Enterprise comprehensive evaluation method and system based on multi-dimensional data analysis

Abstract

The application discloses a kind of enterprise comprehensive evaluation method and system based on multidimensional data analysis, it is related to data analysis and enterprise management technical field, including obtaining enterprise multidimensional data, standardization processing is carried out to the enterprise multidimensional data, and standardization data set is obtained;Based on the standardization data set, construct multilayer neural network model, and the learnable parameter of the multilayer neural network model is optimized using iterative training mode;The standardization data set is input into trained multilayer neural network model, and enterprise comprehensive score is obtained, based on the enterprise comprehensive score, enterprise is divided into different grades, and evaluation report is generated.The application ensures the comprehensiveness of evaluation and the comparability of data by obtaining the multidimensional data of enterprise.Using innovative multilayer neural network structure, including main network and special branch, can better capture the interaction between different types of indicators and the characteristics of specific category indicators.

Description

Technical Field

[0001] This invention relates to the fields of data analysis and enterprise management technology, and in particular to a comprehensive enterprise evaluation method and system based on multi-dimensional data analysis. Background Technology

[0002] With the deepening of economic globalization and digital transformation, the importance of comprehensive enterprise evaluation in investment decisions, risk management, and strategic planning is becoming increasingly prominent. Traditional enterprise evaluation methods mainly rely on financial indicators and expert experience, making it difficult to comprehensively and objectively reflect an enterprise's overall strength and development potential. In recent years, with the rapid development of big data technology and artificial intelligence algorithms, enterprise evaluation methods based on multi-dimensional data analysis have gradually become a research hotspot. These methods integrate data from multiple dimensions such as finance, operations, marketing, and innovation, and utilize machine learning algorithms for in-depth mining and analysis, aiming to build a more comprehensive and accurate enterprise evaluation system.

[0003] However, existing enterprise evaluation methods based on multi-dimensional data analysis still have many shortcomings. First, data standardization is not perfect, making it difficult to effectively eliminate the differences in dimensions and scale effects between different indicators, thus affecting the accuracy of evaluation results. Second, the construction of evaluation models is often too simplistic or rigid, such as linear weighting or fixed analytic hierarchy process (AHP), which cannot fully capture the non-linear relationships and interactions between indicators. Furthermore, the determination of model parameters relies heavily on subjective experience, lacking adaptive optimization mechanisms and making it difficult to adapt to the evaluation needs of different industries and periods. In addition, existing methods are also insufficient in the interpretability and visualization of evaluation results, failing to provide decision-makers with intuitive and valuable insights. These problems seriously restrict the scientific rigor and practicality of comprehensive enterprise evaluation, urgently requiring a more advanced and reliable evaluation method to overcome these limitations. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem that this invention aims to solve is that the construction of existing evaluation models is often too simple or rigid, such as linear weighting or fixed analytic hierarchy process, which cannot fully capture the nonlinear relationships and interactions between indicators.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a comprehensive enterprise evaluation method based on multi-dimensional data analysis, which includes,

[0008] Acquire multi-dimensional data of the enterprise, and perform standardization processing on the multi-dimensional data of the enterprise to obtain a standardized dataset;

[0009] Based on the standardized dataset, a multi-layer neural network model is constructed, and the learnable parameters of the multi-layer neural network model are optimized by iterative training.

[0010] The standardized dataset is input into the trained multilayer neural network model to obtain a comprehensive enterprise score. Based on the comprehensive enterprise score, enterprises are divided into different levels and an evaluation report is generated.

[0011] As a preferred embodiment of the enterprise comprehensive evaluation method based on multi-dimensional data analysis described in this invention, the multi-dimensional enterprise data includes enterprise financial indicator data, enterprise market performance data, enterprise innovation capability data, and enterprise social responsibility data.

[0012] The enterprise's financial indicators include debt-to-equity ratio, net profit margin, revenue growth rate, and cash flow ratio.

[0013] The enterprise's market performance data includes market share, brand awareness, customer satisfaction, and product sales growth rate.

[0014] The data on enterprise innovation capabilities includes the proportion of R&D investment, the number of patent applications, the number of new product launches, and the number of technology innovation awards.

[0015] The corporate social responsibility data includes environmental protection investment, employee welfare expenditures, charitable donations, and social responsibility report scores.

[0016] As a preferred embodiment of the enterprise comprehensive evaluation method based on multi-dimensional data analysis described in this invention, the standardized dataset includes:

[0017] The enterprise's multi-dimensional data is transformed into indicator data, and the indicator data is then standardized.

