Object evaluation method, model training method, device, equipment, storage medium and program product

Abstract

The invention provides an object evaluation method and device, a model training method and device, equipment, a storage medium and a program product, relates to the technical field of financial services, and is used for efficiently and accurately evaluating the green degree of an object. According to the specific technical scheme, index information of a to-be-evaluated object in multiple dimensions is obtained; the plurality of dimensions comprise an environment dimension, a social dimension and a governance dimension; processing the index information of the multiple dimensions based on a pre-trained random forest model to obtain an evaluation result of the to-be-evaluated object; the pre-trained random forest model comprises at least one target decision tree corresponding to each dimension in a plurality of dimensions; the at least one target decision tree corresponding to each dimension is constructed through the following process: aiming at each training set corresponding to each dimension, taking each training set as a root node, and splitting according to the target information gain of each sub-dimension included in each training set to obtain the target decision tree corresponding to each training set, and obtaining at least one target decision tree.

Description

Technical Field

[0001] This application relates to the field of financial services technology, and in particular to an object evaluation method, a model training method, an apparatus, a device, a storage medium, and a program product. Background Technology

[0002] To better support the development of green finance, it is necessary to evaluate the greenness of entities (such as enterprises). Related technologies primarily use manually developed scorecards to score various indicators related to the greenness of the entity, thereby deriving a comprehensive score to evaluate its greenness. This method, relying on manual evaluation, is prone to leading to overly subjective results and is inefficient. Summary of the Invention

[0003] This application provides an object evaluation method, a model training method, an apparatus, a device, a storage medium, and a program product for efficiently and accurately evaluating the greenness of objects.

[0004] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: In a first aspect, embodiments of this application provide an object evaluation method, comprising: acquiring indicator information of multiple dimensions of the object to be evaluated; the multiple dimensions include environmental, social, and governance dimensions; processing the indicator information of the multiple dimensions based on a pre-trained random forest model to obtain an evaluation result of the object to be evaluated; the evaluation result is used to indicate the greenness of the object to be evaluated; the pre-trained random forest model includes at least one target decision tree corresponding to each of the multiple dimensions; the at least one target decision tree corresponding to each dimension is constructed through the following process: for each training set corresponding to each dimension, taking each training set as the root node, splitting according to the target information gain of each sub-dimension included in each training set to obtain a target decision tree corresponding to each training set, thereby obtaining at least one target decision tree; wherein, the target information gain of the sub-dimension is the product of the original information gain of the indicator information of the sub-dimension and the adjusted weight corresponding to the sub-dimension; the adjusted weight is used to represent the importance of the sub-dimension to the greenness of the evaluated object.

[0005] Based on this, this application, combining the characteristics of greenness evaluation indicators, can obtain relevant information about the object to be evaluated according to three dimensions: environment, society, and governance, thereby improving the accuracy of the greenness evaluation. Furthermore, this application can efficiently evaluate the greenness of the object through pre-trained random forest models using target decision trees for the three dimensions of environment, society, and governance, improving the accuracy of the evaluation and avoiding the tedious and time-consuming problems associated with manual methods. Simultaneously, when constructing the target decision tree for each dimension, the target information gain of the sub-dimensions can be referenced to reflect the importance of the sub-dimensions to the greenness of the evaluated enterprise, avoiding the disadvantage of using the original information gain of the sub-dimensions for splitting, which can easily lead to bias towards sub-dimensions with more values, thus improving the accuracy of the random forest model. Therefore, this application can be used to efficiently and accurately evaluate the greenness of objects.

[0006] In one possible implementation, the environmental dimension includes at least one of the following sub-dimensions: climate change, natural capital, waste treatment, and environmental opportunities. The indicators for the climate change sub-dimension include at least one of the following: carbon emissions, carbon emission intensity, carbon emission reduction, and the proportion of green energy. The indicators for the natural capital sub-dimension include at least one of the following: resource consumption intensity, ecological protection investment, and land use efficiency. The indicators for the waste treatment sub-dimension include at least one of the following: compliance rate of pollutant emission concentrations, wastewater treatment compliance rate, and solid waste recycling rate. The indicators for the environmental opportunities sub-dimension include at least one of the following: the proportion of green business and the proportion of green R&D investment. The social dimension includes at least one of the following sub-dimensions: human capital, product responsibility, stakeholders, and social opportunities. The indicators for the human capital sub-dimension include at least one of the following: employee compensation competitiveness. The indicators for the environmental, social, and governance dimensions include: average employee training time, workplace accident rate, and employee satisfaction; product responsibility indicators include at least one of the following: quality inspection pass rate, product recall rate, customer complaint handling timeliness, and proportion of recyclable production materials; stakeholder indicators include at least one of the following: number of stakeholder objections, supplier ESG compliance rate, and number of negative events; social opportunity indicators include at least one of the following: number of people served in public welfare projects, number of jobs created, and proportion of inclusive business; governance indicators include at least one of the following: target governance and target behavior; target governance indicators include at least one of the following: reasonableness of compensation, equity concentration, proportion of independent directors, and type of audit opinion; target behavior indicators include at least one of the following: number of abnormal events per year, number of abnormal competition events, and proportion of material donations. Thus, this application can obtain multiple indicators consistent with greenness evaluation according to the three dimensions of environment, society, and governance, improving the accuracy of the target's greenness evaluation.

[0007] In one possible implementation, the random forest model is trained through the following process: acquiring multiple sample information; each sample information includes multiple dimensions of indicator information for a sample object, as well as the greenness label of the sample object; training a random forest based on the multiple sample information to obtain the random forest model; wherein, the indicator information for each of the multiple dimensions is trained separately using a random forest. Based on this, this application can select and train samples according to the three dimensions of environment, society, and governance, thereby improving the accuracy of the algorithm in evaluating the greenness of objects.

[0008] In one possible implementation, a random forest model is obtained by training a random forest based on multiple sample information. This includes: performing the following process for the indicator information of each of the multiple dimensions to obtain at least one target decision tree corresponding to each dimension: sampling the indicator information of each dimension with replacement to obtain at least one training set corresponding to each dimension; the training set includes the indicator information of at least one sample object in each sub-dimension of the dimension, and the greenness label of the sample object; training a random forest based on each training set to obtain at least one target decision tree corresponding to each dimension; wherein the training of the target decision trees for multiple dimensions is performed in parallel; and voting weights are configured for the at least one target decision tree corresponding to each dimension to generate the random forest model. Thus, during the training of the random forest model, the decision trees of each dimension can be computed in parallel and evaluated according to weighted voting, resulting in high computational speed and efficiency. Furthermore, the parallel execution of the training of the target decision trees for multiple dimensions improves training efficiency and avoids the high time consumption caused by serial training when the number of sample information is large.

[0009] In one possible implementation, a random forest is trained on each training set to obtain at least one target decision tree corresponding to each dimension. This includes: for each training set, using the training set as the root node, splitting it according to the target information gain of each sub-dimension included in the training set to obtain a target decision tree corresponding to the training set, thus obtaining at least one target decision tree. The target information gain of each sub-dimension is the product of the original information gain of the sub-dimension's indicator information and the adjusted weight corresponding to the sub-dimension. The adjusted weight is used to represent the importance of the sub-dimension to the greenness of the evaluation object. In this way, the target information gain of the sub-dimension can reflect the importance of the sub-dimension to the greenness of the evaluation object, avoiding the disadvantage of using the original information gain of the sub-dimension for splitting, which can easily lead to bias towards sub-dimensions with more values, thus improving the accuracy of the random forest model.

