Enterprise data processing method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides an enterprise data processing method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring multi-modal data of a target enterprise; performing feature extraction on the multi-modal data to obtain different multi-modal features; performing feature fusion on different multi-modal features to obtain fused features; identifying feature attributes of the multi-modal features, wherein the feature attributes comprise data integrity, timestamps and risk types; determining a risk assessment type according to the risk assessment task demand of the target enterprise; determining a target processing strategy according to the feature attribute and the risk assessment type; and performing risk assessment processing on the fusion feature according to the target processing strategy through a risk assessment model, and generating a risk assessment result of the corresponding target enterprise. According to the embodiment of the invention, the comprehensiveness and accuracy of enterprise risk assessment can be improved, the resource allocation weight of the risk assessment model is dynamically adjusted according to the target processing strategy, and efficient and accurate enterprise risk assessment can be realized in a financial scene.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an enterprise data processing method and apparatus, electronic device and storage medium. Background Technology

[0002] Currently, risk assessments of enterprises rely heavily on static financial data (such as financial statements, historical transaction records, and historical credit records), lacking integration of non-static financial data (such as supply chain relationships, public opinion dynamics, and industry policy changes). This results in a single assessment dimension, failing to dynamically reflect the real-time operating status of enterprises. Operationally, the entire risk assessment process relies heavily on manual review, leading to inconsistent efficiency and standardization. For example, pre-loan verification of transaction backgrounds requires methods such as telephone, fax, or offline follow-up, resulting in lengthy processes that are difficult to meet the needs of small-amount, high-frequency factoring. Furthermore, risk assessments often employ scoring cards, lacking corresponding refined models. Therefore, current enterprise data processing methods for enterprise risk assessment lack comprehensiveness and accuracy. Consequently, improving the comprehensiveness and accuracy of enterprise risk assessment in financial scenarios has become a pressing technical challenge. Summary of the Invention

[0003] The main objective of this application is to propose an enterprise data processing method, apparatus, electronic device, and storage medium that can improve the comprehensiveness and accuracy of enterprise risk assessment. Furthermore, when fusing and analyzing multimodal data, it can dynamically adjust the resource allocation weights of the risk assessment model according to the target processing strategy, thereby rationally allocating computing resources and reducing computing costs. In financial scenarios, it can achieve efficient and accurate enterprise risk assessment.

[0004] To achieve the above objectives, a first aspect of this application proposes an enterprise data processing method, the method comprising:

[0005] Acquire multimodal data of the target enterprise, including structured data, unstructured data, and semi-structured data; Feature extraction is performed on the multimodal data to obtain different multimodal features; The different multimodal features are fused to obtain fused features; Identify the feature attributes of the multimodal features, including data integrity, timestamp, and risk type; The risk assessment type is determined based on the risk assessment task requirements of the target enterprise. Determine the target treatment strategy based on the aforementioned characteristic attributes and the aforementioned risk assessment type; The risk assessment model performs risk assessment processing on the fusion features according to the target processing strategy, generating risk assessment results corresponding to the target enterprise. The risk assessment model includes multiple target sub-models with different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy.

[0006] In some embodiments, acquiring the multimodal data of the target enterprise includes: The structured data of the target enterprise can be obtained in real time through an application programming interface; The unstructured data of the target company is crawled using web crawlers. The semi-structured data of the target enterprise is received through an IoT device.

[0007] In some embodiments, the target multimodal data includes structured data, unstructured data, and semi-structured data, and the feature extraction of the target multimodal data to obtain different multimodal features includes: Feature extraction is performed on the structured data to obtain structured features; Feature extraction is performed on the unstructured data to obtain unstructured features; Feature extraction is performed on the semi-structured data to obtain semi-structured features; Different multimodal features are obtained based on the structured features, the unstructured features, and the semi-structured features.

[0008] In some embodiments, the feature fusion of the different multimodal features to obtain fused features includes: The feature weights corresponding to the structured features, unstructured features, and semi-structured features are determined based on a multi-head attention mechanism. Based on the feature weights, the structured features, unstructured features, and semi-structured features are concatenated and weighted to obtain the fused features.

[0009] In some embodiments, the step of performing risk assessment processing on the fusion features according to the target processing strategy using a risk assessment model to generate a risk assessment result corresponding to the target enterprise includes: The resource allocation weights of each target sub-model in the risk assessment model are determined according to the target processing strategy, wherein the target sub-model is any one of a text risk scoring model, a graph risk scoring model, and a time-series risk scoring model; If the resource allocation weight of the target sub-model in the risk assessment model is determined to be zero, the target sub-model is disabled based on a gating mechanism. If the resource allocation weight of the target sub-model in the risk assessment model is determined to be greater than zero and less than a preset threshold, the target sub-model is lightweighted. If the resource allocation weight of the target sub-model in the risk assessment model is determined to be greater than the preset threshold, the target operating state of the risk assessment model is determined according to the resource allocation weight of each target sub-model. The risk assessment model, which is in the target operating state, performs risk assessment processing on the fusion features to generate risk assessment results corresponding to the target enterprise.

[0010] In some embodiments, the method further includes: Generate a risk assessment report based on the risk assessment results; The risk assessment report is displayed on the terminal's visual interface.

[0011] In some embodiments, the method further includes: Obtain user interaction information; The user interaction information can be used to query or adjust the assessment content of the risk assessment report.

[0012] To achieve the above objectives, a second aspect of this application provides an enterprise data processing apparatus, the apparatus comprising: The acquisition module is used to acquire multimodal data of the target enterprise, including structured data, unstructured data, and semi-structured data. The extraction module is used to extract features from the multimodal data to obtain different multimodal features; The fusion module is used to fuse the different multimodal features to obtain fused features; The identification module is used to identify the feature attributes of the multimodal features, including data integrity, timestamp, and risk type; The requirements module is used to determine the risk assessment type based on the risk assessment task requirements of the target enterprise; The strategy module is used to determine the target processing strategy based on the feature attributes and the risk assessment type; The assessment module is used to perform risk assessment processing on the fusion features according to the target processing strategy through a risk assessment model, and generate risk assessment results corresponding to the target enterprise. The risk assessment model includes multiple target sub-models with different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy.

