Project evaluation method and device based on feature contribution analysis, medium and product

Abstract

The invention discloses a feature contribution analysis-based project evaluation method and device, a medium and a product. The method comprises the following steps of: constructing a standardized feature vector according to original data of a to-be-evaluated project; according to the standardized feature vector, an improved decision tree evaluation model is adopted to generate an evaluation result and a decision path of the to-be-evaluated project; wherein the improved decision tree evaluation model is obtained by improving a decision tree model by adopting a cost complexity pruning function and a feature weight adjustment factor; performing feature contribution analysis on each feature in the standardized feature vector according to the decision path to obtain a contribution vector; adopting a preset text template and the contribution vector to generate an interpretable text; and generating an evaluation report according to the evaluation result, the decision path, the contribution vector and the interpretability text. A pruning constraint and weight adjustment mechanism is introduced on the basis of a traditional decision tree, and the prediction precision and project interpretation ability of the model can be improved.

Description

Technical Field

[0001] This invention relates to the field of project evaluation technology, and in particular to a project evaluation method, apparatus, medium and product based on feature contribution analysis. Background Technology

[0002] In fields such as system operation innovation projects, engineering construction, supply chain optimization, and government project management, scientific evaluation and quantitative decision-making throughout the entire project process have become crucial. Existing project evaluation methods exhibit a significant contradiction between predictive accuracy and interpretability: first, high-precision models can provide relatively accurate predictions, but they are black-box structures lacking transparent decision-making logic; second, while traditional decision trees offer interpretability, their predictive accuracy is limited, and they are prone to overfitting or distortion in complex, multi-dimensional interactive scenarios. This situation results in managers being unable to simultaneously obtain highly credible evaluation results and traceable explanatory evidence in practical applications such as project budget approval, resource allocation, performance evaluation, and audit supervision, impacting the scientific validity and adoptability of decisions. Summary of the Invention

[0003] This invention provides a project evaluation method, device, medium, and product based on feature contribution analysis. It introduces pruning constraints and weight adjustment mechanisms on the basis of traditional decision trees to ensure the prediction accuracy and generalization ability of the model under complex multidimensional feature conditions. Through the path splitting contribution calculation method, it quantifies the positive and negative contributions of each input feature in the single project evaluation, forming a complete feature contribution vector and enhancing the project interpretability.

[0004] To achieve the above objectives, embodiments of the present invention provide a project evaluation method based on feature contribution analysis, comprising: Obtain the raw data of the project to be evaluated, and construct a standardized feature vector of the project to be evaluated based on the raw data; The evaluation results and decision paths for the project to be evaluated are generated using an improved decision tree evaluation model based on the standardized feature vectors. The improved decision tree evaluation model is obtained by improving the decision tree model using a cost complexity pruning function and a feature weight adjustment factor. Based on the decision path, a feature contribution analysis is performed on each feature in the standardized feature vector to obtain the contribution vector of the standardized feature vector; An interpretable text for the contribution vector is generated using a preset text template and the contribution vector; an evaluation report for the project to be evaluated is generated based on the evaluation results, the decision path, the contribution vector, and the interpretable text.

[0005] As an improvement to the above scheme, the method for obtaining the improved decision tree evaluation model includes: Obtain historical raw data of several historical evaluation projects, construct historical standardized feature vectors based on the historical raw data, form a sample dataset by combining the historical standardized feature vectors and the corresponding historical evaluation results, and divide the sample dataset into a training set and a validation set; Initialize the decision tree model and determine the splitting criteria for the decision tree model; Using the training set, the splitting nodes are recursively split according to the splitting criteria and feature weights until a preset stopping condition is met, generating a preliminary decision tree model; Based on the validation set, the cost complexity pruning function is used to optimize the pruning coefficients of the initial decision tree model through grid search to obtain the pruned decision tree model. The pruned decision tree model is subjected to split path verification according to preset business rules. The target threshold corresponding to the target split node that does not conform to the preset business rules is adjusted. If the adjustment is ineffective, the optimal candidate split combination that conforms to the preset business rules is selected to replace the target split node, thereby obtaining an improved decision tree evaluation model.