[0010] In one possible implementation, the training set is used as the root node, and the target decision tree is obtained by splitting the training set according to the target information gain of each sub-dimension. This includes: dividing the training set into multiple subsets; determining the sub-dimension with the largest target information gain in each subset as the target sub-dimension for each subset; and splitting the training set according to the target sub-dimension with the largest target information gain among the target sub-dimensions of each subset to obtain the target decision tree for each training set. This improves the speed of node splitting during training, achieving efficient construction of the decision tree.

[0011] Secondly, embodiments of this application provide a model training method, comprising: acquiring multiple sample information; each sample information includes indicator information of a sample object across multiple dimensions, and a greenness label of the sample object; the multiple dimensions include environmental, social, and governance dimensions; performing random forest training based on the multiple sample information to obtain a random forest model; wherein, the indicator information of each dimension in the multiple dimensions is respectively trained using random forest to obtain at least one target decision tree corresponding to each dimension in the multiple dimensions; the at least one target decision tree corresponding to each dimension is constructed through the following process: for each training set corresponding to each dimension, taking each training set as the root node, splitting according to the target information gain of each sub-dimension included in each training set to obtain a target decision tree corresponding to each training set, thereby obtaining at least one target decision tree; wherein, the target information gain of the sub-dimension is the product of the original information gain of the indicator information of the sub-dimension and the adjusted weight corresponding to the sub-dimension; the adjusted weight is used to represent the importance of the sub-dimension to the greenness of the evaluation object.

[0012] In one possible implementation, the environmental dimension includes at least one of the following sub-dimensions: climate change, natural capital, waste treatment, and environmental opportunities. The indicators for the climate change sub-dimension include at least one of the following: carbon emissions, carbon emission intensity, carbon emission reduction, and the proportion of green energy. The indicators for the natural capital sub-dimension include at least one of the following: resource consumption intensity, ecological protection investment, and land use efficiency. The indicators for the waste treatment sub-dimension include at least one of the following: compliance rate of pollutant emission concentrations, wastewater treatment compliance rate, and solid waste recycling rate. The indicators for the environmental opportunities sub-dimension include at least one of the following: the proportion of green business and the proportion of green R&D investment. The social dimension includes at least one of the following sub-dimensions: human capital, product responsibility, stakeholders, and social opportunities. The indicators for the human capital sub-dimension include at least one of the following: employee compensation competitiveness. The indicators for the product responsibility sub-dimension include at least one of the following: average employee training time, workplace accident rate, and employee satisfaction; the product responsibility sub-dimension includes at least one of the following: quality inspection pass rate, product recall rate, customer complaint handling timeliness, and proportion of recyclable production materials; the stakeholder sub-dimension includes at least one of the following: number of stakeholder objections, supplier ESG compliance rate, and number of negative events; the social opportunity sub-dimension includes at least one of the following: number of people served in public welfare projects, number of jobs created, and proportion of inclusive business; the governance dimension includes at least one of the following sub-dimensions: target governance and target behavior; the target governance sub-dimension includes at least one of the following: reasonableness of compensation, equity concentration, proportion of independent directors, and type of audit opinion; the target behavior sub-dimension includes at least one of the following: number of abnormal events per year, number of abnormal competition events, and proportion of material donations.

[0013] In one possible implementation, a random forest model is obtained by training a random forest based on multiple sample information. This includes: performing the following process for the indicator information of each of the multiple dimensions to train the random forest and obtain at least one target decision tree corresponding to the multiple dimensions: sampling the indicator information of the dimensions in the multiple sample information with replacement to obtain at least one training set corresponding to the dimension; the training set includes the indicator information of at least one sample object in each sub-dimension of the dimension, and the greenness label of the sample object; training a random forest based on each training set to obtain at least one target decision tree corresponding to the dimension; wherein the training of the target decision trees for the multiple dimensions is performed in parallel; and configuring voting weights for the at least one target decision tree corresponding to each of the multiple dimensions to generate the random forest model.

[0014] In one possible implementation, random forest training is performed on each training set to obtain at least one target decision tree corresponding to a dimension. This includes: for each training set, using the training set as the root node, splitting it according to the target information gain of each sub-dimension included in the training set to obtain the target decision tree corresponding to the training set, thereby obtaining at least one target decision tree; wherein, the target information gain of the sub-dimension is the product of the original information gain of the indicator information of the sub-dimension and the adjusted weight corresponding to the sub-dimension; the adjusted weight is used to represent the importance of the sub-dimension to the greenness of the evaluation object.

[0015] In one possible implementation, the training set is used as the root node, and the training set is split according to the target information gain of each sub-dimension included in the training set to obtain the target decision tree corresponding to the training set. This includes: dividing the training set to obtain multiple subsets; determining the sub-dimension with the largest target information gain in each subset as the target sub-dimension of each subset; and splitting the training set according to the target sub-dimension with the largest target information gain in each subset to obtain the target decision tree corresponding to each training set.

[0016] Thirdly, embodiments of this application provide an object evaluation device, comprising: an acquisition unit for acquiring indicator information of multiple dimensions of the object to be evaluated; the multiple dimensions include environmental, social, and governance dimensions; and a processing unit for processing the indicator information of the multiple dimensions based on a pre-trained random forest model to obtain an evaluation result of the object to be evaluated; the evaluation result is used to indicate the greenness of the object to be evaluated; the pre-trained random forest model includes at least one target decision tree corresponding to each of the multiple dimensions; the at least one target decision tree corresponding to each dimension is constructed through the following process: for each training set corresponding to each dimension, taking each training set as the root node, splitting according to the target information gain of each sub-dimension included in each training set to obtain a target decision tree corresponding to each training set, thereby obtaining at least one target decision tree; wherein, the target information gain of the sub-dimension is the product of the original information gain of the indicator information of the sub-dimension and the adjustment weight corresponding to the sub-dimension; the adjustment weight is used to represent the importance of the sub-dimension to the greenness of the evaluated object.

[0017] Fourthly, embodiments of this application provide a model training apparatus, comprising: an acquisition unit for acquiring multiple sample information; each sample information includes indicator information of multiple dimensions of a sample object, and a greenness label of the sample object; the multiple dimensions include environmental dimension, social dimension, and governance dimension; and a processing unit for performing random forest training based on the multiple sample information to obtain a random forest model; wherein, the indicator information of each dimension in the multiple dimensions is respectively trained using random forest to obtain at least one target decision tree corresponding to each dimension in the multiple dimensions; the at least one target decision tree corresponding to each dimension is constructed through the following process: for each training set corresponding to each dimension, taking each training set as the root node, splitting according to the target information gain of each sub-dimension included in each training set to obtain a target decision tree corresponding to each training set, thereby obtaining at least one target decision tree; wherein, the target information gain of the sub-dimension is the product of the original information gain of the indicator information of the sub-dimension and the adjusted weight corresponding to the sub-dimension; the adjusted weight is used to represent the importance of the sub-dimension to the greenness of the evaluation object.