[0013] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0014] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0015] This application proposes a method, apparatus, electronic device, and storage medium for enterprise data processing. The method acquires multimodal data from a target enterprise, including structured, unstructured, and semi-structured data. It extracts features from the multimodal data to obtain different multimodal features; fuses these features to obtain fused features; identifies the characteristic attributes of the multimodal features, including data integrity, timestamps, and risk types; determines the risk assessment type based on the risk assessment task requirements of the target enterprise; determines a target processing strategy based on the characteristic attributes and risk assessment type; and performs risk assessment processing on the fused features using a risk assessment model according to the target processing strategy, generating a corresponding risk assessment result for the target enterprise. The risk assessment model includes multiple target sub-models of different risk assessment types, and the resource allocation weights of each target sub-model in the risk assessment model are determined by the target processing strategy. By acquiring multidimensional multimodal data from the target enterprise, extracting features from the multimodal data to obtain different multimodal features, and fusing these different multimodal features to obtain fused features, the method achieves the desired results. This application not only combines the feature attributes of multimodal features with the risk assessment task requirements of the target enterprise to conduct multi-dimensional risk assessment of the enterprise, making the assessment dimensions more comprehensive, but also determines targeted target processing strategies based on the feature attributes of multimodal features and risk assessment types. A risk assessment model containing multiple target sub-models processes the fused features according to the target processing strategies to generate risk assessment results for the corresponding target enterprise. Since the resource allocation weights of each target sub-model in the risk assessment model are determined by the target processing strategies, the resource weight allocation is more intelligent and adaptable, thereby improving the accuracy of enterprise risk assessment. Furthermore, through this dynamic activation mechanism of the target processing strategies, the activation or adjustment weights of target sub-models of corresponding risk assessment types can be determined. This allows for flexible adjustment of the risk assessment model structure in cases of data loss, changes in task requirements, or resource constraints, efficiently generating accurate risk assessment results for the target enterprise. Based on this, this application embodiment can improve the comprehensiveness and accuracy of enterprise risk assessment. Moreover, when fusing and analyzing multimodal data, it can dynamically adjust the resource allocation weights of the risk assessment model according to the target processing strategies, thereby rationally allocating computing resources and reducing computing costs. In financial scenarios, it can achieve efficient and accurate enterprise risk assessment. Attached Figure Description

[0016] Figure 1 This is a flowchart of the enterprise data processing method provided in the embodiments of this application; Figure 2 yes Figure 1 The flowchart of step S101 in the text; Figure 3 yes Figure 1 The flowchart of step S103 in the process; Figure 4 yes Figure 1 The flowchart of step S104 in the process; Figure 5 yes Figure 1 The flowchart of step S107 in the process; Figure 6 This is a flowchart of steps S601 to S602; Figure 7 This is a flowchart of steps S701 to S702; Figure 8 This is a schematic diagram of the structure of the enterprise data processing device provided in the embodiments of this application; Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] First, let's analyze some of the terms used in this application: Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0021] The Transformer model is a deep learning architecture based on self-attention. It revolutionized the field of Natural Language Processing (NLP) and became the core foundation of modern large language models such as DeepSeek, BERT, and GPT.

[0022] Graph Neural Networks (GNNs) are a general term for algorithms that use neural networks to learn graph-structured data, extract and discover features and patterns within it, and meet the needs of graph learning tasks such as clustering, classification, prediction, segmentation, and generation. By establishing specific strategies for nodes and edges in a graph, GNNs transform graph-structured data into a standardized representation, which is then fed into various neural networks for training. They have achieved excellent results in tasks such as node classification, edge information propagation, and graph clustering.

[0023] Long Short-Term Memory (LSTM) networks are a type of recurrent neural network specifically designed to address the long-term dependency problem inherent in general RNNs (Recurrent Neural Networks). All RNNs have a chain-like structure of repeating neural network modules. Due to its unique design, LSTM is well-suited for processing and predicting important events with very long intervals and delays in time series.

[0024] Deep Learning (DL) is a new research direction in the field of Machine Learning (ML). It was introduced into machine learning to bring it closer to its original goal—Artificial Intelligence (AI). Deep learning learns the inherent patterns and hierarchical representations of sample data. The information gained during this learning process greatly aids in interpreting data such as text, images, and sound. Its ultimate goal is to enable machines to possess analytical and learning capabilities like humans, capable of recognizing data such as text, images, and sound. Deep learning is a complex machine learning algorithm that has achieved results in speech and image recognition far exceeding previous related technologies. Deep learning has yielded significant achievements in search technology, data mining, machine learning, machine translation, natural language processing, multimedia learning, speech recognition, recommendation and personalization technologies, and other related fields. Deep learning enables machines to mimic human activities such as sight, hearing, and thinking, solving many complex pattern recognition problems and leading to significant advancements in artificial intelligence-related technologies.

[0025] Currently, risk assessments of enterprises rely heavily on static financial data, lacking integration of non-static financial data (such as supply chain relationships, public opinion dynamics, and industry policy changes), resulting in a single assessment dimension. Operationally, the entire risk assessment process involves excessive manual review, leading to inconsistent efficiency and standardization. For example, pre-loan verification of transaction backgrounds often requires phone calls, faxes, or offline follow-ups, a time-consuming process. Therefore, improving the comprehensiveness and accuracy of enterprise risk assessment in financial scenarios has become a pressing technical challenge.

[0026] Based on this, embodiments of this application provide an enterprise data processing method and apparatus, electronic device, and storage medium, which acquires multimodal data of a target enterprise; extracts features from the multimodal data to obtain different multimodal features; fuses the different multimodal features to obtain fused features; identifies the feature attributes of the multimodal features, including data integrity, timestamps, and risk types; determines the risk assessment type according to the risk assessment task requirements of the target enterprise; determines a target processing strategy based on the feature attributes and risk assessment type; and performs risk assessment processing on the fused features according to the target processing strategy using a risk assessment model to generate a risk assessment result corresponding to the target enterprise. The risk assessment model includes multiple target sub-models of different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy. By acquiring multi-dimensional multimodal data of the target enterprise, extracting features from the multimodal data to obtain different multimodal features, and fusing the different multimodal features to obtain fused features, this method achieves the desired results. This application not only combines the feature attributes of multimodal features with the risk assessment task requirements of the target enterprise to conduct multi-dimensional risk assessment of the enterprise, making the assessment dimensions more comprehensive, but also determines targeted target processing strategies based on the feature attributes of multimodal features and risk assessment types. A risk assessment model containing multiple target sub-models processes the fused features according to the target processing strategies to generate risk assessment results for the corresponding target enterprise. Since the resource allocation weights of each target sub-model in the risk assessment model are determined by the target processing strategies, the resource weight allocation is more intelligent and adaptable, thereby improving the accuracy of enterprise risk assessment. Furthermore, through this dynamic activation mechanism of the target processing strategies, the activation or adjustment weights of target sub-models of corresponding risk assessment types can be determined. This allows for flexible adjustment of the risk assessment model structure in cases of data loss, changes in task requirements, or resource constraints, efficiently generating accurate risk assessment results for the target enterprise. Based on this, this application embodiment can improve the comprehensiveness and accuracy of enterprise risk assessment. Moreover, when fusing and analyzing multimodal data, it can dynamically adjust the resource allocation weights of the risk assessment model according to the target processing strategies, thereby rationally allocating computing resources and reducing computing costs. In financial scenarios, it can achieve efficient and accurate enterprise risk assessment.