[0018] Obtain the standardized value X std ;

[0019] For the standardized value X std Perform exponential smoothing correction and adjust the smoothed standardized value X. smooth X is obtained by performing a logarithmic transformation. log ;

[0020] X log The values ​​are used as the final standardization result to construct a standardized dataset;

[0021] Wherein, the standardized value X std The calculation is expressed by the following formula:

[0022]

[0023] Among them, X stdX represents the standardized values, a represents the original data values, a represents the mixture parameter with a range of [0,1], μ represents the mean, σ represents the standard deviation, Min represents the minimum value of the dataset, and Max represents the maximum value of the dataset.

[0024] The exponential smoothing correction is expressed by the following formula:

[0025]

[0026] Among them, X smooth X is the smoothed, standardized value. std This is the current standardized value. is the standardized value of the previous period, and β is the smoothing coefficient, with a value range of [0,1].

[0027] As a preferred embodiment of the enterprise comprehensive evaluation method based on multi-dimensional data analysis described in this invention, the multi-layer neural network model includes a backbone network and multiple specialized branches;

[0028] The backbone network contains multiple hidden layers with N and N / 2 nodes, respectively.

[0029] The specialized branch contains a hidden layer with N / 4 nodes;

[0030] A hidden layer of the backbone network can be represented by the following formula:

[0031] H i =APReLU(W i H i-1 +b i )

[0032] Among them, W i b is the weight matrix of the i-th hidden layer of the backbone network. i is the bias vector of the i-th hidden layer of the backbone network, and APReLU is the adaptive parameterized ReLU activation function;

[0033] When i <= 1, it represents the first layer of the backbone network, and H i-1 =X, where X is the input feature of the backbone network.

[0034] As a preferred embodiment of the enterprise comprehensive evaluation method based on multi-dimensional data analysis described in this invention, the learnable parameters for optimizing the multi-layer neural network model are expressed by the following formula:

[0035] m t =β1*m t-1 +(1-β1)*g t

[0036]

[0037]

[0038]

[0039] Where, m t ,v t For first-order and second-order momentum estimation, β1 is the first-order momentum decay rate, β2 is the second-order momentum decay rate, and g t Let η be the current gradient, t be the current iteration number, and η be the current gradient. t The current learning rate is ε, a small constant to prevent division by zero, and θ is... t These are the parameters to be optimized. This is the first-order momentum estimate after bias correction. The square root of the second-order momentum estimate. It is the t-th power of the first-order momentum decay rate;

[0040] The current learning rate can be calculated using the following formula:

[0041] η t =η base +(η max -η base )*max(0,1-|t mod TT / 2| / (T / 2))

[0042] Where, η base Based on the learning rate, η max The maximum learning rate is T, where T is the learning rate period length.

[0043] As a preferred embodiment of the enterprise comprehensive evaluation method based on multi-dimensional data analysis described in this invention, the enterprise comprehensive score is obtained by:

[0044] Optimize the output of the backbone network and each specialized branch;

[0045] The outputs of the optimized backbone network and each specialized branch are linearly transformed and then aggregated.

[0046] A comprehensive score for the enterprise is obtained through a single-node output layer.

[0047] As a preferred embodiment of the enterprise comprehensive evaluation method based on multi-dimensional data analysis described in this invention, the optimization of the output of the backbone network and each specialized branch adopts an attention mechanism, expressed by the following formula:

[0048]

[0049] Attention k =sofmax((Q k *Kk T ) / sqrt(d k ))*V k

[0050] in, Let H be the query, key, and value weight matrix for the k-th attention head, and let H be the concatenation of the outputs of all branches. k Q represents the dimensions of the query and key vectors. k ,K k V k This is the query, key, and value matrix for the k-th attention head;

[0051] The linear change is expressed by the following formula:

[0052] MultHead=Concat(Attention1,…,Attention h )*W o

[0053] Where h is the number of attention heads, W o The linear transformation weight matrix for the multi-head attention output;

[0054] The summary is expressed by the following formula:

[0055] S = W s *MultHead+b s

[0056] Among them, W s It is the weight matrix of the aggregation layer, b s is the bias vector of the aggregation layer, and S is the output of the aggregation layer;

[0057] The comprehensive score of the enterprise obtained through the single-node output layer is expressed by the following formula:

[0058] output = sigmoid(W o *S+b o )

[0059] Among them, W o b is the weight vector of the output layer. o is the bias scalar of the output layer, and Output is the final enterprise score output.