[0018] Fifthly, embodiments of this application provide an electronic device, the method comprising: a memory and at least one processor. The memory is communicatively connected to the processor. The memory is used to store computer program code, the computer program code including computer instructions. When the processor executes the computer instructions, it causes the electronic device to perform the method as described in the first aspect and any possible implementation thereof, or causes the electronic device to perform the method as described in the second aspect and any possible implementation thereof.

[0019] Sixthly, embodiments of this application provide a computer-readable storage medium storing computer instructions. When executed by a processor, the computer instructions are used to implement the method as described in the first aspect and any possible implementation thereof, or to implement the method as described in the second aspect and any possible implementation thereof.

[0020] In a seventh aspect, embodiments of this application provide a computer program product that, when running on a computer / executed by a computer's processor, implements the method described in the first aspect and any possible design thereof, or implements the method described in the second aspect and any possible design thereof. The computer may be the electronic device described in the fifth aspect and any possible implementation thereof.

[0021] Understandably, the beneficial effects that can be achieved by the model training method of the second aspect, the object evaluation device and model training device of the third and fourth aspects, the electronic device of the fifth aspect, the computer-readable storage medium of the sixth aspect, and the computer program product of the seventh aspect can be referred to as the beneficial effects of the first aspect and any possible implementation thereof, and will not be repeated here. Attached Figure Description

[0022] Figure 1 A flowchart illustrating an object evaluation method provided in an embodiment of this application; Figure 2 A schematic flowchart illustrating a model training method provided in an embodiment of this application; Figure 3 A flowchart illustrating another model training method provided in an embodiment of this application; Figure 4 A flowchart illustrating another model training method provided in an embodiment of this application; Figure 5 A schematic diagram of a decision tree construction process provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an object evaluation device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a model training device provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0023] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0025] The technical solutions provided in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of financial data or object data, comply with relevant laws and regulations and do not violate public order and good morals.

[0026] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0027] First, a brief introduction to the application scenarios involved in the technical solution of this application will be given.

[0028] To better support the development of green finance, it is necessary to evaluate the greenness of entities. The relevant technologies primarily use manually developed scorecards to score various indicators related to the greenness of entities, thereby deriving a comprehensive score to evaluate their greenness. However, this method, which relies on manual evaluation, is prone to leading to overly subjective results and is inefficient.

[0029] To address the aforementioned issues, this application provides an object evaluation method that can acquire indicator information from multiple dimensions of the object to be evaluated. These multiple dimensions include environmental, social, and governance dimensions. Furthermore, the indicator information from these multiple dimensions can be processed using a pre-trained random forest model to obtain the evaluation result of the object, namely, its greenness. Based on this, this application, combining the characteristics of greenness evaluation indicators, can acquire relevant information about the object to be evaluated according to the three dimensions of environment, society, and governance, thereby improving the accuracy of the object's greenness evaluation. Moreover, this application can efficiently achieve object greenness evaluation through a pre-trained random forest model, avoiding the tedious and time-consuming problems associated with manual evaluation. Therefore, this application can be used to efficiently and accurately evaluate the greenness of objects.

[0030] Next, a brief introduction will be given to the implementation environment (implementation architecture) involved in this application.

[0031] The object evaluation method provided in this application can be applied to electronic devices. Optionally, the electronic device can be a touch screen all-in-one machine, a smart self-service terminal, a tablet computer, a laptop computer, or a desktop computer, or it can be a remote terminal, user terminal equipment (TE), or a mobile device. The electronic device can also be a server. This application does not limit the specific form of the electronic device. In addition, in the embodiments of this application, the electronic device used to train the random forest model and the electronic device used to evaluate the greenness of objects using the random forest model can be the same electronic device or different electronic devices; this application does not impose specific limitations here.

[0032] For ease of understanding, the object evaluation method provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0033] like Figure 1 The diagram shown is a flowchart illustrating an object evaluation method provided in an embodiment of this application. Figure 1 The object evaluation method shown can be applied to the aforementioned electronic device. The method includes: S101-S102.

[0034] S101. Obtain indicator information for multiple dimensions of the object to be evaluated.

[0035] The entity to be evaluated refers to any entity requiring a greenness assessment. The greenness assessment results of the entity can be used to determine financial service strategies related to it. Multiple dimensions can be included, such as environmental, social, and governance dimensions. The indicator information from these multiple dimensions can be used to quantitatively reflect the entity's performance across each of the environmental, social, and governance (ESG) dimensions.

[0036] Information on multiple dimensions of the entity being evaluated can be extracted from publicly available ESG reports, annual financial reports, and social responsibility reports on the entity's website, or from publicly available ESG rating data, regulatory data, and statistical data corresponding to the entity on the websites of third-party organizations. Alternatively, staff can collect this information in advance. In this case, the pre-collected information can be retrieved from a storage location specified by the staff.

[0037] The environmental dimension includes at least one of the following sub-dimensions: climate change, natural capital, waste management, and environmental opportunities. The indicators for the climate change sub-dimension include at least one of the following: carbon emissions, carbon emission intensity, carbon emission reductions, and the proportion of green energy. The indicators for the natural capital sub-dimension include at least one of the following: resource consumption intensity, ecological protection investment, and land use efficiency. The indicators for the waste management sub-dimension include at least one of the following: compliance rate of pollutant emission concentrations, wastewater treatment compliance rate, and solid waste recycling rate. The indicators for the environmental opportunities sub-dimension include at least one of the following: the proportion of green business and the proportion of green R&D investment.

[0038] The social dimension includes at least one of the following sub-dimensions: human capital, product responsibility, stakeholders, and social opportunities. The indicators for the human capital sub-dimension include at least one of the following: employee compensation competitiveness, average employee training duration, workplace accident rate, and employee satisfaction. The indicators for the product responsibility sub-dimension include at least one of the following: quality inspection pass rate, product recall rate, customer complaint handling timeliness, and the proportion of recyclable production materials. The indicators for the stakeholders sub-dimension include at least one of the following: number of stakeholder objections, supplier ESG compliance rate, and number of negative events. The indicators for the social opportunities sub-dimension include at least one of the following: number of people served by public welfare projects, number of jobs created, and the proportion of inclusive services.

[0039] The governance dimension includes at least one of the following sub-dimensions: object governance and object behavior. The indicators for the object governance sub-dimension include at least one of the following: reasonableness of compensation, equity concentration, percentage of independent directors, and type of audit opinion. The indicators for the object behavior sub-dimension include at least one of the following: number of abnormal events per year, number of abnormal competition events, and percentage of material donations.

[0040] S102. Based on the pre-trained random forest model, the indicator information of multiple dimensions is processed to obtain the evaluation results of the object to be evaluated.

[0041] The random forest model can be pre-trained based on sample information from multiple sample objects. It can process indicator information across multiple dimensions separately to obtain a greenness classification result for each dimension. Furthermore, the greenness classification results for each dimension can be weighted and summed to obtain the evaluation result for the object being evaluated.

[0042] The evaluation results of the object being evaluated can be used to indicate its greenness. The greenness of an object refers to the degree of environmental impact of its production and operation activities, and is used to quantitatively reflect the object's degree of greening and sustainable development capabilities.