[0027] The enterprise data processing method, apparatus, electronic device, and storage medium provided in this application are specifically described through the following embodiments. First, the enterprise data processing method in this application is described.

[0028] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0029] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0030] The enterprise data processing method provided in this application relates to the field of artificial intelligence technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the enterprise data processing method, but is not limited to the above forms.

[0031] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0032] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0033] Figure 1 This is an optional flowchart of the enterprise data processing method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S107.

[0034] Step S101: Obtain multimodal data of the target enterprise, including structured data, unstructured data, and semi-structured data; Step S102: Extract features from the target multimodal data to obtain different multimodal features; Step S103: Perform feature fusion on different multimodal features to obtain fused features; Step S104: Identify the feature attributes of the multimodal features, including data integrity, timestamp, and risk type; Step S105: Determine the risk assessment type based on the risk assessment task requirements of the target enterprise; Step S106: Determine the target treatment strategy based on the characteristic attributes and risk assessment type; Step S107: The risk assessment model performs risk assessment processing on the fusion features according to the target processing strategy, and generates the risk assessment results of the corresponding target enterprise. The risk assessment model includes multiple target sub-models with different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy.

[0035] In step S101 of some embodiments, multimodal data of the target enterprise is acquired. According to data attribute types, multimodal data can include text data, image data, graph data, and time-series data. Text data includes, but is not limited to, financial data, contract texts, news reports, and social media comments; image data includes, but is not limited to, photos of enterprise facilities, equipment images, and scanned copies of invoices; graph data includes, but is not limited to, supply chain graph data, which is constructed based on supply chain relationship data; time-series data, also known as time series data, includes a series of data obtained in chronological order at different time periods or points in time, such as monthly sales trends of customers. According to data structure types, multimodal data can include structured data, unstructured data, and semi-structured data. Structured data includes, but is not limited to, financial data and transaction records; unstructured data includes, but is not limited to, contract texts, news reports (such as changes in industry policies), social media comments (such as public opinion dynamics), and image data; semi-structured data includes, but is not limited to, supply chain graph data and logistics information in JSON / XML format (such as logistics transportation status and cargo details).

[0036] In step S102 of some embodiments, feature extraction is performed on the multimodal data to obtain different multimodal features. The multimodal data includes structured data, unstructured data, and semi-structured data.

[0037] Feature extraction is performed on structured data to obtain structured features; feature extraction is performed on unstructured data to obtain unstructured features; and feature extraction is performed on semi-structured data to obtain semi-structured features. Multimodal features are then derived based on these structured, unstructured, and semi-structured features.

[0038] In step S103 of some embodiments, feature fusion is performed on the multimodal features to obtain fused features. Specifically, fusion can be performed through weighted summation, attention mechanisms, or gating mechanisms. For example, a multi-head attention mechanism can be used to dynamically weight different modal features to obtain fused features. Subsequently, the fused features can be used as a comprehensive input vector to input into the risk assessment model to improve the predictive ability of the risk assessment model.

[0039] In step S104 of some embodiments, since the multimodal features themselves have feature attributes, the feature attributes include data integrity, timestamps and risk types, such as the data missing rate of features of different modalities such as structured features, unstructured features and semi-structured features, the reliability and real-time nature of timestamps, and whether the risk type is credit risk, market risk or operational risk, etc.

[0040] In step S105 of some embodiments, it is also necessary to determine the risk assessment type of the target enterprise according to the risk assessment task requirements. The risk assessment task requirements include different types of risk assessment tasks such as credit risk assessment, market risk assessment, and operational risk assessment of the target enterprise.

[0041] In step S106 of some embodiments, a target processing strategy for risk assessment of the target enterprise is determined based on the feature attributes of the multimodal features and the risk assessment type. This allows for dynamic activation and scheduling of corresponding resources. For example, when the data missing rate of financial data features is high, the resource allocation weight for analyzing and processing structured features can be reduced; similarly, when the supply chain map features were updated several months ago, the resource allocation weight for analyzing and processing semi-structured features can be reduced; and when keywords such as "policy tightening" appear in news reports or social media comments, the resource allocation weight for analyzing and processing unstructured features can be increased. Based on this dynamic activation mechanism, targeted target processing strategies can be obtained for the target enterprise according to the multimodal features and the risk assessment task requirements of the corresponding target enterprise. This allows for dynamic adjustment of the activation status or weight allocation of target sub-models of different risk assessment types in the risk assessment model, thereby optimizing the use of computing resources and the evaluation effect of the model.

[0042] In step S107 of some embodiments, the risk assessment model performs risk assessment processing on the fused features according to the target processing strategy, generating a risk assessment result for the corresponding target enterprise. The risk assessment model includes multiple target sub-models of different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy. By using a risk assessment model containing multiple target sub-models to perform risk assessment processing on the fused features according to the target processing strategy, a risk assessment result for the corresponding target enterprise is generated. Since the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy, the resource weight allocation is more intelligent and adaptive, thereby improving the accuracy of enterprise risk assessment. It should be noted that the risk assessment results can be presented through a visual interface on a web or mobile device. The risk assessment report can include risk scores, risk levels, risk factor analysis, risk node highlighting, risk distribution heatmaps, etc., and supports user interaction. Users can not only request queries for the risk assessment report but also perform interactive operations, such as customizing risk level thresholds and triggering supply chain map updates by adding real-time transaction data.

