Supplier intelligent portrait construction method based on multi-source heterogeneous data fusion

Abstract

The invention discloses a supplier intelligent portrait construction method based on multi-source heterogeneous data fusion, and the method is characterized in that the method comprises the steps: obtaining supplier multi-source heterogeneous data; wherein the supplier multi-source heterogeneous data comprises graph data, time sequence data, text data and structured data; inputting supplier multi-source heterogeneous data into the trained heterogeneous feature extraction model; splicing and fusing the obtained relation feature, dynamic feature, semantic feature and numerical feature to obtain a first fusion feature; splicing and fusing the fusion feature and the user demand feature to obtain a second fusion feature; obtaining a corresponding supplier qualification grading result according to the second fusion feature; and constructing a supplier qualification radar map according to a supplier qualification grading result. According to the method, multi-source heterogeneous data are effectively integrated, features can be extracted from multi-dimensional multi-source data, the problems of dimension evaluation complexity and user demand diversity are solved, and data can be processed and updated in real time to ensure timeliness.

Description

Technical Field

[0001] The present invention belongs to the field of big data processing, and specifically relates to a method for constructing supplier intelligent portraits based on multi-source heterogeneous data fusion. Background Art

[0002] In modern supply chain management, suppliers are a crucial link in a company's production and operations, and their selection and management are directly linked to its core competitiveness. With the continuous expansion of supply chains and the increasing complexity of business scenarios, traditional supplier management approaches are gradually becoming increasingly limited. These methods often rely on manual experience and simple historical data analysis, making it difficult to comprehensively and dynamically reflect suppliers' performance capabilities, quality levels, and potential risks. This in turn leads to inefficient decision-making and increased management risks. This is particularly true in the power equipment industry, which places extremely high demands on quality, delivery time, and technological innovation. In practice, suppliers from different regions and backgrounds exhibit significant differences in technical capabilities, quality standards, and production scale. Traditional static assessment methods struggle to accurately identify and quantify these differences, often leading to biased decisions. This deficiency can not only impact a company's product quality and production schedule, but can also negatively impact its overall operations and reputation. Therefore, supplier management is evolving from traditional linear analysis to intelligent and dynamic approaches. At the same time, the development of big data and deep learning technologies has provided new approaches and tools for supplier management. However, building supplier profiles still faces challenges, including diverse data sources, complex assessment dimensions, and dynamic requirements. At the same time, traditional methods have limited processing capabilities for large-scale, multi-dimensional data, and are unable to meet the accuracy and real-time requirements of modern supply chain management.

[0003] In particular, current supplier management still faces numerous challenges in data integration and intelligent analysis. First, supplier data comes from a wide range of sources, covering multiple dimensions such as contract performance records, quality indicators, and financial status. This data is distributed across different systems and platforms, lacking unified standards. This leads to varying degrees of data structuring and usability, complicating standardized data processing and in-depth analysis. Traditional supplier evaluation methods often rely on static indicators, making it difficult to dynamically track changes in supplier capabilities, resulting in insufficient timeliness and pertinence in evaluation results. Summary of the Invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a method for constructing supplier intelligent portraits based on multi-source heterogeneous data fusion.

[0005] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0006] In a first aspect, the present invention provides a method for constructing a supplier intelligent profile based on multi-source heterogeneous data fusion, the method comprising:

[0007] Acquire multi-source heterogeneous data of suppliers; wherein the multi-source heterogeneous data of suppliers includes: graph data, time series data, text data and structured data;

[0008] Inputting the supplier's multi-source heterogeneous data into the trained heterogeneous feature extraction model;

[0009] Obtain the relational features corresponding to the graph data, the dynamic features corresponding to the time series data, the semantic features corresponding to the text data, and the numerical features corresponding to the structured data in the supplier's multi-source heterogeneous data;

[0010] Concatenate and fuse the relational feature, the dynamic feature, the semantic feature, and the numerical feature to obtain a first fused feature;

[0011] Concatenate and fuse the fusion feature and the user demand feature to obtain a second fusion feature;

[0012] Obtaining a corresponding supplier qualification grading result according to the second fusion feature;

[0013] A supplier qualification radar chart is constructed based on the supplier qualification grading results to characterize the supplier's qualifications in multiple scoring dimensions.