[0043] As an example, greenness can be divided into six levels: dark green, medium green, light green, yellow, orange, and red. The degree of greenness of the objects indicated by dark green, medium green, light green, yellow, orange, and red decreases in that order, or in other words, the lower the degree of environmental impact of the production and business activities of the indicated objects.

[0044] In one embodiment, in order to train the above-mentioned random forest model, such as Figure 2 The diagram shown is a flowchart of a model training method provided in an embodiment of this application. Figure 2 The model training methods shown include: S201-S202.

[0045] S201. Obtain information from multiple samples.

[0046] One sample information includes multiple dimensions of indicator information for a sample object, as well as the greenness label of the sample object.

[0047] The multiple sample objects corresponding to multiple sample information can be objects from the same domain. Furthermore, the object to be evaluated and these multiple sample objects can also be from the same domain, so that the random forest model can accurately learn knowledge from that domain and thus accurately evaluate the greenness of the object to be evaluated. Alternatively, the multiple sample objects corresponding to multiple sample information can be objects from different domains, such as objects from different domains including manufacturing, energy, and services, so that the random forest model can simultaneously learn knowledge from different domains, and the object to be evaluated can be an object from any of these different domains.

[0048] The greenness label of a sample object can be obtained through evaluation using industry-recognized standards. For example, the greenness label of a sample object can be dark green, medium green, light green, yellow, orange, or red.

[0049] Alternatively, preprocessing can be performed on multiple sample information, such as data cleaning and default value imputation, for example, using default values ​​to fill in missing data. Furthermore, data values ​​in multiple sample information can be type-converted, such as converting strings to integers or floating-point numbers, to ensure that the data in the sample information conforms to the type requirements of the training input.

[0050] S202. Random forest training is performed based on information from multiple samples to obtain a random forest model.

[0051] In this process, the indicator information for each of the multiple dimensions is used to train a random forest.

[0052] As an example, multiple sample information can be divided into training and test sets according to a preset ratio. The training set is used for training and tuning the random forest model, such as for parameter learning. The test set is used to evaluate the generalization ability of the random forest model, such as for verifying its accuracy. For example, 80% of the sample information can be allocated to the training set and 20% to the test set. Furthermore, bootstrap resampling (sampling with replacement) can be performed on the training set for each dimension to obtain N training sets corresponding to each dimension. The sample information in the training set corresponding to a dimension includes the indicator information for that dimension and the greenness label of the sample object. N is a positive integer. It is understandable that the training sets obtained for each dimension are independent of each other.

[0053] Furthermore, for each dimension, N decision trees can be constructed from the N training sets corresponding to each dimension. When constructing a decision tree for a dimension, at each node of the decision tree, the sub-dimension with the highest discriminative power can be selected for splitting, resulting in child nodes, until the decision tree reaches its maximum depth or the number of node samples is less than a preset threshold. In this way, the N decision trees corresponding to a dimension can constitute a random forest sub-model for that dimension. The 3N decision trees corresponding to the three dimensions of environment, society, and governance can constitute a random forest model. The 3N decision trees can determine the evaluation result of the input through weighted voting. Furthermore, the accuracy of the random forest model can be calculated using a test set to ensure that the random forest model has superior predictive performance. Moreover, the sample information obtained from the accurate results calculated based on the test set can be used to form a training set again, and the random forest model can be retrained based on this training set to further improve the accuracy of the random forest model.

[0054] Thus, this application can automatically complete the greenness evaluation of objects by constructing an object greenness evaluation model based on the random forest algorithm, avoiding the tedious and time-consuming problems that easily arise from relying on manual methods. Furthermore, this application combines the characteristics of greenness evaluation indicators and selects and trains samples according to three dimensions: environment, society, and governance, which can improve the accuracy of the algorithm in evaluating the greenness of objects. Moreover, when the random forest model processes the input, the decision trees of each dimension can be computed in parallel and the evaluation result is obtained through weighted voting, exhibiting high computational speed and efficiency.

[0055] In one embodiment, such as Figure 3 The diagram shown is a flowchart illustrating another model training method provided in this application embodiment. In S202 above, when a random forest model is obtained by training based on multiple sample information, this application embodiment provides an optional implementation, including: S2021-S2022.

[0056] S2021. For the indicator information of each of the multiple dimensions, perform the following process to achieve random forest training, so as to obtain at least one target decision tree corresponding to the multiple dimensions: sample the indicator information of the dimension with replacement from the multiple sample information to obtain at least one training set corresponding to the dimension, and perform random forest training based on each training set to obtain at least one target decision tree corresponding to the dimension.

[0057] In each dimension, there is at least one training set, and each training set includes at least one sample object's indicator information for each sub-dimension under that dimension, as well as the greenness label of the sample object.

[0058] Furthermore, the training of target decision trees for multiple dimensions is performed in parallel, improving training efficiency and avoiding the problem of high time consumption caused by serial training when the amount of sample information is large.

[0059] As an example, suppose there are 100 sample information entries, and 50 training sets need to be built corresponding to the environment dimension, with each training set containing 30 samples. Then, 50 samplings with replacement can be performed on the 100 sample information entries, with 30 samples drawn each time. The environment dimension indicator information and greenness label included in each of these 30 samples are considered as one sample, resulting in one training set for the environment dimension. By performing 50 samplings with replacement, 50 training sets corresponding to the environment dimension can be obtained. Each training set is then trained using a random forest to obtain a target decision tree for the environment dimension. In this way, a total of 50 target decision trees corresponding to the environment dimension can be trained.

[0060] S2022. Configure voting weights for at least one objective decision tree corresponding to each of the multiple dimensions to generate a random forest model.

[0061] The target decision tree, consisting of child nodes and leaf nodes, is a hierarchical tree-shaped classification model. The child nodes of the target decision tree indicate the decision criteria for splitting. The leaf nodes indicate the predicted classification result.

[0062] The basic building block of the trained random forest model is the objective decision tree, which includes at least one objective decision tree for each of the multiple dimensions.

[0063] The input to a random forest model is information about the object to be evaluated, and the output is the greenness of the object. Optionally, a random forest model may include an input layer, a decision layer, and a prediction output layer. The input layer receives information about the object to be evaluated and identifies the indicator information for each dimension within that information, further transmitting the indicator information for each dimension to the corresponding target decision tree in the decision layer. The decision layer includes target decision trees for each dimension. A target decision tree for one dimension receives the indicator information for that dimension, makes predictions based on the received indicator information, and further transmits the prediction results to the prediction output layer. The prediction output layer receives the prediction results from each target decision tree in the decision layer, performs a summary calculation, and outputs the greenness label with the highest confidence level (i.e., the highest score after summarization).

[0064] Based on this, in configuring voting weights for at least one target decision tree corresponding to each of multiple dimensions, a preset voting weight for each dimension can be obtained. Then, according to the preset voting weight for each dimension and the number of target decision trees for each dimension, the voting weight of the target decision tree for each dimension is configured, generating a random forest model. Specifically, the target voting weight for each dimension can be obtained by dividing the preset voting weight for each dimension by the number of target decision trees for each dimension. Furthermore, the target voting weight for each dimension can be configured in the leaf nodes of the target decision trees for each dimension. Thus, target decision trees for the same dimension are configured with the same voting weight. The prediction result obtained from the target decision tree of a dimension is the product of the greenness label indicated by the leaf node and the target voting weight.