[0060] As a preferred embodiment of the enterprise comprehensive evaluation method based on multi-dimensional data analysis described in this invention, the enterprise is divided into different levels through clustering, including:

[0061] The cluster number K is iterated according to a preset interval;

[0062] For each cluster number K, calculate the silhouette coefficient and the sum of squares within each group;

[0063] Plot the relationship between each cluster number K and its corresponding silhouette coefficient, and plot the relationship between each cluster number K and its corresponding within-group sum of squares;

[0064] Analyzing these two graphs, we select the K value with the largest silhouette coefficient and the within-group sum of squares starting to level off as the optimal number of clusters;

[0065] Based on the determined optimal number of clusters, a data point is randomly selected as the first center point;

[0066] For the remaining center points, select the point that is farther away from the existing center point as the new center point;

[0067] Each data point is assigned to the cluster represented by the nearest centroid.

[0068] Recalculate the center point of each cluster;

[0069] Repeat the steps of allocating data points and calculating the center point of each cluster until the position of the center point no longer changes significantly or the preset maximum number of iterations is reached, and finally obtain the clustering result.

[0070] The clustering results are analyzed and rank labels are assigned, and each enterprise is assigned a rank label of its cluster.

[0071] Secondly, embodiments of the present invention provide a comprehensive enterprise evaluation system based on multi-dimensional data analysis, comprising:

[0072] The data acquisition and processing module is used to acquire multi-dimensional data of the enterprise, and to perform standardization processing on the multi-dimensional data of the enterprise to obtain a standardized dataset.

[0073] The model building module is used to build a multi-layer neural network model based on the standardized dataset and optimize the learnable parameters of the multi-layer neural network model by iterative training.

[0074] The rating module is used to input the standardized dataset into the trained multi-layer neural network model to obtain a comprehensive enterprise rating. Based on the comprehensive enterprise rating, the enterprise is divided into different levels and an evaluation report is generated.

[0075] The beneficial effects of this invention are as follows: by acquiring multi-dimensional data from enterprises (financial indicators, market performance, innovation capabilities, and social responsibility) and performing standardization processing, the comprehensiveness of the assessment and the comparability of the data are ensured. The improved hybrid standardization method combines the advantages of Z-score and Min-Max standardization, preserving the original data distribution characteristics while limiting the impact of extreme values, thus improving data quality. An innovative multi-layer neural network structure is adopted, including a backbone network and specialized branches, which can better capture the interactions between different types of indicators and the characteristics of specific categories of indicators. The use of the Adaptive Parametric ReLU activation function and multi-head attention mechanism further improves the model's nonlinear modeling ability and its ability to dynamically adjust the importance of different indicators. The Adaptive Momentum Estimation with Cyclical Learning Rate (AME-CLR) algorithm is used for model training, combining adaptive momentum estimation and periodic learning rate. This not only adaptively adjusts the learning rate of each parameter but also explores a broader parameter space through periodic changes, improving the training effect and stability of the model. Attached Figure Description

[0076] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0077] Figure 1 This is a flowchart of a comprehensive enterprise evaluation method based on multi-dimensional data analysis. Detailed Implementation

[0078] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0079] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0080] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0081] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0082] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0083] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0084] Example 1

[0085] Reference Figure 1 The first embodiment of the present invention provides a comprehensive enterprise evaluation method based on multi-dimensional data analysis, comprising:

[0086] S1: Obtain multi-dimensional data of the enterprise, and perform standardization processing on the multi-dimensional data of the enterprise to obtain a standardized dataset;

[0087] Collect corporate financial data, including debt-to-equity ratio, net profit margin, revenue growth rate, and cash flow ratio; collect corporate market performance data, including market share, brand awareness, customer satisfaction, and product sales growth rate; obtain corporate innovation capability data, including R&D investment ratio, number of patent applications, number of new product launches, and number of technology innovation awards; and summarize corporate social responsibility data, including environmental protection investment, employee welfare expenditures, charitable donations, and social responsibility report scores.

[0088] A corporate database is constructed, storing financial indicators, market performance data, innovation capability data, and social responsibility data. The data in the corporate database is cleaned to remove outliers and missing values, ensuring data quality.

[0089] Furthermore, financial indicator data, market performance data, innovation capability data, and social responsibility data are extracted from the enterprise database; the asset-liability ratio, net profit margin, operating income growth rate, and cash flow ratio in the financial indicator data are standardized in terms of dimensions and converted into percentage form.

[0090] For market performance data, market share and brand awareness are expressed as percentages, customer satisfaction is converted into a 1-10 score system, and product sales growth rate remains in the original percentage format.

[0091] When processing innovation capability data, the R&D investment ratio is converted into a percentage, the number of patent applications and the number of new product launches remain at their original values, and the number of technology innovation awards is counted cumulatively.

[0092] For social responsibility data, environmental protection investment and employee welfare expenditures are converted into percentages of operating revenue, and charitable donations are logarithmically converted. The social responsibility report score remains on a scale of 1-100.