[0043] Steps S101 to S107 of this embodiment involve acquiring multi-dimensional, multimodal data of the target enterprise, extracting features from the multimodal data to obtain different multimodal features, and fusing these different multimodal features to obtain fused features. This embodiment not only combines the feature attributes of multimodal features with the risk assessment task requirements of the target enterprise to conduct multi-dimensional risk assessment, resulting in a more comprehensive assessment, but also determines targeted target processing strategies based on the feature attributes of the multimodal features and the risk assessment type. Furthermore, a risk assessment model containing multiple target sub-models processes the fused features according to the target processing strategies to generate risk assessment results for the corresponding target enterprise. Since the resource allocation weights of each target sub-model in the risk assessment model are determined by the target processing strategies, resource weight allocation becomes more intelligent and adaptable, thereby improving the accuracy of enterprise risk assessment. Moreover, through this dynamic activation mechanism of the target processing strategies, the activation or adjustment of the weights of target sub-models for the corresponding risk assessment type can be determined. This allows for flexible adjustment of the risk assessment model structure in cases of data loss, changes in task requirements, or resource constraints, efficiently generating accurate risk assessment results for the target enterprise. Based on this, the embodiments of this application can improve the comprehensiveness and accuracy of enterprise risk assessment. When fusing and analyzing multimodal data, the resource allocation weight of the risk assessment model can be dynamically adjusted according to the target processing strategy, thereby reasonably allocating computing resources and reducing computing costs. In financial scenarios, it can achieve efficient and accurate enterprise risk assessment.

[0044] In some embodiments, before feature extraction from the multimodal data, the multimodal data can be preprocessed to obtain preprocessed multimodal data. The acquired multimodal data may contain missing fields, outliers, duplicates, or lack of standardization. Therefore, preprocessing is necessary, including but not limited to data cleaning, data standardization, and data quality assessment, to facilitate subsequent feature extraction.

[0045] Understandably, the original multimodal data can be cleaned to obtain the first multimodal data. For example, for handling missing values, time series interpolation (ARIMA model) or the KNN algorithm based on similar customers can be used to fill missing fields in financial data (such as net profit margin); for handling outliers, Z-scores can be calculated for financial indicators (such as current ratio), and when |Z|>3, it can be marked as an anomaly; semantic validation, such as unit consistency, can be performed on text data (such as contract amount); for deduplication, duplicate records can be deleted based on unique identifiers such as customer ID and contract number, and the SimHash algorithm can be used to detect similarity and remove duplicates from text data.

[0046] Understandably, raw multimodal data can be standardized to obtain target multimodal data. For example, for structured data, time formats (ISO 8601), currency units (CNY), and field naming can be standardized, such as standardizing "total accounts receivable" to "AR_Total"; for unstructured data, text can be segmented (jieba / spaCy), stop words can be removed, and stemming can be performed, and field parsing (such as amount and date) can be performed on invoice text extracted by OCR; for semi-structured data, JSON / XML data (such as logistics information) can be parsed, key fields can be extracted, and the data can be flattened into a two-dimensional table structure.

[0047] It should be noted that data quality assessment can also be performed on the target multimodal data. Data quality assessment can include integrity checks, consistency verification, and accuracy checks. For example, for integrity checks, the missing rate of key fields can be calculated (e.g., the missing rate of customer industry classification should be <5%); for consistency verification, the difference between the total accounts receivable in ERP and bank statements can be compared (e.g., a difference >5% triggers manual review); for accuracy verification, the data logic can be verified through the reconciliation of financial data (e.g., the correlation between the profit and loss statement and the cash flow statement).

[0048] Please see Figure 2 In some embodiments, step S101 may include, but is not limited to, steps S201 to S203: Step S201: Obtain the target company's structured data in real time through the application programming interface; Step S202: Use web crawlers to capture unstructured data from the target company; Step S203: Receive semi-structured data from the target enterprise via an IoT device.

[0049] It is understandable that the data sources for acquiring multimodal data can include, but are not limited to, Enterprise Resource Planning (ERP) systems, Customer Relationship Management (CRM) systems, factoring systems, third-party credit reporting platforms, logistics platforms, social media, and IoT devices. For example, structured data can be obtained in real time from the above data sources through API interfaces; unstructured data can be crawled from the above data sources using web crawlers; and semi-structured data from IoT devices can be received via MQTT / Kafka protocols.

[0050] Please see Figure 3 In some embodiments, step S103 may include, but is not limited to, steps S301 to S304: Step S301: Extract features from the structured data to obtain structured features; Step S302: Extract features from unstructured data to obtain unstructured features; Step S303: Extract features from the semi-structured data to obtain semi-structured features; Step S304: Based on structured features, unstructured features, and semi-structured features, different multimodal features are obtained.

[0051] In step S301, feature extraction is performed on the structured data to obtain structured features. Taking financial data as an example, calculations can be performed on the financial data (such as current ratio = current assets / current liabilities, net profit margin = net profit / operating revenue) to extract transaction behavior features (such as historical financing frequency, average financing cycle, and overdue frequency) to obtain a structured feature vector, such as [current ratio, net profit margin, overdue frequency]. Taking time-series data as an example, time-series data such as monthly sales trends of customers can be processed using a Long Short-Term Memory (LSTM) network model to extract time-series features and obtain a structured feature vector for predicting future risks.

[0052] In step S302, feature extraction is performed on the unstructured data to obtain unstructured features. Taking news reports as an example, the DeepSeek model (supporting 128k tokens) can be used to perform semantic analysis on long texts of news reports, extracting information such as business conditions (e.g., "market share decline"), market dynamics (e.g., "raw material price increase"), and risk signals (e.g., "executive changes"). Sentiment analysis (e.g., using the VADER model) can be performed on the long texts to generate sentiment scores (-1 to 1), outputting unstructured feature vectors such as [business condition score, market dynamics score, sentiment score]. In other words, for long text data, the DeepSeek model can be used to analyze the long text data, extracting information such as business conditions, market dynamics, and potential risks, and can also be used to assist in risk assessment based on relevant personal preferences and sentiments in the long text data. Taking photos of unstructured enterprise facilities as an example, a pre-trained CNN model (e.g., ResNet) can be used to detect key facilities (e.g., production lines, warehousing equipment) in the images. The number and status of equipment (e.g., "equipment aging") are labeled using object detection algorithms (such as YOLO), and unstructured feature vectors are output, such as [number of equipment, facility status score].

