CN117522132A - Vendor risk assessment system and application method - Google Patents

Abstract

The invention discloses a provider risk assessment system and an application method, comprising the following steps: s1, acquiring basic data comprising numerical type and text type from a provider; s2, preprocessing abnormal values, missing values, repetition and/or nonsensical characters in the basic data through a preprocessing module; s3, performing risk analysis on the preprocessed numerical data based on a random forest algorithm by the numerical model module so as to output a preliminary evaluation result I; s4, performing risk analysis on the preprocessed text data by using a text model module to output a preliminary evaluation result II; and S5, outputting a comprehensive evaluation result by the risk evaluation model module based on the evaluation result I and the evaluation result II. The invention provides a provider risk assessment system and an application method thereof, which can provide a comprehensive, accurate and dynamic provider risk assessment tool for a decision maker by combining numerical value and text data of a provider.

Description

Vendor risk assessment system and application method

Technical Field

The present invention relates to the field of risk assessment. More particularly, the present invention relates to a vendor risk assessment system and method of use.

Background

Supply chain management has become an important component of enterprise core competitiveness in a global economic environment. As a key element in the supply chain, the stability and reliability of the logistics provider directly affects the efficiency and stability of the overall supply chain. Therefore, assessment and management of risk of suppliers is an important content of supply chain management. Conventional vendor risk assessment methods are typically based on historical data, empirical judgment, and simple statistical analysis. Although these methods can provide references to some extent, they often lack depth, accuracy, and pertinence. In particular, in a complex supply chain environment, risk factors for suppliers may involve multiple dimensions, including quality, delivery, financial stability, legitimacy, etc., which are difficult to fully capture by a single assessment method.

In recent years, with the rapid development of data science and artificial intelligence technology, especially the wide application of machine learning and deep learning technology, a new method and thought are provided for risk assessment of suppliers. These techniques are capable of processing large amounts of data, capturing complex nonlinear relationships, and providing more accurate and deep risk assessment results. However, how to combine these advanced techniques with the actual needs of supply chain management to design a practical and efficient evaluation system remains an important topic of current research and application.

Disclosure of Invention

It is an object of the present invention to address at least the above problems and/or disadvantages and to provide at least the advantages described below.

To achieve these objects and other advantages and in accordance with the purpose of the invention, there is provided a vendor risk assessment system, comprising:

s1, acquiring basic data comprising numerical type and text type from a provider;

s2, preprocessing abnormal values, missing values, repetition and/or nonsensical characters in the basic data through a preprocessing module;

s3, performing risk analysis on the preprocessed numerical data based on a random forest algorithm by the numerical model module so as to output a preliminary evaluation result I;

s4, performing risk analysis on the preprocessed text data by using a text model module to output a preliminary evaluation result II;

and S5, outputting a comprehensive evaluation result by the risk evaluation model module based on the evaluation result I and the evaluation result II.

Preferably, the base data includes financial data, operational data, contractual text, social feedback, other related evaluations.

Preferably, the preprocessing module processes the text data in the following manner:

removing duplicate and/or nonsensical text by natural language processing techniques;

key information is extracted from text data by text mining techniques.

Preferably, the preprocessing module includes:

a data partitioning sub-module that partitions the numeric data based on data type and/or data source;

a data conversion sub-module for converting the non-numerical data into numerical data through an algorithm;

a data smoothing sub-module that processes short-term fluctuations in the data;

the data conversion sub-module performs word segmentation, part-of-speech tagging and emotion analysis on the text data through a natural language processing technology so as to convert the text data into word vectors;

and the data conversion submodule completes feature extraction on the picture data by adopting a convolutional neural network.

Preferably, the risk assessment model module includes:

a dynamic feature selection sub-module for adjusting feature weights in real time based on a feature importance feedback mechanism of the model;

the model iterative training sub-module is used for triggering iterative training automatically periodically or when a certain condition is reached;

a hardware optimization sub-module based on GPU;

the model iterative training submodule actively searches valuable data for learning based on an active learning strategy when new data are added or a risk mode is changed, so that model performance is optimized in real time;

the hardware optimization sub-module is accelerated by the GPU and distributes partial risk assessment tasks to various nodes of the supply chain based on an edge calculation strategy.

The invention at least comprises the following beneficial effects: according to the logistics provider risk assessment method, the numerical data and the text data are acquired from the provider, the risk analysis is carried out on the preprocessed numerical data in the numerical model module based on the random forest algorithm, the preliminary assessment result is output, the deep analysis is carried out on the preprocessed text data, the comprehensive provider risk assessment result is obtained by combining the analysis result of the numerical data, and a comprehensive, accurate and dynamic provider risk assessment tool can be provided for a decision maker by combining the numerical data and the text data of the provider.

Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.