[0093] Calculate the mean μ, standard deviation σ, maximum value Max, and minimum value Min for each indicator. A modified mixed standardization method is used to standardize all indicator data. The calculation formula is as follows:

[0094]

[0095] Among them, X std X represents the standardized values, a represents the original data values, a represents the mixture parameter with a range of [0,1], μ represents the mean, σ represents the standard deviation, Min represents the minimum value of the dataset, and Max represents the maximum value of the dataset.

[0096] A superior, improved hybrid standardization method combines the advantages of Z-score standardization and Min-Max standardization, considering both the overall distribution of the data and limiting the impact of extreme values. By adjusting the α parameter, the dispersion and range of the data can be flexibly balanced according to specific data characteristics and analytical needs. For example, when α is close to 1, more of the original data's distribution characteristics are preserved; when α is close to 0, more emphasis is placed on range scaling. This improved hybrid standardization method yields a standardized dataset that retains the original data characteristics while possessing good statistical properties.

[0097] For positive indicators (such as net profit margin, innovation capability, etc.), simply use X.std For negative indicators (such as the debt-to-equity ratio), use 1-X. std As a result of standardization;

[0098] Furthermore, exponential smoothing correction is performed to further optimize the standardization results. The exponential smoothing correction can be expressed by the following formula:

[0099] X smooth =β*X std +(1-β*X stdprev )

[0100] Among them, X smooth X is the smoothed, standardized value. std X is the current standardized value. stdprev The value is the standardized value of the previous period, and β is the smoothing coefficient, which takes the value [0,1].

[0101] To reduce the impact of extreme values, a logarithmic transformation is performed on the smoothed, standardized data. The transformation formula is as follows:

[0102] X log =log(1+X) smooth )

[0103] X log The values ​​are used as the final standardization result to construct a standardized dataset;

[0104] S2: Based on the standardized dataset, construct a multi-layer neural network model and optimize the learnable parameters of the multi-layer neural network model using an iterative training method;

[0105] Specifically, determine the dimensions of the standardized dataset, which is the sum of the number of indicators included in each of the financial indicators, market performance, innovation capabilities, and social responsibility, denoted as N; design the input layer of the neural network, with the number of nodes equal to the dimension N of the standardized dataset, and each node corresponding to a standardized indicator.

[0106] An innovation structure is constructed, consisting of a backbone network and four specialized branches (financial, market, innovation, and social responsibility). The backbone network contains two hidden layers with N and N / 2 nodes respectively, used to capture the global interactions between various indicators. Each specialized branch contains one hidden layer with N / 4 nodes, focusing on processing indicators of a specific category. This structural design takes into account the diversity of enterprise evaluation indicators, and improves the model's feature extraction capability by handling different types of indicators through specialized branch networks.

[0107] For example, the first hidden layer of the backbone layer can be represented by the following formula:

[0108] H1 = APReLU(W1X + b1)

[0109] Where W1 is an N×N weight matrix, b1 is the bias vector of the first hidden layer of the backbone layer, and X is the input feature of the backbone layer.

[0110] The second hidden layer of the backbone layer can be represented by the following formula:

[0111] H2 = APReLU(W2H1 + b2)

[0112] Where W2 is an N / 2×N weight matrix, and b2 is the bias vector of the second hidden layer of the backbone layer. The hidden layer of a specialized branch (taking the finance branch as an example, other branches are similar) can be represented by the following formula:

[0113] F = APReLU(W f X f +b f )

[0114] Among them, X f It is a financially related input feature, W f It is N / 4×N f The weight matrix (N) f (This refers to the quantity of financial characteristics), b f is the bias vector, and F is the output of the hidden layer of the financial branch.

[0115] Preferably, the Adaptive Parametric ReLU activation function is applied to the output of each hidden layer. Each APReLU has its own independent α and β parameters. Through learnable parameters, APReLU can dynamically adjust its shape according to the needs of different layers and different training stages. The independent learning of its positive and negative slopes allows the model to handle various nonlinear relationships more flexibly. APReLU can be expressed as follows:

[0116] f(x)=α*max(0,x)+β*min(0,x)

[0117] Here, α is the learnable slope parameter for the positive part, and β is the learnable slope parameter for the negative part. The initial values ​​are set to 1 and 0.01 respectively, so that the function approximates Leaky ReLU in the early stage of training, ensuring good gradient flow.