[0014] Optionally, the trained heterogeneous feature extraction model includes: a trained graph convolutional network, a trained Transformer, a trained BERT and a trained convolutional neural network.

[0015] Optionally, the trained graph convolutional network is used to extract relational features corresponding to graph data in the supplier's multi-source heterogeneous data; the trained Transformer is used to extract dynamic features of time series data in the supplier's multi-source heterogeneous data; the trained graph convolutional network is used to extract semantic features corresponding to text data in the supplier's multi-source heterogeneous data; the trained graph convolutional network is used to extract numerical features corresponding to structured data in the supplier's multi-source heterogeneous data.

[0016] Optionally, before the step of concatenating and fusing the fused feature and the user demand feature to obtain a second fused feature, the step includes:

[0017] Obtain user demand data;

[0018] The user demand data is input into the trained BERT to obtain the user demand features.

[0019] Optionally, obtaining a corresponding supplier qualification grading result according to the second fusion feature includes:

[0020] Inputting the second fused features into the trained supplier grading model;

[0021] Obtain the supplier qualification grading result.

[0022] Optionally, constructing a supplier qualification radar chart according to the supplier qualification grading result includes:

[0023] Standardizing the supplier qualification grading results to obtain standardized results;

[0024] Using a Sigmoid function to correct the normalized result to obtain a corrected result;

[0025] Obtaining a weighted score based on the revised result and the weights corresponding to the various scoring dimensions in the revised result;

[0026] The supplier qualification radar chart is constructed according to the weighted scores.

[0027] Optionally, the corrected result is expressed as follows:

[0028]

[0029] Among them, s i ' represents the score of the supplier in the i-th scoring dimension in the corrected result, Sigmoid(.) represents the Sigmoid function, k is a parameter that controls the steepness of the Sigmoid function, and e is a natural constant.

[0030] Optionally, the weighted score is expressed as follows:

[0031]

[0032] Among them, s weighted represents the weighted score of the supplier, n is the number of scoring dimensions, ω i Indicates the weight of the supplier in the i-th scoring dimension in the revised result.

[0033] The technical solutions provided by the embodiments of the present invention may have the following beneficial effects:

[0034] In the above technical solution, a heterogeneous feature extraction model is designed to perform heterogeneous feature extraction, and the extracted multi-source features are fused to effectively integrate multi-source heterogeneous data, thereby solving the processing difficulties brought about by data heterogeneity; and it is able to extract features from multi-dimensional multi-source data and perform personalized evaluation based on user needs, thereby solving the problems of complexity of evaluation dimensions and diversity of user needs; by introducing a heterogeneous feature extraction model designed based on deep learning, it is able to process and update data in real time, automatically adapt to changes in supplier behavior, and ensure the timeliness and dynamism of the portrait.

[0035] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of a method for constructing a supplier intelligent profile based on multi-source heterogeneous data fusion provided by an embodiment of the present invention;

[0037] Figure 2 is a schematic diagram of feature splicing and fusion provided by an embodiment of the present invention;

[0038] Figure 3 This is a schematic diagram of a supplier qualification radar chart provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to facilitate the understanding of the present invention, the related art and the inventive concept of the present invention are briefly described first.