[0006] As an improvement to the above scheme, the step of recursively splitting nodes using the training set according to the splitting criteria and feature weights until a preset stopping condition is met, and generating a preliminary decision tree model, includes: The split scores of all candidate split combinations for the current node to be split in the decision tree model are determined according to the split criteria and feature weight adjustment factors; wherein, the candidate split combination is a candidate combination of the candidate split features and the corresponding split threshold of the current node to be split. The candidate splitting combination with the highest splitting score is taken as the optimal splitting combination and recursively splitting nodes until a preset stopping condition is met, thus generating a preliminary decision tree model.

[0007] As an improvement to the above scheme, the step of optimizing the pruning coefficients of the initial decision tree model using a cost complexity pruning function based on the validation set through grid search to obtain the pruned decision tree model includes: Set the initial range and step size of the pruning coefficient, and then assign values ​​to the pruning coefficient based on the initial range and step size; The preliminary decision tree model is pruned according to the pruning coefficient for each value to obtain the corresponding pruned model; Based on the validation set, a cost-complexity pruning function is used to evaluate the performance of each pruned model through grid search; The pruning model corresponding to the pruning coefficient with the best model performance evaluation result is used as the pruned decision tree model.

[0008] As an improvement to the above scheme, the step of performing feature contribution analysis on each feature in the standardized feature vector according to the decision path to obtain the contribution vector of the standardized feature vector includes: The overall prediction result of the decision path is calculated based on the prediction increment of each split node in the decision path. The total contribution of each feature in the standardized feature vector is calculated based on the predicted increment of each split node to obtain the contribution vector of the standardized feature vector.

[0009] As an improvement to the above scheme, after obtaining the improved decision tree evaluation model, the method further includes: The improved decision tree evaluation model shall be audited and managed.

[0010] To achieve the above objectives, embodiments of the present invention provide a project evaluation device based on feature contribution analysis, comprising: The raw data acquisition module is used to acquire the raw data of the project to be evaluated and construct a standardized feature vector of the project to be evaluated based on the raw data. The evaluation result generation module is used to generate the evaluation result and decision path of the project to be evaluated based on the standardized feature vector using an improved decision tree evaluation model; wherein, the improved decision tree evaluation model is obtained by improving the decision tree model by using a cost complexity pruning function and a feature weight adjustment factor; The feature contribution analysis module is used to perform feature contribution analysis on each feature in the standardized feature vector according to the decision path, so as to obtain the contribution vector of the standardized feature vector. The evaluation report generation module is used to generate interpretable text for the contribution vector using a preset text template and the contribution vector; and to generate an evaluation report for the project to be evaluated based on the evaluation results, the decision path, the contribution vector, and the interpretable text.

[0011] To achieve the above objectives, embodiments of the present invention provide a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the above-described project evaluation method based on feature contribution analysis.

[0012] To achieve the above objectives, embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the above-described project evaluation method based on feature contribution analysis.

[0013] To achieve the above objectives, embodiments of the present invention also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the above-described project evaluation method based on feature contribution analysis.