[0053] In step S303, feature extraction is performed on the semi-structured data to obtain semi-structured features. Taking a supply chain graph as an example, the supply chain graph is input into a graph neural network (GNN) model for feature extraction, and a semi-structured feature vector is output. It should be noted that the supply chain graph is constructed based on standardized supply chain relationship data. The construction process is as follows: enterprises, suppliers, and logistics nodes are defined as graph nodes, and transaction relationships are defined as edges (with weights, such as transaction amounts). A graph database (such as Neo4j) is used to store the graph, supporting real-time updates (e.g., triggered by Kafka streaming data), and the supply chain graph (node-edge structure) is output. For the obtained supply chain graph, a graph neural network (GNN) model can also be used to calculate node embeddings, analyze the customer's position in the supply chain (e.g., centrality PageRank), assess supply chain stability through edge weights (transaction amount volatility), detect abnormal subgraphs (e.g., excessively high supplier concentration), and predict node risks (e.g., default probability), outputting the supply chain graph features.

[0054] Please see Figure 4 In some embodiments, step S104 may include, but is not limited to, steps S401 to S402: Step S401: Determine the feature weights corresponding to structured features, unstructured features, and semi-structured features based on the multi-head attention mechanism; Step S402: Based on the feature weights, perform feature concatenation and weighted summation on the structured features, unstructured features, and semi-structured features to obtain the fused features.

[0055] In some embodiments, during feature fusion, corresponding feature weights can be assigned to structured features, unstructured features, and semi-structured features based on a multi-head attention mechanism. For example, a first feature weight is assigned to structured features, a second feature weight to unstructured features, and a third feature weight to semi-structured features. Then, feature concatenation and weighted summation are performed on the structured features, unstructured features, and semi-structured features based on the first feature weight, the second feature weight, and the third feature weight, respectively. That is, the fused feature = the first feature weight. Structured features + weight of second feature Unstructured features + third feature weights Semi-structured features.

[0056] Please see Figure 5 In some embodiments, step S107 may include, but is not limited to, steps S501 to S505: Step S501: Determine the resource allocation weights of each target sub-model in the risk assessment model according to the target processing strategy, wherein the target sub-model is any one of the text risk scoring model, graph risk scoring model and time series risk scoring model; Step S502: If the resource allocation weight of the target sub-model in the risk assessment model is determined to be zero, the target sub-model is disabled based on the gating mechanism. Step S503: If the resource allocation weight of the target sub-model in the risk assessment model is greater than zero and less than a preset threshold, the target sub-model is lightened. Step S504: If the resource allocation weight of the target sub-model in the risk assessment model is greater than the preset threshold, determine the target operating state of the risk assessment model according to the resource allocation weight of each target sub-model. Step S505: The fusion features are processed by a risk assessment model in the target operating state to generate the risk assessment results for the corresponding target enterprise.

[0057] It is understandable that the risk assessment model includes multiple target sub-models of different risk assessment types. These target sub-models include, but are not limited to, text risk scoring models, graph risk scoring models, and time-series risk scoring models. Among them, the text risk scoring model is used to output text risk scores, including but not limited to the Transformer model, which can be used to analyze risk signals in text (such as contract terms, public opinion news); the graph risk scoring model is used to output graph risk scores, including but not limited to the Graph Neural Network (GNN) model, which can be used to calculate node risks (such as supplier concentration) in the supply chain graph; and the time-series risk scoring model is used to output time-series risk scores, including but not limited to the Long Short-Term Memory (LSTM) network model, which can be used to analyze trends in time-series data (such as accounts payable turnover).

[0058] The resource allocation weights of each objective sub-model in the risk assessment model are determined by the objective processing strategy.

[0059] In one embodiment, when financial data is missing, the resource allocation weight for the target sub-model used to process structured features is zero. Based on the access control mechanism, the Long Short-Term Memory (LSTM) network model is disabled to reduce the corresponding resource waste, thereby achieving dynamic scheduling of computing resources.

[0060] In one embodiment, when the supply chain graph features were updated several months ago, the resource allocation weights allocated to the target sub-model used to process semi-structured features will be reduced. When it is determined that the resource allocation weight of the target sub-model is lower than a preset threshold, i.e., it is a low-weight model, the target sub-model can be lightweighted, such as model pruning or quantization, so that the allocation of computing resources to it can be reduced in resource-constrained scenarios, thereby achieving dynamic scheduling of computing resources.

[0061] In one embodiment, when the supply chain graph features are updated one day ago, the resource allocation weights allocated to the target sub-model for processing semi-structured features are increased, such as increasing the resource allocation weights of the graph neural network (GNN) model, thereby achieving dynamic scheduling of computing resources.

[0062] In one embodiment, when keywords such as "policy tightening" appear in news reports or social media comments, the Transformer model can be activated or the resource allocation weight of the Transformer model can be increased, thereby achieving dynamic scheduling of computing resources.

[0063] In one embodiment, if it is determined that the resource allocation weights of the target sub-models in the risk assessment model are all greater than a preset threshold, for example, if the Transformer model, the Graph Neural Network (GNN) model, and the Long Short-Term Memory (LSTM) network model are all activated and are not low-weight models, then the resource allocation weights of the corresponding Transformer model, GNN model, and LSTM network model are determined according to the target processing strategy to operate the risk assessment model. For example, the resource allocation weight of the Transformer model is 40%, the resource allocation weight of the GNN model is 30%, and the resource allocation weight of the LSTM network model is 30%. The target operating state of the risk assessment model is determined based on this resource allocation weight ratio.

[0064] In one embodiment, the fusion features are processed by a risk assessment model in the target operating state to generate the risk assessment results of the corresponding target enterprise.

[0065] For example, in conducting credit risk assessment of a target company, a target processing strategy is determined by analyzing the company's multimodal data. Based on this strategy, resources are allocated to the Transformer, Graph Neural Network (GNN), and Long Short-Term Memory (LSTM) models within the risk assessment model. If the Transformer model outputs a text risk score of 0.65 with a resource allocation weight of 40%, the GNN model outputs a graph risk score of 0.8 with a resource allocation weight of 30%, and the LSTM model outputs a time-series risk score of 0.75 with a resource allocation weight of 30%, then the final credit risk score calculated by the risk assessment model is 0.65. 0.4 + 0.8 0.3 + 0.75 0.3 = 0.725. If the high-risk level scoring threshold is 0.7, then the risk assessment result of the target company can be determined as high-risk, and a warning can be issued to the risk control department and corresponding measures can be proposed.

[0066] Please see Figure 6 In some embodiments, the enterprise evaluation method of this application may also include, but is not limited to, steps S601 to S602: Step S601: Generate a risk assessment report based on the risk assessment results; Step S602: Display the risk assessment report on the terminal's visual interface.