[0040] Intelligent supplier profiling involves leveraging machine learning or deep learning technologies to analyze various supplier information (such as historical transaction data, quality control indicators, and production capacity) to construct a comprehensive, dynamic, and accurate multi-dimensional description of suppliers. This process uses a neural network model to automatically learn underlying patterns in supplier data and generate a "profile" that comprehensively reflects characteristics such as supplier fulfillment capabilities and quality levels. This intelligent profile not only optimizes supplier selection and management but also possesses the ability to self-update and continuously optimize to adapt to market changes and business needs, thereby helping companies make more informed decisions. Researchers have proposed a method for constructing multi-dimensional behavioral profiles and achieving efficient detection of Android malware. This method, focusing on Android malware behavior, extracts an information-rich behavioral dataset encompassing both static and dynamic behaviors, defines behavioral labels for different types of malware behavior, and integrates machine learning algorithms to achieve label association. The paper also investigates an Android malware behavioral profiling architecture based on behavioral analysis and designs an optimized random forest algorithm that combines behavioral profiling with malware detection. Other researchers have proposed a knowledge graph-based method for constructing user profiles for power company suppliers. This approach addresses the low granularity and poor accuracy of traditional user profile construction methods, which ignore the correlations between different attribute features and suffer from biased analytical perspectives. This approach also proposes a student competency profile system based on big data, aiming to help students more intuitively understand their abilities and provide visual data support for improving their learning process. Research results show that student competency profiles can help students self-monitor, enhance their self-efficacy, and adjust their learning behaviors in a timely manner, thereby improving learning efficiency and effectiveness. Other researchers have proposed a supplier profile generation method based on big data analysis and deep learning, aiming to help users make informed decisions in core processes such as procurement and contract signing. This method establishes a label element analysis model for each level of the power company's vertical labeling system, dividing the labeling system into target, standard, and solution layers, and constructing a hierarchical structure. This method also proposes a supplier management system based on portrait technology to address the challenges faced by power companies in developing information and intelligent supplier management. This method utilizes portrait technology to extract label information from massive data for data analysis, innovatively combining portrait technology with supplier management theory to design a power supplier profile system. The aforementioned works explored various aspects of portrait construction and demonstrated strong performance.

[0041] However, existing supplier portrait construction technologies mainly rely on traditional data mining methods and shallow machine learning models, which have obvious shortcomings when processing complex, multi-source, and heterogeneous data. First, traditional methods have limited processing capabilities for unstructured data (such as text and images), making it difficult to fully capture the multi-dimensional characteristics of suppliers, resulting in one-sidedness of portrait information. Secondly, these methods are not adaptable enough to cope with dynamic changes and real-time updates of data, and are unable to reflect changes in supplier status in a timely manner, affecting the accuracy of decision-making. In addition, traditional technologies have limited capabilities in mining deep associations and complex patterns between data, making it difficult to reveal the underlying laws of supplier behavior, limiting the depth and accuracy of the portrait. Therefore, existing technologies are difficult to meet the growing demand of enterprises for intelligent and refined supplier management. Therefore, the present invention proposes a supplier intelligent portrait construction method based on multi-source heterogeneous data fusion to solve this technical problem.

[0042] Figure 1 This is a flow chart of a method for constructing a supplier intelligent portrait based on multi-source heterogeneous data fusion provided by an embodiment of the present invention. Figure 1 As shown, the method may include the following steps:

[0043] S101. Acquire supplier multi-source heterogeneous data; wherein the supplier multi-source heterogeneous data includes: graph data, time series data, text data, and structured data.

[0044] It is understandable that multi-source heterogeneous data may include information such as the supplier's main business, risk level, past transaction records, etc. These data are often in different modalities. Therefore, in order to facilitate further analysis and processing through deep learning models, it is necessary to design corresponding models for each type of data to extract corresponding features and embed them into fixed-length vectors.

[0045] S102. Input the supplier's multi-source heterogeneous data into the trained heterogeneous feature extraction model.

[0046] Optionally, the trained heterogeneous feature extraction model includes: a trained graph convolutional network, a trained Transformer, a trained BERT, and a trained convolutional neural network.

[0047] Optionally, the trained graph convolutional network is used to extract relational features corresponding to graph data in supplier multi-source heterogeneous data; the trained Transformer is used to extract dynamic features of time series data in supplier multi-source heterogeneous data; the trained graph convolutional network is used to extract semantic features corresponding to text data in supplier multi-source heterogeneous data; the trained graph convolutional network is used to extract numerical features corresponding to structured data in supplier multi-source heterogeneous data.