[0065] As an example, suppose that in a random forest model, the weights of the three dimensions of environment, society and governance when voting are weighted can be 0.5, 0.3 and 0.2 respectively, and each dimension contains 50 objective decision trees.

[0066] In this scenario, the voting weight for each target decision tree corresponding to the environment dimension can be set to 0.01 (i.e., 0.5 divided by 50). That is, a weight of 0.01 is configured in the output layer (leaf node) of each tree-based classification model corresponding to the environment dimension, multiplying the predicted classification result by this weight before output. Similarly, the voting weight for each target decision tree corresponding to the social dimension can be set to 0.006 (i.e., 0.3 divided by 50). The voting weight for each target decision tree corresponding to the governance dimension can be set to 0.004 (i.e., 0.2 divided by 50). Furthermore, the output results of each target decision tree can be aggregated and calculated, and the output layer logic that outputs the greenness label with the highest confidence (i.e., the highest score after aggregation) can be generated to produce a random forest model.

[0067] As an example, in the process of evaluating the greenness of the object to be evaluated based on the generated random forest model, the information of the input object to be evaluated can be calculated in parallel based on 150 target decision trees in the random forest model, and prediction result queues for three dimensions: environmental dimension, social dimension and governance dimension are established respectively. The prediction results obtained by the target decision trees of each dimension are included in the queue of each dimension, so as to further extract the prediction results from the queues of each dimension in order and calculate them in a weighted form.

[0068] Assuming that for the environmental dimension, 30 of the 50 decision trees predict a "dark green" and 20 predict a "medium green" outcome for the evaluated object, then the overall voting result for the environmental dimension is: 30 × 0.01 = 0.3 for dark green and 20 × 0.01 = 0.2 for medium green. For the social dimension, 40 of the 50 decision trees predict a "medium green" and 10 predict a "light green" outcome, then the overall voting result for the social dimension is: 40 × 0.006 = 0.24 for medium green and 10 × 0.006 = 0.06 for light green. For the governance dimension, 25 of the 50 decision trees predict a "medium green" and 25 predict a "light green" outcome for the evaluated object. In the overall voting results for the governance dimension, the score for medium green is 25 × 0.004 = 0.1, and the score for light green is 25 × 0.004 = 0.1. Furthermore, after weighted summation of all dimensions, the score for dark green is 0.3, the score for medium green is 0.54, the score for light green is 0.16, and the scores for other categories (yellow, orange, red) are 0. Therefore, the greenness classification of the object to be evaluated is medium green.

[0069] In one embodiment, when performing random forest training based on each training set to obtain at least one target decision tree corresponding to that dimension, this application provides an optional implementation method, including: S301.

[0070] S301. For each training set, take the training set as the root node and split it according to the target information gain of each sub-dimension included in the training set to obtain the target decision tree corresponding to the training set, so as to obtain at least one target decision tree.

[0071] The target information gain of a sub-dimension is the product of the original information gain of the sub-dimension's indicator information and the corresponding adjusted weight of the sub-dimension.

[0072] The original information gain of the sub-dimension's indicator information can be calculated using the following information gain formula.

[0073] Information gain formula: .

[0074] in, Let be the nth sub-dimension in the training set T. Let T be the information entropy of the training set T. The number of samples in the training set T. For sub-dimensions Information gain when splitting a node. (Based on sub-dimensions) When splitting a node, it can be based on the sub-dimensions of each sample in the training set. The value of is used to divide each sample in the training set T into . , … There are m subsets in total. After partitioning, the sub-dimensions within the same subset are... The values ​​meet the same conditions, such as all being greater than the set threshold, all being less than the set threshold, or all being within the set value range. Let i be the i-th subset among m subsets. For subset Information entropy. For subset The number of samples in the sample.

[0075] Information entropy formula: .

[0076] Where s is the number of green color label categories in the training set T. This is the j-th greenness label. for Frequency of occurrence in training set T.

[0077] Similarly, It can be calculated based on the information entropy formula.

[0078] Adjusting weights is used to represent the importance of sub-dimensions to the greenness of the evaluated object. For example, the importance index value of the greenness of the evaluated object corresponding to each sub-dimension can be predefined. In this way, the sub-dimensions can be sorted in descending order according to their importance index values, and weights can be assigned to each sub-dimension in descending order according to an arithmetic sequence or exponential form.

[0079] In this way, the target information gain of the sub-dimension can reflect the importance of the sub-dimension to the greenness of the evaluation object, avoiding the disadvantage of using the original information gain of the sub-dimension for splitting, which can easily lead to bias towards the sub-dimension with more values, and improving the accuracy of the random forest model.

[0080] In one embodiment, such as Figure 4The diagram shown is a flowchart illustrating another model training method provided in this application embodiment. When using the training set as the root node and splitting it according to the target information gain of each sub-dimension included in the training set to obtain the target decision tree corresponding to the training set, this application embodiment provides an optional implementation method, including: S401-S403.

[0081] S401. Divide the training set into multiple subsets.

[0082] S402. Determine the sub-dimension with the largest target information gain in each subset, and define the target sub-dimension for each subset.

[0083] S403. Split the target sub-dimension with the largest target information gain in each subset to obtain the target decision tree corresponding to each training set.

[0084] As an example, suppose a training set corresponding to the environmental dimension includes 30 samples. If, during the first split, the target sub-dimension with the largest target information gain in the training set is determined to be climate change, then climate change is taken as the first child node. For example, splitting is performed according to index values ​​greater than and less than or equal to climate change. Based on the index values ​​of climate change, the training set can be divided into set 1 (containing 20 samples) and set 2 (containing 10 samples). If, during the second split, the target sub-dimension with the largest target information gain in set 1 is determined to be natural capital, then natural capital is taken as the second child node. For example, splitting is performed according to index values ​​greater than and less than or equal to natural capital. Based on the index values ​​of natural capital, set 1 can be divided into set 3 (containing 10 samples) and set 4 (containing 10 samples).

[0085] Repeat the above splitting process until all samples in the resulting sets have the same greenness label. Then, use that greenness label as the leaf node for that set. Furthermore, if there are no remaining sub-dimensions to split, or if the number of samples in the resulting sets is less than a set threshold (e.g., 3), then use the greenness label with the highest frequency in the set as the leaf node. In this way, the target decision tree corresponding to each training set can be obtained.

[0086] Therefore, when the number of samples in the training set is large, the training set can be divided into multiple subsets. Different subsets may include the same or different numbers of samples. Then, the target sub-dimension with the largest target information gain in each subset can be calculated in parallel, and further, the target sub-dimension with the largest target information gain from the target sub-dimensions of each subset can be split to obtain the corresponding target decision tree. In this way, the speed of node splitting can be improved during training, achieving efficient construction of the decision tree.

[0087] As an example, such as Figure 5 The diagram illustrates a decision tree construction process provided in an embodiment of this application. The training set can be divided into multiple environmental dimension training sets, multiple social dimension training sets, and multiple governance dimension training sets. Decision tree construction processes can be performed in parallel on the multiple environmental dimension training sets, multiple social dimension training sets, and multiple governance dimension training sets to obtain multiple target decision trees and generate a random forest model.