[0014] Compared with existing technologies, this invention discloses a project evaluation method, apparatus, medium, and product based on feature contribution analysis. The method involves acquiring the original data of the project to be evaluated, constructing a standardized feature vector for the project based on the original data, generating an evaluation result and decision path for the project using an improved decision tree evaluation model based on the standardized feature vector, wherein the improved decision tree evaluation model is obtained by improving the decision tree model using a cost complexity pruning function and a feature weight adjustment factor; performing feature contribution analysis on each feature in the standardized feature vector based on the decision path to obtain a contribution vector; generating interpretable text for the contribution vector using a preset text template and the contribution vector; and generating an evaluation report for the project based on the evaluation result, the decision path, the contribution vector, and the interpretable text. By introducing pruning constraints and weight adjustment mechanisms on the basis of traditional decision trees, the method ensures the prediction accuracy and generalization ability of the model under complex multi-dimensional feature conditions. Through a path splitting contribution calculation method, the positive and negative contributions of each input feature in the single project evaluation are quantified to form a complete feature contribution vector, enhancing the project interpretability. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a project evaluation method based on feature contribution analysis provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a project evaluation device based on feature contribution analysis provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of a terminal device provided in an embodiment of the present invention. Detailed Implementation

[0016] 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.

[0017] It should be noted that the terms "comprising" and "specific" in this invention, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0018] Please see Figure 1 , Figure 1 This is a flowchart illustrating a project evaluation method based on feature contribution analysis provided in an embodiment of the present invention. The project evaluation method based on feature contribution analysis includes: S1, Obtain the original data of the project to be evaluated, and construct a standardized feature vector of the project to be evaluated based on the original data; S2, Based on the standardized feature vector, an improved decision tree evaluation model is used to generate the evaluation results and decision paths for the project to be evaluated; wherein, the improved decision tree evaluation model is obtained by improving the decision tree model using a cost complexity pruning function and a feature weight adjustment factor; S3, perform feature contribution analysis on each feature in the standardized feature vector according to the decision path to obtain the contribution vector of the standardized feature vector; S4, using a preset text template and the contribution vector to generate interpretable text for the contribution vector; and generating an evaluation report for the project to be evaluated based on the evaluation results, the decision path, the contribution vector, and the interpretable text.

[0019] For example, standardized modeling and feature construction of raw data are performed to ensure the integrity, consistency, and computability of the input data. Since project management scenarios often involve multi-source heterogeneous data, including budget information, schedule plans, team configurations, historical performance, and external environmental indicators, failure to process this data can lead to distorted modeling results. Preprocessing of raw data is crucial, such as data cleaning: removing logically contradictory records and imputing variables with few missing values ​​using mean, median, or interpolation. Feature standardization: normalizing numerical features to ensure they are distributed within a uniform scale, avoiding bias introduced by differences in units of measurement. For categorical features (such as industry classification, project level, etc.), one-hot encoding or embedding encoding is used to convert them into computable vectors. Feature enhancement: combining business logic to construct composite indicators, such as unit duration output rate, budget input-output ratio, and team experience average, further enhances feature expressiveness. Label setting: defining continuous or categorical labels based on project evaluation objectives. Continuous labels are often comprehensive evaluation scores, while categorical labels are often level classifications used for classification modeling. After the above processing, the project data is represented by feature vectors, which have consistency and trainability.

[0020] Construct input feature vectors (standardized feature vectors) for the project to be evaluated: , in, Indicates the first Standardized feature vectors of the projects to be evaluated; Indicates the first The project in the Values ​​can be taken in each feature dimension; This represents the total number of feature dimensions. The original data for the project to be evaluated includes, but is not limited to: budget, duration, phase, unit type, business affiliation, historical completion rate, delay rate, and member experience values. Output labels can be continuous scores. (Regression task), or rating labels (excellent, good, average, poor, etc., categorized tasks).

[0021] Perform missing value imputation, categorical feature encoding (such as One-Hot), numerical feature standardization or binning, outlier pruning, and feature selection on the original data; if necessary, introduce oversampling / undersampling strategies to balance sample categories for classification tasks.

[0022] The pre-set text template includes: "Concluding sentence + list of key factors + direction / magnitude of influence + actionable suggestions". For example: "The project received an evaluation score of 86.5 points, mainly due to factors such as a budget of less than 3 million (+5.2 points), a project cycle of less than 6 months (+3.7 points), and a high level of experience in the person in charge (+4.8 points)." The module supports threshold filtering (output only). Key factors), synonym rewriting, and automatic unit / range adaptation ensure that the results are manager-friendly and can be directly used in evaluation reports.