Drawings

FIG. 1 is a flow chart of a logistics provider risk assessment system framework of the present invention;

FIG. 2 is a block diagram illustrating a data preprocessing module of the risk assessment system of the present invention;

FIG. 3 is a block diagram of a risk assessment model module of the logistics provider risk assessment system of the present invention.

Detailed Description

The present invention is described in further detail below with reference to the drawings to enable those skilled in the art to practice the invention by referring to the description.

A vendor risk assessment system, comprising:

and a data preprocessing module: when the numerical data is processed, abnormal values are automatically detected and processed through an advanced algorithm, so that the data distribution normality is ensured. For missing values, selecting a proper filling strategy according to the characteristics of the data; in text data processing, natural language processing techniques are applied to remove duplicate and nonsensical text, such as common filler words or idioms. The key information is extracted by a text mining technology, and high-quality text data is provided for subsequent analysis.

Numerical model module: the numerical data of the suppliers are trained using a random forest model. The integrated learning characteristic of the random forest model is considered, so that the robustness and the accuracy of the model can be improved. The training process of the model is visualized, so that a decision maker can know the training state and important parameters of the model.

A text model module: training the text data of the provider by adopting RNN and LSTM models in deep learning. These models can capture long-term dependencies in text, providing accurate interpretation of complex text content. And providing a parameter visualization of model training, and helping a user to know the training process and effect of the model.

The result display module: and comprehensively evaluating risks of the suppliers by combining the output of the numerical model and the text model, and giving risk grades. Visual visualization tools such as bar charts, pie charts, and thermodynamic diagrams are provided to enable a decision maker to quickly understand the assessment results.

The real-time assessment interface is designed, so that enterprises can be allowed to input new provider data at any time, and the system can give risk assessment results in real time.

A method of application of a vendor risk assessment system, comprising:

s1: collecting basic information of suppliers, including financial data, operation data, contract text, social feedback and other relevant evaluations;

s2: the basic information of the provider is processed in detail through a data preprocessing module, namely, the obtained numerical data is processed by using a numerical model module to obtain abnormal values and missing values, and repeated characters and/or nonsensical characters in the text data are extracted and deleted by using a text model module;

s3: and (3) carrying out risk level assessment on the processed data by using a risk assessment model module, wherein the risk assessment model module adopts the following risk assessment formula to assess the overall risk:

total risk= (provider history score x 0.4) + (provider financial status score x 0.3) + (text data risk score x 0.2) + (other risk factor score x 0.1);

in actual operation, after quantification according to theoretical research, a risk value greater than 0.5 is defined as high risk, a risk value between 0.3 and 0.5 is defined as medium risk, and a risk value below 0.3 is defined as low risk;

s4: the comprehensive risk assessment results of the suppliers are visually displayed by utilizing the result presentation module, so that a decision maker can intuitively know the risk level of the suppliers and provide a corresponding risk management strategy;

in the step S2, the data preprocessing module mainly uses a numerical model module to perform risk analysis on the preprocessed numerical data based on a random forest algorithm, and outputs a preliminary vendor risk assessment result; meanwhile, if the numerical data of the provider shows a high risk trend, but the text data contains some positive evaluation or information, the two data are weighted and a comprehensive risk evaluation result is output, and further, when the numerical model module weights the numerical data and the text data, the system adopts a hybrid scoring mechanism, and the mechanism considers industry standards, historical data and related data of other providers to output a final risk evaluation result, and in a specific operation, the hybrid scoring mechanism mainly comprises the following mechanisms:

1. integrating a rule-based judging module (or adopting industry standard), when the calculated risk score deviates from the judging suggestion of the judging module (the deviation range can be determined according to the actual working requirement, namely if the deviation is not processed in a certain range and exceeds the range to carry out warning prompt), triggering a warning by the system to prompt a decision maker to review;

2. a real-time feedback mechanism that allows a decision maker or other user to provide feedback, such as indicating the variability of a certain evaluation result, which the system considers to fine tune the weights and parameters of the model;

3. the multi-model fusion, except for random forest algorithm, the numerical model module can introduce other machine learning models, such as a support vector machine, a gradient elevator and the like, and carries out weighted fusion on the prediction results of the machine learning models, so as to improve the accuracy of risk assessment;