[0067] Understandably, the risk assessment results can be presented in a visual interface on terminals, including web or mobile devices. The risk assessment report can include risk scores, risk levels, risk factor analysis, risk node highlights, risk distribution heatmaps, and other content.

[0068] Please see Figure 7 In some embodiments, the enterprise evaluation method of this application may also include, but is not limited to, steps S701 to S702: Step S701: Obtain user interaction operation information; Step S702: Query or adjust the assessment content of the risk assessment report based on user interaction operation information.

[0069] Understandably, the enterprise data processing method in this application also supports user interaction. Based on user interaction information, users can not only request queries for risk assessment reports but also perform interactive operations, such as customizing risk level thresholds and triggering supply chain map updates by adding real-time transaction data. It should be noted that real-time transaction data is streamed to the Graph Neural Network (GNN) model via Kafka, which can trigger supply chain map updates and incremental training of the GNN model.

[0070] By acquiring multi-dimensional, multimodal data from target companies, such as financial, text, image, and graph data, and extracting features from this data to obtain different multimodal features, and then fusing these features to obtain fused features, this application not only combines the feature attributes of multimodal features with the risk assessment task requirements of the target company to conduct multi-dimensional risk assessments, resulting in a more comprehensive assessment, but also determines targeted processing strategies based on the feature attributes of the multimodal features and the risk assessment type. A risk assessment model containing multiple target sub-models processes the fused features according to the target processing strategies to generate risk assessment results for the corresponding target company. Since the resource allocation weights of each target sub-model in the risk assessment model are determined by the target processing strategies, resource weight allocation becomes more intelligent and adaptable, thereby improving the accuracy of enterprise risk assessment. Furthermore, through this dynamic activation mechanism of the target processing strategies, the activation or adjustment of the weights of target sub-models for corresponding risk assessment types can be determined. This allows for flexible adjustment of the risk assessment model structure in cases of data loss, changes in task requirements, or resource constraints, efficiently generating accurate risk assessment results for the target company.

[0071] Based on this, the embodiments of this application can improve the comprehensiveness and accuracy of enterprise risk assessment. When fusing and analyzing multimodal data, the resource allocation weight of the risk assessment model can be dynamically adjusted according to the target processing strategy, thereby reasonably allocating computing resources and reducing computing costs. In financial scenarios, it can achieve efficient and accurate enterprise risk assessment.

[0072] Please see Figure 8 This application also provides an enterprise data processing apparatus that can implement the above-described enterprise data processing method. The apparatus includes: The acquisition module 810 is used to acquire multimodal data of the target enterprise, including structured data, unstructured data and semi-structured data; The extraction module 820 is used to extract features from multimodal data to obtain different multimodal features; The fusion module 830 is used to fuse different multimodal features to obtain fused features; The identification module 840 is used to identify the feature attributes of multimodal features, including data integrity, timestamps, and risk types. Module 850 is used to determine the type of risk assessment based on the risk assessment task requirements of the target enterprise. Strategy module 860 is used to determine the target treatment strategy based on feature attributes and risk assessment type; The assessment module 870 is used to perform risk assessment processing on the fusion features according to the target processing strategy through the risk assessment model, and generate the risk assessment results of the corresponding target enterprise. The risk assessment model includes multiple target sub-models with different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy.

[0073] In some embodiments of this application, the acquisition module 810 acquires multimodal data of the target enterprise; the extraction module 820 extracts features from the target multimodal data to obtain different multimodal features; the fusion module 830 fuses the different multimodal features to obtain fused features; the identification module 840 identifies the feature attributes of the multimodal features, including data integrity, timestamp, and risk type; the requirement module 850 determines the risk assessment type based on the risk assessment task requirements of the target enterprise; the strategy module 860 determines the target processing strategy based on the feature attributes and the risk assessment type; and the evaluation module 870 performs risk assessment processing on the fused features according to the target processing strategy through the risk assessment model to generate the corresponding risk assessment result for the target enterprise. The risk assessment model includes multiple target sub-models of different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy.

[0074] In some embodiments of this application, multimodal data of the target enterprise is acquired. Classified by data attribute type, multimodal data can include text data, image data, graph data, and time-series data. Text data includes, but is not limited to, financial data, contract texts, news reports, and social media comments; image data includes, but is not limited to, photos of enterprise facilities, images of equipment, and scanned copies of invoices; graph data includes, but is not limited to, supply chain graph data, which is constructed based on supply chain relationship data; time-series data, also known as time series data, includes a series of data obtained in chronological order at different time periods or points in time, such as monthly sales trends of customers. Classified by data structure type, multimodal data can include structured data, unstructured data, and semi-structured data. Structured data includes, but is not limited to, financial data and transaction records; unstructured data includes, but is not limited to, contract texts, news reports (such as changes in industry policies), and social media comments (such as public opinion dynamics); semi-structured data includes, but is not limited to, supply chain graph data and logistics information in JSON / XML format (such as logistics transportation status and cargo details).

[0075] In some embodiments of this application, feature extraction is performed on the target multimodal data to obtain different multimodal features. The target multimodal data includes structured data, unstructured data, and semi-structured data.

[0076] Feature extraction is performed on structured data to obtain structured features; feature extraction is performed on unstructured data to obtain unstructured features; and feature extraction is performed on semi-structured data to obtain semi-structured features. Multimodal features are then derived based on these structured, unstructured, and semi-structured features.

[0077] In some embodiments of this application, feature fusion is performed on multimodal features to obtain fused features. Specifically, fusion can be performed through weighted summation, attention mechanisms, or gating mechanisms. For example, a multi-head attention mechanism can be used to dynamically weight different modal features to obtain fused features. Subsequently, the fused features can be used as a comprehensive input vector to input into a risk assessment model to improve the predictive ability of the risk assessment model.

[0078] In some embodiments of this application, since multimodal features themselves have feature attributes, the feature attributes include data integrity, timestamps and risk types, such as the data missing rate of features of different modalities such as structured features, unstructured features and semi-structured features, the reliability and real-time nature of timestamps, and whether the risk type is credit risk, market risk or operational risk, etc.

[0079] In some embodiments of this application, it is also necessary to determine the type of risk assessment for the target enterprise based on the requirements of the risk assessment task. The requirements of the risk assessment task include different types of risk assessment tasks such as credit risk assessment, market risk assessment, and operational risk assessment for the target enterprise.