[0048] It is understandable that the heterogeneous feature extraction model can fully utilize the supplier's multi-source heterogeneous data (such as structured data, text data, time series data, and graph data) by designing a specialized deep learning model for each data type to extract its unique features. For structured data, a trained convolutional neural network (CNN) can be used to extract numerical features; for text data, a trained BERT can be used to extract semantic features; for time series data, a trained Transformer can be used to extract dynamic features; and for graph data, a trained graph convolutional network (GCN) can be used to extract relational features. The goal of heterogeneous feature extraction is to convert raw data into high-dimensional feature vectors to provide high-quality input for subsequent multimodal fusion and task output.

[0049] S103: Obtain relational features corresponding to graph data, dynamic features corresponding to time series data, semantic features corresponding to text data, and numerical features corresponding to structured data in the supplier's multi-source heterogeneous data.

[0050] S104: Concatenate and fuse the relational features, dynamic features, semantic features, and numerical features to obtain a first fused feature.

[0051] It is understandable that Figure 2 is a schematic diagram of feature splicing and fusion provided by an embodiment of the present invention, such as Figure 2 As shown in the figure, features from different modalities are integrated, using feature concatenation as a fusion method. The multimodal fusion output of relational, dynamic, semantic, and numerical features is a comprehensive feature vector, the first fused feature, which fully reflects the multidimensional characteristics of suppliers and supports subsequent tasks.

[0052] For example, the numerical features extracted by the trained CNN of structured data can be expressed as f struct ∈R 64 , the semantic features extracted by BERT after training the text data can be expressed as f text ∈R 128 , the dynamic features extracted by the trained Transformer of time series data can be expressed as f time ∈R 32 , the relational features extracted by the trained GCN of the graph data can be expressed as f graph ∈R 128 It can ensure that the features of each modality have the same vector size in all dimensions except data merging. The feature vectors of each modality can be connected in the feature dimension, and the first fusion feature obtained can be expressed as f supplier =[f struct ;f text ;f time;f graph ]∈R 64+128+32+128 =R 352 .

[0053] S105: Concatenate and fuse the fusion feature and the user demand feature to obtain a second fusion feature.

[0054] It's understandable that multimodal feature fusion design consists of two steps: the first is to fuse the supplier's multimodal features, and the second is to combine the fused features with the user's demand characteristics. Multimodal feature fusion design effectively integrates supplier characteristics and user demand characteristics to generate a comprehensive and accurate supplier profile, providing more precise support for personalized recommendations.

[0055] S106. Obtain corresponding supplier qualification grading results based on the second fusion feature.

[0056] Optionally, S106 may include:

[0057] Inputting the second fused features into the trained supplier grading model;

[0058] Get the supplier qualification grading results.

[0059] It is understandable that the second fused feature is fed into the trained supplier grading model for grading on a single dimension. The trained supplier grading model is a classification model whose input is the second fused feature and whose output is the supplier qualification grading result. This model is an MLP network consisting of multiple linear and activation layers. The output of each layer is added to the input of the previous layer to accelerate model training and avoid the vanishing gradient problem. An attention layer is then added to enable the model to better capture information within the data. Finally, the classification task is performed, and the output is the model's capability grading for the requirement. It is worth noting that the model training requires independent training of the grading model for each evaluation dimension. After the second fused feature is fed into the trained supplier grading model, the classification task is performed. The final output of the supplier qualification grading result is the probability of different evaluation results for the supplier on the requirement. The rating with the highest probability is the supplier's rating for the requirement.

[0060] It's worth noting that the supplier grading model consists of an input layer, hidden layers, and an output layer. Specifically, the model input layer accepts the second fused feature, which has a dimension of 480. The model contains three hidden layers, with the output of each layer connected to the input of the previous layer to further improve the model's learning efficiency and avoid the vanishing gradient problem. At the same time, the output of each layer undergoes a nonlinear transformation using the ReLU activation function. The model output layer outputs the supplier's evaluation results on that dimension, namely the probability of each level of the four-level indicator, and the level with the highest probability is selected as the classification result.

[0061] During the training process, a supervised learning method is adopted, and cross entropy loss is used as the loss function to measure the difference between the predicted results and the true labels. The cross entropy loss can be expressed as follows:

[0062]

[0063] Among them, y i' is the true label corresponding to the second fusion feature of the i'th i The model predicts the probability corresponding to the i'th second fused feature. The optimization process uses the Adam optimizer to adjust the learning rate of each parameter to accelerate the training process and avoid gradient explosion or vanishing problems.