[0118] Furthermore, a multi-head attention mechanism is applied to the outputs of the backbone network and each specialized branch. This attention mechanism dynamically adjusts the importance of different indicators. Ideally, it fully utilizes the multi-branch structure of the backbone network and four specialized branches, enabling the model to automatically adjust the weights of each indicator based on the characteristics of the current input data, highlighting key information and thus improving the model's adaptability to different enterprise characteristics. Specifically, for each head k:

[0119]

[0120] Attention k =sofmax((Q k *K k T ) / sqrt(d k ))*V k

[0121] in, Let H be the query, key, and value weight matrix for the k-th attention head, and let H be the concatenation of the outputs of all branches. k Q represents the dimensions of the query and key vectors. k ,K k V k Let be the query, key, and value matrix of the k-th attention head.

[0122] Finally, the outputs of all the heads are concatenated and subjected to a linear transformation:

[0123] MultHead=Concat(Attention1,…,Attention h )*W o

[0124] Where h is the number of attention heads, W o This is the linear transformation weight matrix for the multi-head attention output. Further, the features processed by attention are summarized:

[0125] S = W s *MultHead+b s

[0126] Among them, W s It is the weight matrix of the aggregation layer, b s is the bias vector of the aggregation layer, and S is the output of the aggregation layer.

[0127] Then, through the single-node output layer, a comprehensive score for the enterprise is generated:

[0128] output = sigmoid(W o *S+b o )

[0129] Among them, W o b is the weight vector of the output layer. o is the bias scalar of the output layer, and Output is the final enterprise score output.

[0130] Finally, a comprehensive loss function is used to optimize the model's prediction accuracy while preventing overfitting. This loss function can be further optimized by adding additional loss terms, such as domain-knowledge-based constraints, to ensure that the model's predictions conform to the basic principles of enterprise evaluation. For example, monotonicity constraints can be added to ensure that improvements in certain key indicators necessarily lead to higher scores. This part of the loss function can be expressed as follows:

[0131] L = MSE(Qutput,Target) + λ1 * L1 reg +λ2*L2 reg

[0132] Where MSE is the mean squared error, L1 reg and L2 reg These are L1 and L2 regularization terms, with λ1 and λ2 being the weighting coefficients.

[0133] It should be noted that throughout the training process, the Adaptive Momentum Estimation with Cyclical Learning Rate (AME-CLR) algorithm is used to update all learnable parameters in the network. For each parameter θ:

[0134] m t =β1*m t-1 +(1-β1)*g t

[0135]

[0136]

[0137]

[0138] Where, m t ,v t For first-order and second-order momentum estimation, β1 is the first-order momentum decay rate, β2 is the second-order momentum decay rate, and g t Let η be the current gradient, t be the current iteration number, and η be the current gradient. t The current learning rate is ε, a small constant to prevent division by zero, and θ is... t These are the parameters to be optimized. This is the first-order momentum estimate after bias correction. The square root of the second-order momentum estimate. It is the t-th power of the first-order momentum decay rate.

[0139] It should be noted that the current learning rate can be calculated using the following formula:

[0140] η t =η base+(η max -η base )*max(0,1-|t mod TT / 2| / (T / 2))

[0141] Where, η base Based on the learning rate, η max The maximum learning rate is T, where T is the learning rate period length.

[0142] A superior approach is to combine adaptive momentum estimation and a periodic learning rate. This allows the model to adaptively adjust the learning rate for each parameter while exploring a broader parameter space through periodic changes. The periodic learning rate helps the model escape local optima, increasing the likelihood of finding the global optimum. Adaptive momentum estimation considers historical gradient information, helping to maintain training stability even with significant metric fluctuations. In implementation, the period length T can be dynamically adjusted based on the validation set performance during training to further enhance the algorithm's adaptability.

[0143] S3: Input the standardized dataset into the trained multilayer neural network model to obtain the comprehensive enterprise score, and classify the enterprises into different levels based on the comprehensive enterprise score.

[0144] First, collect and organize the comprehensive scores for all companies. These scores are calculated using the innovative multi-layer neural network model built earlier. Ensure all scores are standardized for subsequent processing. Create a dataset containing company identifiers (such as company names or IDs) and their corresponding comprehensive scores.

[0145] Considering the characteristics of enterprise ratings, an improved K-means++ algorithm is used. This algorithm optimizes the traditional K-means algorithm, better handling the uneven distribution of enterprise ratings. K-means++ improves the stability and effectiveness of clustering by optimizing the selection of initial centroids.

[0146] Furthermore, the optimal number of clusters is determined using a combination of the silhouette coefficient and the elbow method. The process is as follows:

[0147] a. Iterate through the number of clusters K from 2 to 10.

[0148] b. For each K value, run the K-means++ algorithm to calculate the silhouette coefficient and within-group sum of squares (WCSS).