[0088] The process of constructing a decision tree for an environment dimension training set may include: identifying the sub-dimensions included in the environment dimension, dividing the environment dimension training set into multiple subsets, determining the target sub-dimension of each subset, determining the target sub-dimension corresponding to the environment dimension training set, splitting based on the target sub-dimension corresponding to the environment dimension training set, and generating a decision tree.

[0089] The process of constructing a decision tree for a social dimension training set may include: identifying the sub-dimensions included in the social dimension, dividing the social dimension training set into multiple subsets, determining the target sub-dimension of each subset, determining the target sub-dimension corresponding to the social dimension training set, splitting based on the target sub-dimension corresponding to the social dimension training set, and generating a decision tree.

[0090] The process of constructing a decision tree for a governance dimension training set may include: identifying the sub-dimensions included in the governance dimension, dividing the governance dimension training set into multiple subsets, determining the target sub-dimension of each subset, determining the target sub-dimension corresponding to the governance dimension training set, splitting based on the target sub-dimension corresponding to the governance dimension training set, and generating a decision tree.

[0091] In the above embodiments of this application, the random forest model constructed based on multidimensional indicators of objects can be processed according to the importance of multidimensional indicators of objects, which can better match the greenness evaluation scenario of objects. Moreover, using this model to evaluate the greenness of objects can improve the evaluation efficiency and accuracy. Furthermore, multiple decision trees in the random forest model can be calculated in parallel, which has high efficiency.

[0092] Figure 6 This is a schematic diagram of the structure of an object evaluation device provided in an embodiment of this application. Figure 6 As shown, the object evaluation device includes an acquisition unit 501 and a processing unit 502.

[0093] The acquisition unit 501 is used to acquire indicator information of the object to be evaluated from multiple dimensions, including environmental, social, and governance dimensions. The processing unit 502 is used to process the indicator information of the multiple dimensions based on a pre-trained random forest model to obtain the evaluation result of the object to be evaluated. The evaluation result is used to indicate the greenness of the object to be evaluated. The pre-trained random forest model includes at least one target decision tree corresponding to each of the multiple dimensions. The at least one target decision tree corresponding to each dimension is constructed through the following process: for each training set corresponding to each dimension, with each training set as the root node, the target decision tree is split according to the target information gain of each sub-dimension included in each training set to obtain the target decision tree corresponding to each training set, so as to obtain at least one target decision tree. Among them, the target information gain of the sub-dimension is the product of the original information gain of the indicator information of the sub-dimension and the adjustment weight corresponding to the sub-dimension. The adjustment weight is used to represent the importance of the sub-dimension to the greenness of the evaluated object.

[0094] In other embodiments, the environmental dimension includes at least one of the following sub-dimensions: climate change, natural capital, waste treatment, and environmental opportunities; the indicators for the climate change sub-dimension include at least one of the following: carbon emissions, carbon emission intensity, carbon emission reduction, and the proportion of green energy; the indicators for the natural capital sub-dimension include at least one of the following: resource consumption intensity, ecological protection investment, and land use efficiency; the indicators for the waste treatment sub-dimension include at least one of the following: compliance rate of pollutant emission concentration, wastewater treatment compliance rate, and solid waste recycling rate; the indicators for the environmental opportunities sub-dimension include at least one of the following: the proportion of green business and the proportion of green R&D investment; the social dimension includes at least one of the following sub-dimensions: human capital, product responsibility, stakeholders, and social opportunities; the indicators for the human capital sub-dimension include at least one of the following: employee compensation competitiveness, employee... The indicators for the product responsibility sub-dimension include at least one of the following: average training time per employee, workplace accident rate, and employee satisfaction; the product responsibility sub-dimension includes at least one of the following: quality inspection pass rate, product recall rate, customer complaint handling timeliness, and proportion of recyclable production materials; the stakeholder sub-dimension includes at least one of the following: number of stakeholder objections, supplier ESG compliance rate, and number of negative events; the social opportunity sub-dimension includes at least one of the following: number of people served in public welfare projects, number of jobs created, and proportion of inclusive business; the governance dimension includes at least one of the following sub-dimensions: target governance and target behavior; the target governance sub-dimension includes at least one of the following: reasonableness of compensation, equity concentration, proportion of independent directors, and type of audit opinion; the target behavior sub-dimension includes at least one of the following: number of abnormal events per year, number of abnormal competition events, and proportion of material donations.

[0095] In other embodiments, the random forest model is trained through the following process: acquiring multiple sample information; each sample information includes multiple dimensions of indicator information of a sample object, as well as the greenness label of the sample object; training a random forest based on the multiple sample information to obtain the random forest model; wherein, the indicator information of each dimension in the multiple dimensions is trained separately using a random forest.

[0096] In other embodiments, random forest training is performed based on multiple sample information to obtain a random forest model, including: performing the following process for the indicator information of each of the multiple dimensions to achieve random forest training, so as to obtain at least one target decision tree corresponding to the multiple dimensions: sampling with replacement for the indicator information of the dimensions in the multiple sample information to obtain at least one training set corresponding to the dimension; the training set includes the indicator information of at least one sample object in each sub-dimension of the dimension, and the greenness label of the sample object; performing random forest training based on each training set to obtain at least one target decision tree corresponding to the dimension; wherein, the training of the target decision trees for the multiple dimensions is performed in parallel; and configuring voting weights for at least one target decision tree corresponding to each of the multiple dimensions to generate a random forest model.

[0097] In other embodiments, random forest training is performed based on each training set to obtain at least one target decision tree corresponding to a dimension, including: for each training set, taking the training set as the root node, splitting according to the target information gain of each sub-dimension included in the training set to obtain the target decision tree corresponding to the training set, so as to obtain at least one target decision tree; wherein, the target information gain of the sub-dimension is the product of the original information gain of the indicator information of the sub-dimension and the adjusted weight corresponding to the sub-dimension; the adjusted weight is used to represent the importance of the sub-dimension to the greenness of the evaluation object.

[0098] In other embodiments, the training set is used as the root node, and the target decision tree corresponding to the training set is obtained by splitting the training set according to the target information gain of each sub-dimension. This includes: dividing the training set into multiple subsets; determining the sub-dimension with the largest target information gain in each subset as the target sub-dimension of each subset; and splitting the training set according to the target sub-dimension with the largest target information gain in each subset to obtain the target decision tree corresponding to each training set.

[0099] The object evaluation device provided in this application embodiment can execute the method shown in the above method embodiment. Its implementation principle and beneficial effects can be referred to the relevant description in the method embodiment, and will not be repeated here.

[0100] Figure 7 This is a schematic diagram of another model training device provided in an embodiment of this application. Figure 7As shown, the model training device includes an acquisition unit 601 and a processing unit 602.