[0023] In step S4, based on different feature combinations, corresponding natural language generation templates are matched and automatically concatenated to form explanatory text. Factors are categorized and displayed hierarchically according to their contribution, such as key factors, secondary factors, and general factors, facilitating managers' quick focus on key areas. In conjunction with an industry knowledge base, decision-making support suggestions are appended to the explanatory text, making the results not only readable but also usable. The invention supports output in a combination of charts and text, enhancing the intuitiveness of the results. This embodiment of the invention transforms the numerical contribution results output by the model into highly readable textual expressions, enabling both technical and managerial personnel to quickly understand the evaluation basis and utilize it in actual decision-making processes, thereby improving the understandability and application value of the results.

[0024] Specifically, the method for obtaining the improved decision tree evaluation model includes: Obtain historical raw data of several historical evaluation projects, construct historical standardized feature vectors based on the historical raw data, form a sample dataset by combining the historical standardized feature vectors and the corresponding historical evaluation results, and divide the sample dataset into a training set and a validation set; Initialize the decision tree model and determine the splitting criteria for the decision tree model; Using the training set, the splitting nodes are recursively split according to the splitting criteria and feature weights until a preset stopping condition is met, generating a preliminary decision tree model; Based on the validation set, the cost complexity pruning function is used to optimize the pruning coefficients of the initial decision tree model through grid search to obtain the pruned decision tree model. The pruned decision tree model is subjected to split path verification according to preset business rules. The target threshold corresponding to the target split node that does not conform to the preset business rules is adjusted. If the adjustment is ineffective, the optimal candidate split combination that conforms to the preset business rules is selected to replace the target split node, thereby obtaining an improved decision tree evaluation model.

[0025] For example, when constructing historical standardized feature vectors, a threshold-based feature selection mechanism is introduced: when the correlation coefficient between a feature and its label is greater than a set correlation threshold, a feature selection mechanism is introduced. When the noise level is below the threshold, retain the feature; when the noise level is below the threshold, retain the feature. In cases where the value is 1, it is removed; otherwise, its weight is reduced through regularization penalties.

[0026] The correlation coefficient function can be expressed as: , In the formula, Features With tags The correlation coefficient between them; Features With tags covariance; Features Standard deviation; For tags The standard deviation.

[0027] when When describing features It has a strong positive correlation with the label and can be used as a retained feature; when When this occurs, it can be identified as a noise feature and removed. To retain a threshold for correlation, The noise threshold is set based on business needs or experience.

[0028] The decision tree model is trained using the CART (Classification and Regression Tree) framework, with information gain or the Gini index as the splitting criterion. The expression for information gain is: , In the formula, In the sample set Selecting features and threshold The information gain resulting from splitting; This is the current set of samples to be split; Indicating candidate splitting features With the corresponding splitting threshold The combination; For sample set The impurity measure can be entropy or the Gini index; after splitting, several subsets are obtained. , For the first The number of samples in each subset; For the original sample set The number of samples; For subset Impurity measure; summation symbol This indicates that all subsets are summed; the greater the information gain, the better the splitting effect under this feature-threshold combination.

[0029] To control model complexity, a cost complexity pruning function is introduced: , in, This is due to empirical error. The number of leaf nodes, This is the pruning coefficient, used to control model complexity; To the pruning coefficient Below, decision tree The total cost.

[0030] Introducing a feature weight adjustment factor during split selection The formula for calculating the split score is: , in, Features At the threshold The overall split score below; Features The weighting adjustment factor is used to increase the priority of key business features in the splitting process; In the sample set Features adopted and threshold Information gain; during node splitting, compare the information gain of each candidate feature-threshold combination. The combination with the highest score is the optimal split combination.

[0031] Used to increase the priority of key business features in the split.