[0064] The training process uses a training set and a validation set to iteratively train and evaluate the model. Backpropagation is used to update the network weights, gradually reducing the loss function. After each iteration, the validation set is used for evaluation to check the model's performance on unseen data, and early stopping is implemented to prevent overfitting. After training, cross-validation is used to assess the model's accuracy and generalization ability. Hyperparameters are adjusted, and the optimal model is selected for final testing. Metrics such as F1 score, precision, and recall are used to measure the model's performance across different dimensions.

[0065] S107. Construct a supplier qualification radar chart based on the supplier qualification grading results to represent the supplier's qualifications in multiple scoring dimensions.

[0066] Optionally, S107 may include:

[0067] Standardize the supplier qualification grading results to obtain standardized results;

[0068] Use the Sigmoid function to correct the standardized result and obtain the corrected result;

[0069] According to the revised results and the weights corresponding to each scoring dimension in the revised results, a weighted score is obtained;

[0070] Build a supplier qualification radar chart based on weighted scores.

[0071] Optionally, the corrected result is expressed as follows:

[0072]

[0073] Among them, s i ' represents the supplier's score in the i-th scoring dimension in the corrected results, Sigmoid(.) represents the Sigmoid function, k is the parameter that controls the steepness of the Sigmoid function, and e is a natural constant;

[0074] The weighted scores are expressed as follows:

[0075]

[0076] Among them, s weighted represents the weighted score of the supplier, n is the number of scoring dimensions, ω i It represents the weight of the supplier in the i-th scoring dimension in the revised result. It is worth mentioning that the weight of each dimension ω1, ω2, ..., ω3 can be manually determined according to the user's emphasis on each dimension, and the weight satisfies the condition 0≤ω i ≤1(0≤i≤n) and The weighted score can be given to the user as a comprehensive recommendation index.

[0077] The Supplier Qualification Radar chart is designed to map and calculate supplier qualification grading results, deriving weighted scores across various dimensions and visually displaying them. By processing and modifying the supplier qualification grading results, which include evaluation scores across multiple dimensions, an intuitive and actionable visual supplier qualification radar chart is generated, helping users gain a comprehensive understanding of each supplier's evaluation dimension. Figure 3 is a schematic diagram of a supplier qualification radar chart provided by an embodiment of the present invention, such as Figure 3 As shown, the present invention uses a radar chart as a visualization form to generate a supplier qualification radar chart for each supplier. Each dimension of the supplier corresponds to an axis in the chart, and the length of the axis represents the supplier's score value in that dimension. The supplier qualification radar charts of multiple suppliers can be merged to facilitate comparison of the differences between different suppliers in various dimensions.

[0078] It is understandable that, before S105, the method may further include:

[0079] Obtain user demand data;

[0080] Input user demand data into the trained BERT to obtain user demand features.

[0081] It is understandable that by analyzing the multi-dimensional needs of users (such as price sensitivity, delivery time requirements, quality standards, etc.), they can be converted into feature representations. For user demand data, its semantic features can also be extracted through the trained BERT. Combining the already integrated supplier multimodal features with the user demand features, f user ∈R 128 , f concat =[f supplier ;f user ]∈R 352+128 =R 480 .

[0082] The present invention has significant advantages over traditional technologies. First, deep learning can effectively process multi-source heterogeneous data, combine data in different forms such as text, images, transaction records, etc., and comprehensively construct supplier portraits, overcoming the limitation that traditional methods can only rely on structured data; the heterogeneous feature extraction model constructed based on deep learning can learn potential and deep-level correlations from a large amount of complex data, and no longer relies on manually set rules, thereby improving the accuracy and depth of the portrait; secondly, deep learning has strong adaptive capabilities, can process and update data in real time, automatically adapt to changes in supplier behavior, and ensure the timeliness and dynamism of the portrait, which enables the supplier portrait based on deep learning to more accurately reflect the real-time status of the supplier, and provide more intelligent and precise support for the company's supplier management and decision-making; in addition, deep learning can also avoid the complexity of manual feature selection and data preprocessing in traditional methods through an end-to-end learning process, greatly simplifying the model construction process and improving efficiency and scalability.