[0080] In some embodiments of this application, a target processing strategy for risk assessment of a target enterprise is determined based on the feature attributes of multimodal features and the risk assessment type. This allows for the dynamic activation and scheduling of corresponding resources. For example, when the data missing rate of financial data features is high, the resource allocation weight for analyzing and processing structured features can be reduced. Similarly, when the supply chain map features were updated several months ago, the resource allocation weight for analyzing and processing semi-structured features can be reduced. Furthermore, when keywords such as "policy tightening" appear in news reports or social media comments, the resource allocation weight for analyzing and processing unstructured features can be increased. Based on this dynamic activation mechanism, targeted target processing strategies can be obtained according to the multimodal features and the risk assessment task requirements of the corresponding target enterprise. This allows for the dynamic adjustment of the activation status or weight allocation of target sub-models of different risk assessment types within the risk assessment model, thereby optimizing the use of computational resources and the model's assessment effectiveness.

[0081] In some embodiments of this application, a risk assessment model performs risk assessment processing on fused features according to a target processing strategy, generating risk assessment results for the corresponding target enterprise. The risk assessment model includes multiple target sub-models of different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy. By using a risk assessment model containing multiple target sub-models to perform risk assessment processing on fused features according to the target processing strategy, risk assessment results for the corresponding target enterprise are generated. Since the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy, the resource weight allocation is more intelligent and adaptive, thereby improving the accuracy of enterprise risk assessment. It should be noted that the risk assessment results can be presented through a visual interface on a web or mobile device. The risk assessment report can include risk scores, risk levels, risk factor analysis, risk node highlighting, risk distribution heatmaps, etc., and supports user interaction. Users can not only request queries for the risk assessment report but also perform interactive operations, such as customizing risk level thresholds and triggering supply chain map updates by adding real-time transaction data.

[0082] Based on this, the enterprise data processing apparatus of this application embodiment includes: an acquisition module 810 acquiring multimodal data of the target enterprise; an extraction module 820 extracting features from the multimodal data to obtain different multimodal features; a fusion module 830 fusing the different multimodal features to obtain fused features; an identification module 840 identifying the feature attributes of the multimodal features, including data integrity, timestamps, and risk types; a requirement module 850 determining the risk assessment type based on the risk assessment task requirements of the target enterprise; a strategy module 860 determining the target processing strategy based on the feature attributes and the risk assessment type; and an evaluation module 870 performing risk assessment processing on the fused features according to the target processing strategy using a risk assessment model to generate the corresponding risk assessment result for the target enterprise. The risk assessment model includes multiple target sub-models of different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy. This application acquires the target company's original multimodal data; extracts features from the multimodal data to obtain different multimodal features; fuses these different multimodal features to obtain fused features; identifies the feature attributes of the multimodal features, including data integrity, timestamps, and risk types; determines the risk assessment type based on the target company's risk assessment task requirements; determines the target processing strategy based on the feature attributes and risk assessment type; and performs risk assessment processing on the fused features using a risk assessment model according to the target processing strategy to generate the corresponding risk assessment results for the target company. The risk assessment model includes multiple target sub-models with different risk assessment types, and the resource allocation weights of each target sub-model in the risk assessment model are determined by the target processing strategy. By acquiring multi-dimensional multimodal data of the target company, extracting features from the target multimodal data to obtain different multimodal features, and fusing these different multimodal features to obtain fused features, this application achieves the desired results. This application not only combines the feature attributes of multimodal features with the risk assessment task requirements of the target enterprise to conduct multi-dimensional risk assessment of the enterprise, making the assessment dimensions more comprehensive, but also determines targeted target processing strategies based on the feature attributes of multimodal features and risk assessment types. Furthermore, a risk assessment model containing multiple target sub-models processes the fused features according to the target processing strategies to generate risk assessment results for the corresponding target enterprise. Since the resource allocation weights of each target sub-model in the risk assessment model are determined by the target processing strategies, the resource weight allocation is more intelligent and adaptable, thereby improving the accuracy of enterprise risk assessment. Moreover, through this dynamic activation mechanism of the target processing strategies, the activation or adjustment of the weights of target sub-models for corresponding risk assessment types can be determined. This allows for flexible adjustment of the risk assessment model structure in cases of data loss, changes in task requirements, or resource constraints, efficiently generating accurate risk assessment results for the target enterprise.Based on this, the embodiments of this application can improve the comprehensiveness and accuracy of enterprise risk assessment. When fusing and analyzing multimodal data, the resource allocation weight of the risk assessment model can be dynamically adjusted according to the target processing strategy, thereby reasonably allocating computing resources and reducing computing costs. In financial scenarios, it can achieve efficient and accurate enterprise risk assessment.

[0083] The specific implementation of the enterprise data processing device is basically the same as the specific implementation of the enterprise data processing method described above, and will not be repeated here.

[0084] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned enterprise data processing method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0085] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0086] The memory 902 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 902 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and called and executed by the processor 901. Specifically, the method involves: acquiring multimodal data of the target enterprise; extracting features from the multimodal data to obtain different multimodal features; fusing the different multimodal features to obtain fused features; identifying the feature attributes of the multimodal features, including data integrity, timestamps, and risk types; determining the risk assessment type based on the risk assessment task requirements of the target enterprise; determining the target processing strategy based on the feature attributes and risk assessment type; and performing risk assessment processing on the fused features according to the target processing strategy using a risk assessment model to generate the corresponding risk assessment result for the target enterprise. The risk assessment model includes multiple target sub-models of different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy. By acquiring multi-dimensional, multimodal data of the target enterprise and extracting features from this data, different multimodal features are obtained. Furthermore, these different multimodal features are fused to obtain fused features. This embodiment not only combines the feature attributes of multimodal features with the risk assessment task requirements of the target enterprise to conduct multi-dimensional risk assessment, resulting in a more comprehensive assessment, but also determines targeted processing strategies based on the feature attributes of the multimodal features and the risk assessment type. A risk assessment model containing multiple target sub-models processes the fused features according to the target processing strategy, generating risk assessment results for the corresponding target enterprise. Since the resource allocation weights of each target sub-model in the risk assessment model are determined by the target processing strategy, resource weight allocation becomes more intelligent and adaptable, thereby improving the accuracy of enterprise risk assessment. Moreover, through this dynamic activation mechanism of the target processing strategy, the activation or adjustment of the weights of the target sub-models for the corresponding risk assessment type can be determined. This allows for flexible adjustment of the risk assessment model structure in cases of data loss, changes in task requirements, or resource constraints, efficiently generating accurate risk assessment results for the target enterprise. Based on this, the embodiments of this application can improve the comprehensiveness and accuracy of enterprise risk assessment. When fusing and analyzing multimodal data, the resource allocation weight of the risk assessment model can be dynamically adjusted according to the target processing strategy, thereby reasonably allocating computing resources and reducing computing costs. In financial scenarios, it can achieve efficient and accurate enterprise risk assessment.