[0032] At the same time, a business rule verification module is set up to filter out splits that do not conform to business logic (such as the "stage irreversibility" rule) to ensure the rationality of the model results.

[0033] This invention prioritizes key business features during feature selection to ensure their full representation in model splitting; it introduces pruning during model generation to prevent excessively deep tree structures and avoid reduced generalization ability due to overfitting; it embeds business constraints into splitting nodes based on project management logic to ensure model partitioning results conform to the basic laws of the project lifecycle; and it enhances overall robustness through multi-tree ensemble, maintaining interpretability while ensuring result stability. This allows for the construction of an improved decision tree model that balances interpretability and predictive performance, thereby improving the credibility of the evaluation system.

[0034] More specifically, the step of using the training set to recursively split nodes according to the splitting criteria and feature weights until a preset stopping condition is met, thereby generating a preliminary decision tree model, includes: The split scores of all candidate split combinations for the current node to be split in the decision tree model are determined according to the split criteria and feature weight adjustment factors; wherein, the candidate split combination is a candidate combination of the candidate split features and the corresponding split threshold of the current node to be split. The candidate splitting combination with the highest splitting score is taken as the optimal splitting combination and recursively splitting nodes until a preset stopping condition is met, thus generating a preliminary decision tree model.

[0035] For example, the preset stopping conditions include any of the following: (1) the number of child node samples is less than or equal to the minimum threshold, and further splitting is not possible; (2) the splitting score of all candidate feature-threshold combinations is less than or equal to the preset threshold, and there is no gain in splitting; (3) the depth of the tree reaches the preset maximum value to avoid the structure from being too deep.

[0036] More specifically, the step of optimizing the pruning coefficients of the initial decision tree model using a cost complexity pruning function based on the validation set through grid search to obtain the pruned decision tree model includes: Set the initial range and step size of the pruning coefficient, and then assign values ​​to the pruning coefficient based on the initial range and step size; The preliminary decision tree model is pruned according to the pruning coefficient for each value to obtain the corresponding pruned model; Based on the validation set, a cost-complexity pruning function is used to evaluate the performance of each pruned model through grid search; The pruning model corresponding to the pruning coefficient with the best model performance evaluation result is used as the pruned decision tree model.

[0037] Specifically, step S3 includes: S31, calculate the overall prediction result of the decision path based on the prediction increment of each split node in the decision path; S32, calculate the total contribution of each feature in the standardized feature vector based on the predicted increment of each split node, so as to obtain the contribution vector of the standardized feature vector.

[0038] For example, for any project sample The model generates a unique path (decision path) from the root node to the leaf node: , in, The decision path for a given project sample in the decision tree represents the sequence of nodes visited from the root node to the leaf node. For the first on the path One split node; This represents the path length, i.e., the number of split nodes.

[0039] Let the prediction increment brought about by each node split be . The overall prediction result is as follows: , in, This refers to the final predicted value or evaluation result of the project. The root node baseline value, For the first The incremental contribution of each split node.

[0040] If a certain feature If a path is used multiple times, its total contribution is represented in the path. The cumulative impact value of this feature, and the expression for the total contribution, is as follows: , in, For the first The total contribution of each feature to the current project decision-making path; the summation range is the path. All splitting features in the middle are equal to The set of nodes; For the first The predicted increment for each node is as described above.

[0041] This leads to the formation of a contribution vector: , in, The feature contribution vector for the current project, including its dimension and number of features. Consistent.

[0042] It sorts the data by absolute value and outputs the Top-K key influencing factors (positive values ​​add points, negative values ​​deduct points), achieving transparent explanation at the single-item level.

[0043] The contribution ranking mechanism is defined as follows: , in, Features The absolute value of contribution; As the contribution threshold, when At that time, this characteristic was considered a key influencing factor; This represents the set of key features obtained by sorting and filtering the contribution vectors. It can output Top-K features in descending order of absolute contribution value to explain the model's decision-making.