[0149] c. Draw a graph showing the relationship between K and the profile coefficient, and a graph showing the relationship between K and WCSS.

[0150] d. Analyze these two graphs and select the K value with the largest silhouette coefficient and where WCSS begins to flatten out as the optimal number of clusters. The formula for calculating the silhouette coefficient is:

[0151] s(i)=(b(i)-a(i)) / max(a(i),b(i))

[0152] Where a(i) is the average distance between sample i and other samples in the same cluster, and b(i) is the average distance between sample i and all samples in the nearest neighbor cluster.

[0153] Based on the determined optimal number of clusters K, the K-means++ algorithm is run to perform clustering. The specific steps are as follows:

[0154] Randomly select a data point as the first center point.

[0155] For the remaining center points, select the point that is farther away from the existing center point as the new center point.

[0156] Each data point is assigned to the cluster represented by the nearest centroid.

[0157] Recalculate the centroid (mean) of each cluster.

[0158] Repeat the steps of allocating data points and calculating the center point of each cluster until the center point position no longer changes significantly or the preset maximum number of iterations is reached.

[0159] Then, multiple metrics were used to evaluate the clustering effect, including the sum of squares within clusters, silhouette coefficient, Calinski-Harabasz index, and Davies-Bouldin index. Each cluster was then analyzed and interpreted in detail: common characteristics of enterprises within each cluster, such as size and industry distribution, were analyzed, and a meaningful rating label was assigned to each cluster based on the cluster characteristics, such as "excellent", "good", "medium", and "needs improvement".

[0160] Finally, a detailed clustering report is generated, including:

[0161] Each company's rating label.

[0162] The relative position of an enterprise within its cluster.

[0163] Statistical characteristics of each cluster (such as mean score, standard deviation, number of firms).

[0164] Furthermore, this embodiment also provides a comprehensive enterprise evaluation system based on multi-dimensional data analysis, including:

[0165] The data acquisition and processing module is used to acquire multi-dimensional data of the enterprise, and to perform standardization processing on the multi-dimensional data of the enterprise to obtain a standardized dataset.

[0166] The model building module is used to build a multi-layer neural network model based on the standardized dataset and optimize the learnable parameters of the multi-layer neural network model by iterative training.

[0167] The rating module is used to input the standardized dataset into the trained multi-layer neural network model to obtain a comprehensive enterprise rating. Based on the comprehensive enterprise rating, the enterprise is divided into different levels and an evaluation report is generated.

[0168] This embodiment also provides a computer device applicable to the enterprise comprehensive evaluation method based on multi-dimensional data analysis, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the enterprise comprehensive evaluation method based on multi-dimensional data analysis as proposed in the above embodiment.

[0169] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0170] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the enterprise comprehensive evaluation method based on multi-dimensional data analysis as proposed in the above embodiments.

[0171] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0172] Example 2

[0173] The second embodiment of the present invention provides a comprehensive enterprise evaluation method based on multi-dimensional data analysis. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0174] This embodiment aims to verify the effectiveness and advantages of the proposed enterprise evaluation method. To this end, 100 enterprises of different sizes and industries were selected as research subjects, representing companies from multiple sectors including manufacturing, services, and technology.

[0175] First, multi-dimensional data on the companies was collected following step S1. Detailed data on the financial indicators, market performance, innovation capabilities, and social responsibility of these 100 companies were obtained through publicly available financial reports, market research reports, patent databases, and corporate social responsibility reports. Specifically, financial indicators such as debt-to-equity ratio, net profit margin, revenue growth rate, and cash flow ratio were collected; market performance data such as market share, brand awareness, customer satisfaction, and product sales growth rate were collected; innovation capability data such as R&D investment ratio, number of patent applications, number of new product launches, and number of technological innovation awards were collected; and social responsibility data such as environmental protection investment, employee welfare expenditures, charitable donations, and social responsibility report scores were collected. The raw data and processed data are shown in Tables 1 and 2 below, respectively.

[0176] Table 1: Raw Data for Enterprise Evaluation Indicators (Partial)

[0177]

[0178] Table 2: Standardized Enterprise Evaluation Indicator Data (Partial)

[0179]

[0180]

[0181] The comprehensive scores of 100 companies were obtained using the constructed multi-layer neural network model, and the results were divided into levels, as shown in Table 3 below:

[0182] Table 3: Enterprise Comprehensive Scores and Rating Results (Partial)

[0183] Enterprise ID Overall score grade E001 0.67 good E002 0.59 medium E003 0.45 Needs improvement E004 0.77 excellent E005 0.52 medium E006 0.85 excellence

[0184] Analysis of the data in Tables 1, 2, and 3 yields the following conclusions, which fully demonstrate the innovation and advantages of the method of this invention:

[0185] Comparing Tables 1 and 2, we can see that after standardization, the indicators with different dimensions and scales were transformed into the same range (between 0 and 1). This process makes different indicators comparable and effectively solves the problem of inconsistent dimensions between different indicators in the original data. For example, "number of patent applications" and "amount of charitable donations," which originally had very different units and orders of magnitude, can be directly compared after standardization. This standardization process provides more suitable input for subsequent neural network models, helping to improve the training effect and prediction accuracy of the models.