[0101] The acquisition unit 601 is used to acquire multiple sample information; each sample information includes indicator information of multiple dimensions of a sample object, as well as the greenness label of the sample object; the multiple dimensions include environmental dimension, social dimension, and governance dimension; the processing unit 602 is used to train a random forest based on the multiple sample information to obtain a random forest model; wherein, the indicator information of each dimension in the multiple dimensions is trained separately using a random forest to obtain at least one target decision tree corresponding to each dimension in the multiple dimensions; the at least one target decision tree corresponding to each dimension is constructed through the following process: for each training set corresponding to each dimension, with each training set as the root node, the target decision tree is split according to the target information gain of each sub-dimension included in each training set to obtain the target decision tree corresponding to each training set, so as to obtain at least one target decision tree; wherein, the target information gain of the sub-dimension is the product of the original information gain of the indicator information of the sub-dimension and the adjusted weight of the sub-dimension; the adjusted weight is used to represent the importance of the sub-dimension to the greenness of the evaluation object.

[0102] In other embodiments, the environmental dimension includes at least one of the following sub-dimensions: climate change, natural capital, waste treatment, and environmental opportunities; the indicators for the climate change sub-dimension include at least one of the following: carbon emissions, carbon emission intensity, carbon emission reduction, and the proportion of green energy; the indicators for the natural capital sub-dimension include at least one of the following: resource consumption intensity, ecological protection investment, and land use efficiency; the indicators for the waste treatment sub-dimension include at least one of the following: compliance rate of pollutant emission concentration, wastewater treatment compliance rate, and solid waste recycling rate; the indicators for the environmental opportunities sub-dimension include at least one of the following: the proportion of green business and the proportion of green R&D investment; the social dimension includes at least one of the following sub-dimensions: human capital, product responsibility, stakeholders, and social opportunities; the indicators for the human capital sub-dimension include at least one of the following: employee compensation competitiveness, employee... The indicators for the product responsibility sub-dimension include at least one of the following: average training time per employee, workplace accident rate, and employee satisfaction; the product responsibility sub-dimension includes at least one of the following: quality inspection pass rate, product recall rate, customer complaint handling timeliness, and proportion of recyclable production materials; the stakeholder sub-dimension includes at least one of the following: number of stakeholder objections, supplier ESG compliance rate, and number of negative events; the social opportunity sub-dimension includes at least one of the following: number of people served in public welfare projects, number of jobs created, and proportion of inclusive business; the governance dimension includes at least one of the following sub-dimensions: target governance and target behavior; the target governance sub-dimension includes at least one of the following: reasonableness of compensation, equity concentration, proportion of independent directors, and type of audit opinion; the target behavior sub-dimension includes at least one of the following: number of abnormal events per year, number of abnormal competition events, and proportion of material donations.

[0103] In other embodiments, the processing unit 602 is specifically configured to: perform the following process for the indicator information of each of the multiple dimensions to implement random forest training, so as to obtain at least one target decision tree corresponding to the multiple dimensions: sampling the indicator information of the dimensions in the multiple sample information with replacement to obtain at least one training set corresponding to the dimension; the training set includes the indicator information of at least one sample object in each sub-dimension of the dimension, and the greenness label of the sample object; perform random forest training based on each training set to obtain at least one target decision tree corresponding to the dimension; wherein, the training of the target decision trees for the multiple dimensions is performed in parallel; and configure voting weights for at least one target decision tree corresponding to each of the multiple dimensions to generate a random forest model.

[0104] In other embodiments, the processing unit 602 is specifically configured to: for each training set, using the training set as the root node, split according to the target information gain of each sub-dimension included in the training set to obtain a target decision tree corresponding to the training set, so as to obtain at least one target decision tree; wherein, the target information gain of the sub-dimension is the product of the original information gain of the indicator information of the sub-dimension and the adjustment weight corresponding to the sub-dimension; the adjustment weight is used to represent the importance of the sub-dimension to the greenness of the evaluation object.

[0105] In other embodiments, the processing unit 602 is specifically used to: divide the training set to obtain multiple subsets; determine the sub-dimension with the largest target information gain in each subset as the target sub-dimension of each subset; and split according to the target sub-dimension with the largest target information gain in each subset to obtain the target decision tree corresponding to each training set.

[0106] The model training device provided in this application embodiment can execute the method shown in the above method embodiment. Its implementation principle and beneficial effects can be referred to the relevant description in the method embodiment, and will not be repeated here.

[0107] like Figure 8 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application. Figure 8 The illustrated electronic device may include a processor 701, a memory 702, a communication interface 703, and a bus 704. The processor 701, the memory 702, and the communication interface 703 can be connected via the bus 704.

[0108] The processor 701 is the control center of the electronic device. It can be a general-purpose central processing unit (CPU) or other general-purpose processors. The general-purpose processor can be a microprocessor or any conventional processor.

[0109] As an example, processor 701 may include one or more CPUs, for example Figure 8 CPU0 and CPU1 are shown in the diagram.

[0110] The memory 702 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0111] In one possible implementation, the memory 702 can exist independently of the processor 701. The memory 702 can be connected to the processor 701 via a bus 704 and is used to store data, instructions, or program code. When the processor 701 calls and executes the instructions or program code stored in the memory 702, it can implement the object evaluation method or model training method in the above embodiments of this application.

[0112] In another possible implementation, the memory 702 can also be integrated with the processor 701.

[0113] Communication interface 703 is used for connecting electronic devices to other devices via a communication network, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc. Communication interface 703 may include a receiving unit for receiving data and a transmitting unit for transmitting data.

[0114] Bus 704 can be an industry standard architecture (ISA) bus, a peripheral component interconnect (PCI) bus, or an extended industry standard architecture (EISA) bus. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0115] It should be pointed out that, Figure 8 The structures shown do not constitute a limitation on electronic devices, except... Figure 8 In addition to the components shown, electronic devices may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0116] This application also provides a chip for executing instructions, which is used to execute the technical solutions of the object evaluation method or model training method in the above embodiments.

[0117] This application also provides a computer-readable storage medium storing computer instructions. When these computer instructions are executed by a processor, they are used to implement the technical solutions of the object evaluation method or model training method described in the above embodiments. Specifically, when the computer instructions are executed by a processor, the electronic device can perform the technical solutions of the object evaluation method or model training method described in the above embodiments.

[0118] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solutions of the object evaluation method or model training method in the above embodiments. The aforementioned computer-readable storage media can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage media can be any available medium accessible to a general-purpose or special-purpose computer. An exemplary computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the computer-readable storage medium can exist as discrete components in an electronic control unit or main control device; this application does not limit this.

[0119] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0120] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0121] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0122] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0123] It should be understood that the steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

[0124] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. An object evaluation method characterized by comprising: The method comprises: obtaining index information of multiple dimensions of a to-be-evaluated object; the multiple dimensions comprise an environmental dimension, a social dimension and a governance dimension; processing the index information of the multiple dimensions based on a pre-trained random forest model to obtain an evaluation result of the to-be-evaluated object; the evaluation result is used to indicate the green degree of the to-be-evaluated object; the pre-trained random forest model comprises at least one target decision tree corresponding to each dimension of the multiple dimensions; the at least one target decision tree corresponding to each dimension is constructed by the following process: for each training set corresponding to each dimension, taking each training set as a root node, splitting according to the target information gain of each sub-dimension included in each training set to obtain a target decision tree corresponding to each training set, so as to obtain the at least one target decision tree; wherein the target information gain of the sub-dimension is the product of the original information gain of the index information of the sub-dimension and an adjustment weight corresponding to the sub-dimension; the adjustment weight is used to represent the importance of the sub-dimension to the green degree of the evaluation object.