[0044] When making predictions, the model generates a unique path (decision path) from the root node to a leaf node for each input item. Each split node corresponds to a specific feature condition; the presence of this feature at that node signifies its concrete impact on the final prediction result. Through path backtracking, the prediction result of a single item is decomposed into the contribution of multiple features. Interpretive analysis of the prediction process for a single item within the model generates a contribution vector for that item, and a ranking mechanism identifies the most critical influencing factors. This mechanism enables transparent tracing of the individual item's outcome. Contribution can be represented positively or negatively: a positive contribution means the feature improves the item's evaluation result, while a negative contribution means the feature inhibits the result.

[0045] Furthermore, after obtaining the improved decision tree evaluation model, the method also includes: The improved decision tree evaluation model shall be audited and managed.

[0046] For example, establish model version management, save the tree structure and node parameters of different versions of the improved decision tree evaluation model, and compare the differences in paths and contributions: Structural drift detection: Comparing the paths of old and new models and If the edit distance of the decision paths or the difference in feature sets between the two exceed a threshold This triggers an audit alert; Contribution stability test: Compare the rank correlation coefficients of the old and new contribution vectors. The time stamp is inconsistent and requires auditing; In model auditing and management, a dual detection method is used: path difference and contribution stability. The formula for measuring the difference in decision paths is: , in, The difference in decision paths between the old and new models on the same baseline sample, with a value range of [value missing]. ; The set of decision path nodes generated for the old version of the model; The set of decision path nodes generated for the new version of the model; symbols The symmetric difference of sets (elements that appear only in one of the sets); It is the union of the node sets of the two paths; Indicates the number of elements in the set; when ( When the preset path difference threshold is reached, a model structure drift alarm is triggered.

[0047] The formula for measuring contribution stability is: , In the formula, The stability coefficient of the new and old models in terms of feature contribution ranking; The feature contribution vector calculated for the old version model; The feature contribution vector calculated for the new version of the model; , These are the order columns of the old and new contribution vectors, sorted by absolute value from largest to smallest. For the rank correlation function, the Spearman rank correlation coefficient can be preferably used in the embodiments; when hour( (Based on a preset stability threshold), it is believed that there is a significant difference in the ranking of feature importance between the old and new models, which requires triggering manual auditing or model retraining.

[0048] Performance degradation monitoring: Monitor metrics such as MAE / ACC / AUC online, and trigger retraining when they fall below a set threshold; Retraining strategy: Use rolling time windows and incremental sample updates, and combine grid search to determine the optimal combination of pruning coefficients and feature weights; save logs for all update processes to meet compliance requirements.

[0049] This invention discloses a project evaluation method based on feature contribution analysis. The method involves acquiring the original data of the project to be evaluated, constructing a standardized feature vector for the project based on the original data, generating an evaluation result and decision path for the project using an improved decision tree evaluation model based on the standardized feature vector, wherein the improved decision tree evaluation model is obtained by improving the decision tree model using a cost complexity pruning function and a feature weight adjustment factor; performing feature contribution analysis on each feature in the standardized feature vector based on the decision path to obtain a contribution vector; generating interpretable text for the contribution vector using a preset text template and the contribution vector; and generating an evaluation report for the project based on the evaluation result, the decision path, the contribution vector, and the interpretable text. This method introduces pruning constraints and weight adjustment mechanisms on the basis of traditional decision trees to ensure the prediction accuracy and generalization ability of the model under complex multi-dimensional feature conditions. Through a path splitting contribution calculation method, the positive and negative contributions of each input feature in the single project evaluation are quantified to form a complete feature contribution vector, enhancing the project interpretability.