[0186] By comparing the standardized data in Table 2 and the comprehensive scores in Table 3, it can be seen that this method can effectively capture the nonlinear relationships between indicators. For example, the standardized scores of companies E004 and E006 are very close on most indicators, but the comprehensive score of E006 (0.85) is significantly higher than that of E004 (0.77). This shows that the multi-layer neural network model used in this method can learn the complex interactions between indicators, rather than simple linear combinations. This nonlinear modeling capability is a significant advantage of this method compared to traditional linear evaluation methods.

[0187] As can be seen from the results in Table 3, this method can provide reasonable evaluation results even when dealing with companies of different industries and sizes. For example, there is a significant difference between the lowest-scoring company E003 (0.45 points) and the highest-scoring company E006 (0.85 points), but both are reasonably classified into their respective levels. This demonstrates the strong adaptability and robustness of this method when dealing with diverse enterprise data.

[0188] In summary, this embodiment clearly demonstrates the significant innovation and advantages of the method of this invention in the field of enterprise evaluation. It not only comprehensively and objectively evaluates multiple aspects of an enterprise, but also captures the complex relationships between indicators through advanced data processing and modeling techniques, providing more accurate and meaningful evaluation results. The application of this method will provide a more reliable basis for enterprise management, investment decisions, and policy formulation, driving technological progress in the field of enterprise evaluation.

[0189] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A comprehensive enterprise evaluation method based on multi-dimensional data analysis, characterized in that: include, Acquire multi-dimensional data of the enterprise, and perform standardization processing on the multi-dimensional data of the enterprise to obtain a standardized dataset; Based on the standardized dataset, a multi-layer neural network model is constructed, and the learnable parameters of the multi-layer neural network model are optimized by iterative training. The standardized dataset is input into the trained multilayer neural network model to obtain a comprehensive enterprise score. Based on the comprehensive enterprise score, enterprises are divided into different levels and an evaluation report is generated.

2. The enterprise comprehensive evaluation method based on multi-dimensional data analysis as described in claim 1, characterized in that: The multi-dimensional data of the enterprise includes enterprise financial indicators, enterprise market performance, enterprise innovation capabilities, and enterprise social responsibility. The enterprise's financial indicators include debt-to-equity ratio, net profit margin, revenue growth rate, and cash flow ratio. The enterprise's market performance data includes market share, brand awareness, customer satisfaction, and product sales growth rate. The data on enterprise innovation capabilities includes the proportion of R&D investment, the number of patent applications, the number of new product launches, and the number of technology innovation awards. The corporate social responsibility data includes environmental protection investment, employee welfare expenditures, charitable donations, and social responsibility report scores.

3. The enterprise comprehensive evaluation method based on multi-dimensional data analysis as described in claim 2, characterized in that: The standardized dataset obtained includes: The enterprise's multi-dimensional data is transformed into indicator data, and the indicator data is then standardized. Obtain the standardized value X std ; For the standardized value X std Perform exponential smoothing correction and adjust the smoothed standardized value X. smooth X is obtained by performing a logarithmic transformation. log ; X log The values ​​are used as the final standardization result to construct a standardized dataset; Wherein, the standardized value X std The calculation is expressed by the following formula: Among them, X std X represents the standardized values, a represents the original data values, a represents the mixture parameter with a range of [0,1], μ represents the mean, σ represents the standard deviation, Min represents the minimum value of the dataset, and Max represents the maximum value of the dataset. The exponential smoothing correction is expressed by the following formula: Among them, X smooth X is the smoothed, standardized value. std X is the current standardized value. stdprev is the standardized value of the previous period, and β is the smoothing coefficient, with a value range of [0,1].

4. The enterprise comprehensive evaluation method based on multi-dimensional data analysis as described in claim 3, characterized in that: The multi-layer neural network model includes a backbone network and multiple specialized branches; The backbone network contains multiple hidden layers with N and N / 2 nodes, respectively. The specialized branch contains a hidden layer with N / 4 nodes; A hidden layer of the backbone network can be represented by the following formula: H i =APReLU(W i H i-1 +b i ) Among them, W i b is the weight matrix of the i-th hidden layer of the backbone network. i is the bias vector of the i-th hidden layer of the backbone network, and APReLU is the adaptive parameterized ReLU activation function; When i <= 1, it represents the first layer of the backbone network, and H i-1 =X, where X is the input feature of the backbone network.