4. historical data backtracking analysis, wherein the system dynamically adjusts the weight of the historical data in the mixed score according to the historical risk assessment score of the provider, and if the provider always shows stability in the past, the weight of the historical data in the assessment is gradually increased;

in addition, it should be noted that: the text evaluation results are based on data labels of the model, and the data labels have corresponding text risk classification;

in S2, the data preprocessing module mainly uses a text model module to perform deep analysis on the preprocessed text data, combines the analysis result of the numerical data, outputs a comprehensive risk assessment result of the provider, and also analyzes the similarity between the text model module and the text data of other providers according to the historical text data of the provider, and further judges the reputation and stability of the provider according to the similarity, and further, the text model module can identify potential risks and opportunities through the deep analysis on the historical text data of the provider, for example: if a provider has been criticized for some reason in the past, but recently begun to gain positive assessment, the system will incorporate this positive change into the assessment, thereby providing a more comprehensive and accurate risk assessment result;

further, as shown in fig. 2, the data preprocessing module in step S2 specifically includes:

a data segmentation sub-module: orderly partitioning data according to data type or source, wherein orderly partitioning refers to data partitioning according to a certain logic sequence, such as time sequence (such as year, quarter, month), data size, alphabetical sequence, etc.;

a data conversion sub-module: converting non-numeric data, such as text or pictures, into numeric data via a specific (i.e., word embedding) algorithm;

data smoothing submodule: short-term fluctuation in the data is processed by adopting methods such as moving average and the like, so that the accuracy of the subsequent model prediction is improved;

wherein, the data conversion submodule refines into:

preprocessing the text data by using a natural language processing technology, such as word segmentation, part-of-speech tagging, emotion analysis and the like, and finally converting the text data into word vectors;

and the image data adopts a convolutional neural network to extract the characteristics.

In practical application, the data preprocessing module firstly identifies the missing value in the data and selects a proper method for filling; secondly, identifying and processing abnormal values by utilizing an algorithm such as an isolated forest; finally, selecting a proper normalization method according to the data distribution characteristics;

further, as shown in fig. 3, the risk assessment model module includes:

dynamic feature selection submodule: besides the conventional methods of correlation coefficient, mutual information and the like, a model-based feature importance feedback mechanism is adopted to adjust the weight of the features in real time, so that the system can still accurately capture key features of risks over time, and the feature importance feedback mechanism comprises:

1. depth feature extraction network: first, a high-dimensional representation or embedding is extracted for each feature through a deep neural network structure, which captures the potential relationship between each feature and other features.

2. Feature importance evaluator: next, a sub-network is used to evaluate the importance of each feature.

3. Reinforcement learning environment: a reinforcement learning environment is built for the model, wherein feature importance assessors are agents whose goal is to maximize some long-term return (e.g., the model's predictive accuracy). Each time a new data point or batch arrives, the agent makes an action, i.e., adjusts the weight of the feature.

4. Rewarding mechanism: when new data points are classified or predicted, if the prediction of the model is accurate, the agent will get positive feedback; if the prediction is wrong, the agent is penalized. This feedback will drive the feature importance evaluator to continually adjust feature weights, ensuring that the most critical features are always correctly weighted;

model iteration training submodule: the system not only trains based on the features of initial screening, but also periodically or automatically triggers iterative training when a certain condition (such as a certain specific error rate) is reached, in addition, when new data is added or a risk mode is changed, the system can actively search the most valuable data for learning and optimize model performance in combination with an active learning strategy;

hardware optimization sub-module: besides GPU acceleration, an edge computing strategy is considered, part of risk assessment tasks are distributed to all nodes of a supply chain, the burden of a central server is reduced, and meanwhile, the real-time performance of assessment is improved.

Further, the result presentation module not only can display the risk level of the provider, but also can provide explanation of risk sources, give possible risk countermeasures and display the historical trend of the risk assessment of the provider, so that a decision maker can more easily identify the risk change of the provider;

in practical applications, the vendor risk assessment system also has the following functions:

the risk assessment is carried out on the provider data input in real time, corresponding advice or warning is provided for a decision maker according to the risk level of the provider data, meanwhile, when the advice or warning is provided for the decision maker, the system can detail specific factors which cause high risk assessment, such as financial conditions, historical delivery records and the like of the provider, so that the decision maker can know the problem more clearly, in practical application, the numerical data is judged according to a threshold value to be endowed with a corresponding risk data label, the text data has a risk label, and the risk assessment can be realized based on the overall risk assessment formula and the text model part-of-speech label;

new vendor data can be automatically updated and learned to maintain accuracy and timeliness of the assessment model, while new vendor data is automatically acquired and learned by continuously monitoring external data sources or industry news, as well as new events or changes that may affect risk assessment;

when new risk factors appear in the provider data, the system can automatically adjust the evaluation model to ensure the accuracy of the evaluation result, meanwhile, the built-in self-adaptive algorithm of the system can periodically check the evaluation accuracy of the model, and if systematic deviation or error is found, the system can automatically adjust and optimize the model;

the system also provides a real-time interactive assessment function that allows a decision maker or other user to enter newly acquired vendor-related text. When new vendor text data is entered, the system immediately proceeds to:

1. the data is read in real time and preprocessed, such as text cleansing, key information extraction and vectorization.