[0050] See Figure 2 , Figure 2 This is a schematic diagram of a project evaluation device 10 based on feature contribution analysis provided in an embodiment of the present invention. The project evaluation device 10 based on feature contribution analysis includes: The raw data acquisition module 11 is used to acquire the raw data of the project to be evaluated and construct a standardized feature vector of the project to be evaluated based on the raw data. The evaluation result generation module 12 is used to generate the evaluation result and decision path of the project to be evaluated based on the standardized feature vector using an improved decision tree evaluation model; wherein, the improved decision tree evaluation model is obtained by improving the decision tree model by using a cost complexity pruning function and a feature weight adjustment factor; The feature contribution analysis module 13 is used to perform feature contribution analysis on each feature in the standardized feature vector according to the decision path, so as to obtain the contribution vector of the standardized feature vector. The evaluation report generation module 14 is used to generate interpretable text of the contribution vector using a preset text template and the contribution vector; and to generate an evaluation report of the project to be evaluated based on the evaluation results, the decision path, the contribution vector and the interpretable text.

[0051] Furthermore, the project evaluation device 10 based on feature contribution analysis also includes: The project evaluation device 10 based on feature contribution analysis provided in this embodiment of the invention can realize all the processes of the project evaluation method based on feature contribution analysis in the above embodiments. The functions and technical effects of each module in the device are the same as the functions and technical effects of the project evaluation method based on feature contribution analysis in the above embodiments, and will not be repeated here.

[0052] See Figure 3 , Figure 3 This is a schematic diagram of the structure of a terminal device 20 provided in an embodiment of the present invention. The terminal device 20 of this embodiment includes: a processor 21, a memory 22, and a computer program stored in the memory 22 and executable on the processor 21. When the processor 21 executes the computer program, it implements the steps in the above-described embodiment of the project evaluation method based on feature contribution analysis. Alternatively, when the processor 21 executes the computer program, it implements the functions of each module in the above-described embodiment of the project evaluation device based on feature contribution analysis.

[0053] For example, the computer program may be divided into one or more modules, which are stored in the memory 22 and executed by the processor 21 to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device 20.

[0054] The terminal device 20 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. The terminal device 20 may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art will understand that the schematic diagram is merely an example of the terminal device 20 and does not constitute a limitation on the terminal device 20. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal device 20 may also include input / output devices, network access devices, buses, etc.

[0055] The processor 21 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor 21 is the control center of the terminal device 20, connecting all parts of the terminal device 20 via various interfaces and lines.

[0056] The memory 22 can be used to store the computer programs and / or modules. The processor 21 implements various functions of the terminal device 20 by running or executing the computer programs and / or modules stored in the memory 22 and calling the data stored in the memory 22. The memory 22 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0057] If the modules integrated in the terminal device 20 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processor 21, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0058] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units 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 achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0059] This invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the project evaluation method based on feature contribution analysis as described above.

[0060] Furthermore, embodiments of the present invention also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the project evaluation method based on feature contribution analysis described above.

[0061] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A project evaluation method based on feature contribution analysis, characterized by, The method comprises the following steps: obtaining original data of a project to be evaluated, and constructing a standardized feature vector of the project to be evaluated according to the original data; generating an evaluation result and a decision path of the project to be evaluated by using an improved decision tree evaluation model according to the standardized feature vector, wherein the improved decision tree evaluation model is obtained by improving a decision tree model by using a cost complexity pruning function and a feature weight adjustment factor; performing feature contribution analysis on each feature in the standardized feature vector according to the decision path, and obtaining a contribution vector of the standardized feature vector; generating an explainability text of the contribution vector by using a preset text template and the contribution vector, and generating an evaluation report of the project to be evaluated according to the evaluation result, the decision path, the contribution vector and the explainability text.