2. The method of claim 1, wherein, The environmental dimension comprises at least one of the following sub-dimensions: climate change, natural capital, waste disposal, environmental opportunity; The social dimension comprises at least one of the following sub-dimensions: human capital, product responsibility, stakeholder, social opportunity; The governance dimension comprises at least one of the following sub-dimensions: object governance, object behavior.

3. The method according to claim 1 or 2, characterized in that, The random forest model is trained by the following process: obtaining a plurality of sample information; one sample information comprises index information of a plurality of dimensions of a sample object, and a green degree label of the sample object; training the random forest based on the plurality of sample information to obtain the random forest model; wherein the index information of each dimension in the plurality of dimensions is respectively trained by the random forest.

4. The method of claim 3, wherein, The random forest training based on the plurality of sample information to obtain the random forest model comprises: for the index information of each dimension in the plurality of dimensions, the following process is executed to realize random forest training to obtain at least one target decision tree corresponding to the plurality of dimensions: sampling with replacement is performed on the index information of the dimension in the plurality of sample information to obtain at least one training set corresponding to the dimension; the training set comprises index information of each sub-dimension of at least one sample object under the dimension, and a green degree label of the sample object; random forest training is performed based on each training set to obtain at least one target decision tree corresponding to the dimension; wherein the training of the target decision trees of the plurality of dimensions is performed in parallel; configuring a voting weight for the at least one target decision tree corresponding to each dimension in the plurality of dimensions to generate the random forest model.

5. The method of claim 1, wherein, taking each training set as a root node, splitting according to the target information gain of each sub-dimension included in each training set to obtain a target decision tree corresponding to each training set, comprises: dividing each training set to obtain a plurality of subsets; determining a sub-dimension with the maximum target information gain in each subset as a target sub-dimension of each subset; Split according to the target sub-dimension with the maximum target information gain in the target sub-dimension of each subset, to obtain the target decision tree corresponding to each training set.

6. A model training method, comprising: Comprise: Obtain a plurality of sample information; One sample information includes a plurality of dimension index information of a sample object, and a green degree label of the sample object; the plurality of dimensions include environmental dimension, social dimension and governance dimension; Based on the plurality of sample information, random forest training is performed to obtain a random forest model; Wherein, the index information of each dimension in the plurality of dimensions is respectively subjected to random forest training to obtain at least one target decision tree corresponding to each dimension in the plurality of dimensions; at least one target decision tree corresponding to each dimension is constructed by the following process: for each training set corresponding to each dimension, taking each training set as a root node, splitting according to the target information gain of each sub-dimension included in each training set, to obtain the target decision tree corresponding to each training set, to obtain the at least one target decision tree; wherein, the target information gain of the sub-dimension is the product of the original information gain of the index information of the sub-dimension and the adjustment weight corresponding to the sub-dimension; the adjustment weight is used to represent the importance of the sub-dimension to the green degree of the evaluation object.

7. The method of claim 6, wherein, The environmental dimension includes at least one of the following sub-dimensions: climate change, natural capital, waste disposal, environmental opportunity; The social dimension includes at least one of the following sub-dimensions: human capital, product responsibility, stakeholder, social opportunity; The governance dimension includes at least one of the following sub-dimensions: object governance, object behavior.

8. The method according to claim 6 or 7, characterized in that, The random forest training based on the plurality of sample information to obtain the random forest model comprises: For the index information of each dimension in the plurality of dimensions, the following process is performed to realize random forest training to obtain at least one target decision tree corresponding to the plurality of dimensions: resample the index information of the dimension in the plurality of sample information to obtain at least one training set corresponding to the dimension; the training set includes the index information of each sub-dimension of at least one sample object under the dimension, and the green degree label of the sample object; based on each training set, random forest training is performed to obtain at least one target decision tree corresponding to the dimension; wherein, the training of the target decision tree of the plurality of dimensions is performed in parallel; For at least one target decision tree corresponding to each dimension in the plurality of dimensions, configure a voting weight to generate the random forest model.

9. The method of claim 6, wherein, Taking each training set as a root node, splitting according to the target information gain of each sub-dimension included in each training set to obtain the target decision tree corresponding to each training set, comprises: Divide each training set to obtain a plurality of subsets; Determine the target sub-dimension with the maximum target information gain in each subset as the target sub-dimension of each subset; Split according to the target sub-dimension with the maximum target information gain in the target sub-dimension of each subset to obtain the target decision tree corresponding to each training set.

10. An object evaluation device characterized by, Comprise: An acquisition unit is configured to acquire index information of a plurality of dimensions of an object to be evaluated; the plurality of dimensions include environmental dimension, social dimension and governance dimension; The processing unit is configured to process the index information of the plurality of dimensions based on a pre-trained random forest model to obtain an evaluation result of the to-be-evaluated object, and the evaluation result is used to indicate the green degree of the to-be-evaluated object. The pre-trained random forest model comprises at least one target decision tree corresponding to each dimension of the plurality of dimensions, and each target decision tree corresponding to each dimension is constructed by the following process: for each training set corresponding to each dimension, taking each training set as a root node, splitting according to a target information gain of each sub-dimension included in each training set to obtain a target decision tree corresponding to each training set, so as to obtain the at least one target decision tree; wherein the target information gain of the sub-dimension is a product of an original information gain of the index information of the sub-dimension and an adjustment weight corresponding to the sub-dimension; and the adjustment weight is used to indicate the importance of the sub-dimension to the green degree of the evaluation object.

11. A model training apparatus, comprising: The method comprises: The acquisition unit is configured to acquire a plurality of sample information. Each sample information comprises index information of a plurality of dimensions of a sample object and a green degree label of the sample object, and the plurality of dimensions comprise an environmental dimension, a social dimension and a governance dimension. The processing unit is configured to perform random forest training based on the plurality of sample information to obtain a random forest model. Each dimension of the plurality of dimensions is subjected to random forest training to obtain at least one target decision tree corresponding to each dimension of the plurality of dimensions, and each target decision tree corresponding to each dimension is constructed by the following process: for each training set corresponding to each dimension, taking each training set as a root node, splitting according to a target information gain of each sub-dimension included in each training set to obtain a target decision tree corresponding to each training set, so as to obtain the at least one target decision tree; wherein the target information gain of the sub-dimension is a product of an original information gain of the index information of the sub-dimension and an adjustment weight corresponding to the sub-dimension; and the adjustment weight is used to indicate the importance of the sub-dimension to the green degree of the evaluation object.

12. An electronic device, comprising: The electronic device comprises a memory and at least one processor; the memory is in communication connection with the processor; the memory is configured to store computer program codes, the computer program codes comprise computer instructions; and when the processor executes the computer instructions, the electronic device performs the method according to any one of claims 1-5 or 6-9. The computer readable storage medium stores computer instructions, and the computer instructions are executed by a processor to implement the method according to any one of claims 1-5 or 6-9.

13. A computer-readable storage medium, characterized in that, When the computer program product is running on the computer / being executed by the processor of the computer, the method according to any one of claims 1-5 or 6-9 is implemented.

14. A computer program product, characterised in that, ​

Images