[0083] It should be noted that the terms "first," "second," and the like are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention.

[0084] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification.

[0085] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art can understand and implement other variations of the disclosed embodiments by viewing the drawings and the disclosed content. In the description of the present invention, the word "comprising" does not exclude other components or steps, "one" or "a" does not exclude multiple situations, and "multiple" means two or more, unless otherwise clearly and specifically defined. In addition, certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0086] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for constructing supplier intelligent portraits based on multi-source heterogeneous data fusion, characterized in that: The method comprises: Acquire multi-source heterogeneous data of suppliers; wherein the multi-source heterogeneous data of suppliers includes: graph data, time series data, text data and structured data; Inputting the supplier's multi-source heterogeneous data into the trained heterogeneous feature extraction model; Obtain the relational features corresponding to the graph data, the dynamic features corresponding to the time series data, the semantic features corresponding to the text data, and the numerical features corresponding to the structured data in the supplier's multi-source heterogeneous data; Concatenate and fuse the relational feature, the dynamic feature, the semantic feature, and the numerical feature to obtain a first fused feature; Concatenate and fuse the fusion feature and the user demand feature to obtain a second fusion feature; Obtaining a corresponding supplier qualification grading result according to the second fusion feature; A supplier qualification radar chart is constructed based on the supplier qualification grading results to characterize the supplier's qualifications in multiple scoring dimensions.

2. The method for constructing supplier intelligent portrait based on multi-source heterogeneous data fusion according to claim 1 is characterized in that: The trained heterogeneous feature extraction model includes: a trained graph convolutional network, a trained Transformer, a trained BERT and a trained convolutional neural network.

3. The method for constructing supplier intelligent portrait based on multi-source heterogeneous data fusion according to claim 2 is characterized in that: The trained graph convolutional network is used to extract relational features corresponding to graph data in the supplier's multi-source heterogeneous data; The trained Transformer is used to extract dynamic features of time series data in the supplier's multi-source heterogeneous data; the trained graph convolutional network is used to extract semantic features corresponding to text data in the supplier's multi-source heterogeneous data; The trained graph convolutional network is used to extract numerical features corresponding to structured data in the supplier's multi-source heterogeneous data.

4. The method for constructing supplier intelligent portrait based on multi-source heterogeneous data fusion according to claim 2 is characterized in that: Before the fusion feature and the user demand feature are combined and fused to obtain a second fusion feature, the method includes: Obtain user demand data; The user demand data is input into the trained BERT to obtain the user demand features.

5. The method for constructing supplier intelligent portrait based on multi-source heterogeneous data fusion according to claim 1 is characterized in that: Obtaining a corresponding supplier qualification grading result according to the second fusion feature includes: Inputting the second fused features into the trained supplier grading model; Obtain the supplier qualification grading result.

6. The method for constructing supplier intelligent portrait based on multi-source heterogeneous data fusion according to claim 1 is characterized in that: The step of constructing a supplier qualification radar chart based on the supplier qualification grading results includes: Standardizing the supplier qualification grading results to obtain standardized results; Using a Sigmoid function to correct the normalized result to obtain a corrected result; Obtaining a weighted score based on the revised result and the weights corresponding to the various scoring dimensions in the revised result; The supplier qualification radar chart is constructed according to the weighted scores.

7. The method for constructing supplier intelligent portrait based on multi-source heterogeneous data fusion according to claim 6 is characterized in that: The corrected results are shown below: Among them, s i ' represents the score of the supplier in the i-th scoring dimension in the corrected result, Sigmoid(.) represents the Sigmoid function, k is a parameter that controls the steepness of the Sigmoid function, and e is a natural constant.

8. The method for constructing supplier intelligent portrait based on multi-source heterogeneous data fusion according to claim 7 is characterized in that: The weighted scores are expressed as follows: Among them, s weighted represents the weighted score of the supplier, n is the number of scoring dimensions, ω i Indicates the weight of the supplier in the i-th scoring dimension in the revised result.

Images