2. The project evaluation method based on feature contribution analysis as claimed in claim 1, wherein, The method for obtaining the improved decision tree evaluation model comprises the following steps: obtaining historical original data of a plurality of historical evaluation projects, constructing historical standardized feature vectors according to the historical original data, forming a sample data set by combining the historical standardized feature vectors and corresponding historical evaluation results, and dividing the sample data set into a training set and a validation set; initializing a decision tree model and determining a splitting criterion of the decision tree model; recursively splitting nodes according to the splitting criterion and a feature weight adjustment factor by using the training set until a preset stop condition is met, and generating a preliminary decision tree model; optimizing a pruning coefficient by using a cost complexity pruning function through grid search according to the validation set, and obtaining a pruned decision tree model; verifying a splitting path of the pruned decision tree model according to a preset business rule, adjusting a target threshold corresponding to a target splitting node that does not meet the preset business rule, and if the adjustment is invalid, replacing the target splitting node with an optimal candidate splitting combination that meets the preset business rule, and obtaining an improved decision tree evaluation model.

3. The project evaluation method based on feature contribution analysis as claimed in claim 2, wherein, The method for recursively splitting nodes according to the splitting criterion and the feature weight adjustment factor by using the training set until a preset stop condition is met, and generating a preliminary decision tree model comprises the following steps: determining a splitting score of all candidate splitting combinations of a current node to be split of the decision tree model according to the splitting criterion and the feature weight adjustment factor, wherein the candidate splitting combination is a candidate combination of a candidate splitting feature and a corresponding splitting threshold of the current node to be split; recursively splitting nodes by using the candidate splitting combination with the highest splitting score as an optimal splitting combination until a preset stop condition is met, and generating a preliminary decision tree model.

4. The project evaluation method based on feature contribution analysis as claimed in claim 2, wherein, The method for optimizing a pruning coefficient by using a cost complexity pruning function through grid search according to the validation set, and obtaining a pruned decision tree model comprises the following steps: setting an initial range and a step length of the pruning coefficient, and taking a value of the pruning coefficient according to the initial range and the step length; pruning the preliminary decision tree model according to the pruning coefficient with each value, and obtaining a corresponding pruned model; performing model performance evaluation on each pruned model by using a cost complexity pruning function through grid search according to the validation set; The pruned model corresponding to the pruning coefficient with the optimal model performance evaluation result is taken as the pruned decision tree model.

5. The project evaluation method based on feature contribution analysis as claimed in claim 1, wherein, The feature contribution analysis of each feature in the standardized feature vector according to the decision path comprises: The overall prediction result of the decision path is calculated according to the prediction increment of each split node in the decision path. The total contribution of each feature in the standardized feature vector is calculated according to the prediction increment of each split node to obtain the contribution vector of the standardized feature vector.

6. The feature contribution analysis based project evaluation method of claim 1, wherein, After obtaining the improved decision tree evaluation model, the method further comprises: The improved decision tree evaluation model is audited and managed.

7. An item evaluation device based on feature contribution analysis, characterized by, Comprise: An original data acquisition module is configured to acquire original data of a project to be evaluated and construct a standardized feature vector of the project to be evaluated according to the original data; An evaluation result generation module is configured to generate an evaluation result and a decision path of the project to be evaluated by using an improved decision tree evaluation model according to the standardized feature vector; wherein the improved decision tree evaluation model is obtained by improving a decision tree model by using a cost complexity pruning function and a feature weight adjustment factor; A feature contribution analysis module is configured to perform feature contribution analysis of each feature in the standardized feature vector according to the decision path to obtain a contribution vector of the standardized feature vector; An evaluation report generation module is configured to generate an explainability text of the contribution vector by using a preset text template and the contribution vector; and generate an evaluation report of the project to be evaluated according to the evaluation result, the decision path, the contribution vector and the explainability text.

8. A terminal device, characterized by comprising: The processor executes the computer program to implement the project evaluation method based on feature contribution analysis according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to execute the project evaluation method based on feature contribution analysis according to any one of claims 1-6 when the computer program runs.

10. A computer program product, characterised in that, The computer program product is stored in a storage medium, and the program product is executed by at least one processor to implement the steps of the project evaluation method based on feature contribution analysis according to any one of claims 1-6.

Images