements and problems deployed locally, as well as various decision models for various business requirements and problems; each decision model corresponds to a corresponding attribute label, and each dataset includes various types of data related to the business requirements and problems at each sampling time; S2. Using a large model for data discovery, based on a business requirement question to be predicted in each dialogue segment and the dataset of all collected attributes, determine the most relevant dataset for the business and multiple meta-features of the most relevant dataset for the business. Specifically, the process involves using a large model for data discovery to search for multiple datasets related to the current business needs problem to be predicted, fusing the multiple datasets to obtain the most relevant dataset, and extracting the meta-features of the most relevant dataset. S3. Search for decision models related to each meta-feature, and combine them with the most relevant dataset for the business and the given performance metrics to optimize the optimal decision model using an automatic machine learning strategy. The implementation method for searching decision models related to each meta-feature is as follows: The attribute labels of each meta-feature and the decision model corresponding to the business requirement problem to be predicted are mapped to semantic vectors. The similarity between the semantic vector corresponding to each meta-feature and the semantic vector corresponding to each decision model is calculated. The decision model corresponding to the maximum similarity value is determined to be related to the meta-feature. S4. Use the optimal decision model to predict the business demand problem to be predicted, and output various types of data related to the business demand problem to be predicted. S5. Repeat steps S2 to S4 until all types of data related to the business needs to be predicted in the dialogue are obtained, and report them as decision results.

2. The large model based intelligent enterprise decision making method of claim 1, wherein, There are three types of decision-making models: machine learning models, statistical models, and deep learning models.

3. The large model based intelligent enterprise decision making method of claim 1, wherein, It also includes step S6, which collects user feedback on the decision results to guide the selection of automatic machine learning strategies in the next dialogue segment.

4. The large model based intelligent enterprise decision making method of claim 1, wherein, Step S3: Performance metrics include one or more of the following: model execution speed, model result accuracy, and model generalization ability.

5. The large model based intelligent enterprise decision making method of claim 1, wherein, Step S3, the first implementation of automatically optimizing the machine learning strategy to obtain the optimal decision model, is as follows: The decision model related to all meta-features is trained using the business demand problem to be predicted and the most relevant business dataset. The business demand problem to be predicted is used as input, and the various types of data related to the business demand problem at each sampling time in the most relevant business dataset are used as output. The optimal decision model is selected from all trained decision models based on a given performance metric.

6. The large model based intelligent enterprise decision making method of claim 1, wherein, Step S3, the second implementation of automatically optimizing the machine learning strategy to obtain the optimal decision model, is as follows: Subsampling is performed on the dataset most relevant to the business to obtain a subsample set; The decision model related to all meta-features is trained using the business demand problem to be predicted and the subsample set. The business demand problem to be predicted is used as input, and the various types of data related to the business demand problem at each sampling time in the subsample set are used as output. Based on a given performance metric, select the candidate decision model with the best performance from all trained decision models. The candidate decision model is trained using the business requirement problem to be predicted and the most relevant business dataset. The trained candidate decision model is used as the optimal decision model. The business requirement problem to be predicted is used as the input, and the various types of data related to the business requirement problem at each sampling time in the most relevant business dataset are used as the output.

7. The intelligent enterprise decision-making method based on a large model according to claim 1, characterized in that, Business requirements include: ① Market analysis and trend forecasting; ② Inventory management and demand forecasting; ③ Customer insights and personalization; ④ Cost accounting and financial optimization; ⑤ Supply chain optimization; ⑥ Quality control and defect detection; ⑦ Employee performance and training; ⑧ Predictive maintenance of equipment; ⑨ Assistance with regulatory compliance; ⑩ Sales forecasting and revenue optimization.

8. The intelligent enterprise decision-making method based on a large model according to claim 7, characterized in that, ① The data in the dataset corresponding to market analysis and trend forecasting includes historical sales data, customer feedback and preference data, and fashion industry trend and forecast data; ② The data in the datasets corresponding to inventory management and demand forecasting include sales and inventory data, customer purchase history data, market trend and seasonal pattern data; ③ The data within the dataset corresponding to customer insights and personalization includes: customer interaction and feedback data, customer preferences and purchase history data; ④ The data within the dataset corresponding to cost accounting and financial optimization includes: production cost data, pricing and revenue data; ⑤ The data within the dataset corresponding to supply chain optimization includes: supplier and logistics data, and historical supply chain performance data; ⑥ The data in the dataset corresponding to quality control and defect detection includes: manufacturing and quality inspection data; ⑦ The data within the dataset corresponding to employee performance and training includes: employee performance data and training plan data; ⑧ The dataset corresponding to predictive maintenance of equipment includes: equipment sensor data and maintenance history data; ⑨ The dataset corresponding to regulatory compliance assistance includes: legal and regulatory data; ⑩ The data in the dataset corresponding to sales forecasting and revenue optimization includes: historical sales data, market and economic data.

Images