2. And quickly calling a risk assessment model, combining the existing numerical data and historical text data of other suppliers, carrying out risk assessment on the newly input text data, and providing a new supplier risk assessment result for a decision maker in a few seconds.

A supply chain risk assessment system, comprising: a data preprocessing module, a risk assessment model module, a result presentation (or expansion) module.

The system comprises a data acquisition system, a risk assessment system, a risk management suggestion system, a data preprocessing module matched with the data acquisition system, a risk assessment model module matched with the risk assessment system and a result presentation module matched with the risk management suggestion system, and meanwhile, the risk management suggestion system performs assessment according to a constructed corpus.

A supply chain risk assessment maintenance system further incorporates the following databases:

vendor base information database: all historical and current data of the provider is contained;

risk assessment results database: storing risk assessment results of all suppliers, and supporting retrieval according to conditions such as date, risk grade and the like;

risk management advice database: professional risk management strategies and suggestions are provided for suppliers with different risk levels, and in actual risk assessment, a risk assessment model module can be called when the strategy is given in assessment according to actual conditions.

The invention provides a risk assessment method for a logistics provider. First, basic data including numeric type and text type data is acquired from a provider. The data is then subjected to outliers, missing values and repeated nonsensical words by a preprocessing module. Based on a random forest algorithm, the preprocessed numerical data is subjected to risk analysis in a numerical model module, and a preliminary evaluation result is output. And simultaneously, the text model module performs deep analysis on the text data and combines the analysis result of the numerical data to obtain a comprehensive vendor risk assessment result. The system may also preferably provide visual presentation of vendor risk levels, providing real-time advice or warnings to the decision maker. In addition, the system can automatically update and learn new vendor data, ensure accuracy and timeliness of the model, and automatically adjust the assessment model when new risk factors occur. By combining numeric and textual data of the provider, the present invention provides a comprehensive, accurate and dynamic provider risk assessment tool for the decision maker.

The above is merely illustrative of a preferred embodiment, but is not limited thereto. In practicing the present invention, appropriate substitutions and/or modifications may be made according to the needs of the user.

The number of equipment and the scale of processing described herein are intended to simplify the description of the present invention. Applications, modifications and variations of the present invention will be readily apparent to those skilled in the art.

Although embodiments of the invention have been disclosed above, they are not limited to the use listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Additional modifications will readily occur to those skilled in the art. Therefore, the invention is not to be limited to the specific details and illustrations shown and described herein, without departing from the general concepts defined in the claims and their equivalents.

Claims (6)

1. A vendor risk assessment system, comprising: the system comprises a data preprocessing module, a numerical model module, a text model module, a risk assessment model module and a result display module.

2. A method of using the vendor risk assessment system of claim 1, comprising:

s1, acquiring basic data comprising numerical type and text type from a provider;

s2, preprocessing abnormal values, missing values, repetition and/or nonsensical characters in the basic data through a preprocessing module;

s3, performing risk analysis on the preprocessed numerical data based on a random forest algorithm by the numerical model module so as to output a preliminary evaluation result I;

s4, performing risk analysis on the preprocessed text data by using a text model module to output a preliminary evaluation result II;

and S5, outputting a comprehensive evaluation result by the risk evaluation model module based on the evaluation result I and the evaluation result II.

3. The method of claim 2, wherein the base data includes financial data, operational data, contractual text, social feedback, other related evaluations.

4. The method for applying the provider risk assessment system according to claim 2, wherein the preprocessing module processes the text data in a manner that:

removing duplicate and/or nonsensical text by natural language processing techniques;

key information is extracted from text data by text mining techniques.

5. The method of claim 2, wherein the preprocessing module comprises:

a data partitioning sub-module that partitions the numeric data based on data type and/or data source;

a data conversion sub-module for converting the non-numerical data into numerical data through an algorithm;

a data smoothing sub-module that processes short-term fluctuations in the data;

the data conversion sub-module performs word segmentation, part-of-speech tagging and emotion analysis on the text data through a natural language processing technology so as to convert the text data into word vectors;

and the data conversion submodule completes feature extraction on the picture data by adopting a convolutional neural network.

6. The method of claim 2, wherein the risk assessment model module comprises:

a dynamic feature selection sub-module for adjusting feature weights in real time based on a feature importance feedback mechanism of the model;

the model iterative training sub-module is used for triggering iterative training automatically periodically or when a certain condition is reached;

a hardware optimization sub-module based on GPU;

the model iterative training submodule actively searches valuable data for learning based on an active learning strategy when new data are added or a risk mode is changed, so that model performance is optimized in real time;

the hardware optimization sub-module is accelerated by the GPU and distributes partial risk assessment tasks to various nodes of the supply chain based on an edge calculation strategy.