5. The enterprise comprehensive evaluation method based on multi-dimensional data analysis as described in claim 4, characterized in that: The learnable parameters for optimizing the multilayer neural network model are expressed by the following formula: m t =β1*m t-1 +(1-β1)*g t Where, m t ,v t For first-order and second-order momentum estimation, β1 is the first-order momentum decay rate, β2 is the second-order momentum decay rate, and g t Let η be the current gradient, t be the current iteration number, and η be the current gradient. t The current learning rate is ε, a small constant to prevent division by zero, and θ is the learning rate. t These are the parameters to be optimized. This is the first-order momentum estimate after bias correction. The square root of the second-order momentum estimate. It is the t-th power of the first-order momentum decay rate; The current learning rate can be calculated using the following formula: or t =the base +(the max -or base )*max(0.1-|t mod TT / 2| / (T / 2)) Where, η base Based on the learning rate, η max The maximum learning rate is T, where T is the learning rate period length.

6. The enterprise comprehensive evaluation method based on multi-dimensional data analysis as described in claim 5, characterized in that: The comprehensive score for the enterprise includes: Optimize the output of the backbone network and each specialized branch; The outputs of the optimized backbone network and each specialized branch are linearly transformed and then aggregated. A comprehensive score for the enterprise is obtained through a single-node output layer.

7. The enterprise comprehensive evaluation method based on multi-dimensional data analysis as described in claim 6, characterized in that: The output of the backbone network and each specialized branch is optimized using an attention mechanism, expressed as follows: Attention k =sofmax((Q k *K k T ) / sqrt(d k ))*V k in, Let H be the query, key, and value weight matrix for the k-th attention head, and let H be the concatenation of the outputs of all branches. k Q represents the dimensions of the query and key vectors. k ,K k V k This is the query, key, and value matrix for the k-th attention head; The linear change is expressed by the following formula: MultHead=Concat(Attention1,…,Attention h )*W o Where h is the number of attention heads, W o The linear transformation weight matrix for the multi-head attention output; The summary is expressed by the following formula: S=W s *MultHead+b s Among them, W s It is the weight matrix of the aggregation layer, b s is the bias vector of the aggregation layer, and S is the output of the aggregation layer; The comprehensive score of the enterprise obtained through the single-node output layer is expressed by the following formula: output=sigmoid(W o *S+b o ) Among them, W o b is the weight vector of the output layer. o is the bias scalar of the output layer, and Output is the final enterprise score output.

8. The enterprise comprehensive evaluation method based on multi-dimensional data analysis as described in claim 7, characterized in that: Classifying enterprises into different levels is achieved through clustering, including: The cluster number K is iterated according to a preset interval; For each cluster number K, calculate the silhouette coefficient and the sum of squares within each group; Plot the relationship between each cluster number K and its corresponding silhouette coefficient, and plot the relationship between each cluster number K and its corresponding within-group sum of squares; Analyzing these two graphs, we select the K value with the largest silhouette coefficient and the within-group sum of squares starting to level off as the optimal number of clusters; Based on the determined optimal number of clusters, a data point is randomly selected as the first center point; For the remaining center points, select the point that is farther away from the existing center point as the new center point; Each data point is assigned to the cluster represented by the nearest centroid. Recalculate the center point of each cluster; Repeat the steps of allocating data points and calculating the center point of each cluster until the position of the center point no longer changes significantly or the preset maximum number of iterations is reached, and finally obtain the clustering result. The clustering results are analyzed and rank labels are assigned, and each enterprise is assigned a rank label of its cluster.

9. A comprehensive enterprise evaluation system based on multi-dimensional data analysis, based on the comprehensive enterprise evaluation method based on multi-dimensional data analysis as described in any one of claims 1 to 8, characterized in that: include, The data acquisition and processing module is used to acquire multi-dimensional data of the enterprise, and to perform standardization processing on the multi-dimensional data of the enterprise to obtain a standardized dataset. The model building module is used to build a multi-layer neural network model based on the standardized dataset and optimize the learnable parameters of the multi-layer neural network model by iterative training. The rating module is used to input the standardized dataset into the trained multi-layer neural network model to obtain a comprehensive enterprise rating. Based on the comprehensive enterprise rating, the enterprise is divided into different levels and an evaluation report is generated.

Images