[0087] The input / output interface 903 is used to implement information input and output.

[0088] The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0089] The bus transmits information between various components of the device, such as processor 901, memory 902, input / output interface 903, and communication interface 904.

[0090] The processor 901, memory 902, input / output interface 903, and communication interface 904 communicate with each other within the device via a bus.

[0091] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described enterprise data processing method.

[0092] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0093] The enterprise data processing method, enterprise data processing device, electronic device, and storage medium provided in this application embodiment acquire multimodal data of a target enterprise; extract features from the multimodal data to obtain different multimodal features; fuse the different multimodal features to obtain fused features; identify the feature attributes of the multimodal features, including data integrity, timestamps, and risk types; determine the risk assessment type according to the risk assessment task requirements of the target enterprise; determine the target processing strategy based on the feature attributes and risk assessment type; and perform risk assessment processing on the fused features using a risk assessment model according to the target processing strategy to generate the corresponding risk assessment result for the target enterprise. The risk assessment model includes multiple target sub-models of different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy. By acquiring multi-dimensional multimodal data of the target enterprise, extracting features from the multimodal data to obtain different multimodal features, and fusing the different multimodal features to obtain fused features, this process achieves the desired results. This application not only combines the feature attributes of multimodal features with the risk assessment task requirements of the target enterprise to conduct multi-dimensional risk assessment of the enterprise, making the assessment dimensions more comprehensive, but also determines targeted target processing strategies based on the feature attributes of multimodal features and risk assessment types. A risk assessment model containing multiple target sub-models processes the fused features according to the target processing strategies to generate risk assessment results for the corresponding target enterprise. Since the resource allocation weights of each target sub-model in the risk assessment model are determined by the target processing strategies, the resource weight allocation is more intelligent and adaptable, thereby improving the accuracy of enterprise risk assessment. Furthermore, through this dynamic activation mechanism of the target processing strategies, the activation or adjustment weights of target sub-models of corresponding risk assessment types can be determined. This allows for flexible adjustment of the risk assessment model structure in cases of data loss, changes in task requirements, or resource constraints, efficiently generating accurate risk assessment results for the target enterprise. Based on this, this application embodiment can improve the comprehensiveness and accuracy of enterprise risk assessment. Moreover, when fusing and analyzing multimodal data, it can dynamically adjust the resource allocation weights of the risk assessment model according to the target processing strategies, thereby rationally allocating computing resources and reducing computing costs. In financial scenarios, it can achieve efficient and accurate enterprise risk assessment.

[0094] Those skilled in the art will understand that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable programs, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable programs, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0095] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0096] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0098] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0099] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0100] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0102] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0103] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0104] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for processing enterprise data, characterized in that, The method includes: Acquire multimodal data of the target enterprise, including structured data, unstructured data, and semi-structured data; Feature extraction is performed on the multimodal data to obtain different multimodal features; The different multimodal features are fused to obtain fused features; Identify the feature attributes of the multimodal features, including data integrity, timestamp, and risk type; The risk assessment type is determined based on the risk assessment task requirements of the target enterprise. Determine the target treatment strategy based on the aforementioned characteristic attributes and the aforementioned risk assessment type; The risk assessment model performs risk assessment processing on the fusion features according to the target processing strategy, generating risk assessment results corresponding to the target enterprise. The risk assessment model includes multiple target sub-models with different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy.

2. The method according to claim 1, characterized in that, The acquisition of multimodal data of the target enterprise includes: The structured data of the target enterprise can be obtained in real time through an application programming interface; The unstructured data of the target company is crawled using web crawlers. The semi-structured data of the target enterprise is received through an IoT device.

3. The method according to claim 1, characterized in that, The feature extraction of the target multimodal data to obtain different multimodal features includes: Feature extraction is performed on the structured data to obtain structured features; Feature extraction is performed on the unstructured data to obtain unstructured features; Feature extraction is performed on the semi-structured data to obtain semi-structured features; Different multimodal features are obtained based on the structured features, the unstructured features, and the semi-structured features.

4. The method according to claim 3, characterized in that, The feature fusion of different multimodal features to obtain fused features includes: The feature weights corresponding to the structured features, unstructured features, and semi-structured features are determined based on a multi-head attention mechanism. Based on the feature weights, the structured features, unstructured features, and semi-structured features are concatenated and weighted to obtain the fused features.

5. The method according to claim 1, characterized in that, The step of performing risk assessment processing on the fusion features according to the target processing strategy using a risk assessment model to generate risk assessment results corresponding to the target enterprise includes: The resource allocation weights of each target sub-model in the risk assessment model are determined according to the target processing strategy, wherein the target sub-model is any one of a text risk scoring model, a graph risk scoring model, and a time-series risk scoring model; If the resource allocation weight of the target sub-model in the risk assessment model is determined to be zero, the target sub-model is disabled based on a gating mechanism. If the resource allocation weight of the target sub-model in the risk assessment model is determined to be greater than zero and less than a preset threshold, the target sub-model is lightweighted. If the resource allocation weight of the target sub-model in the risk assessment model is determined to be greater than the preset threshold, the target operating state of the risk assessment model is determined according to the resource allocation weight of each target sub-model. The risk assessment model, which is in the target operating state, performs risk assessment processing on the fusion features to generate risk assessment results corresponding to the target enterprise.

6. The method according to claim 1, characterized in that, The method further includes: Generate a risk assessment report based on the risk assessment results; The risk assessment report is displayed on the terminal's visual interface.

7. The method according to claim 6, characterized in that, The method further includes: Obtain user interaction information; The user interaction information can be used to query or adjust the assessment content of the risk assessment report.

8. An enterprise data processing device, characterized in that, The device includes: The acquisition module is used to acquire multimodal data of the target enterprise, including structured data, unstructured data, and semi-structured data. The extraction module is used to extract features from the multimodal data to obtain different multimodal features; The fusion module is used to fuse the different multimodal features to obtain fused features; The identification module is used to identify the feature attributes of the multimodal features, including data integrity, timestamp, and risk type; The requirements module is used to determine the risk assessment type based on the risk assessment task requirements of the target enterprise; The strategy module is used to determine the target processing strategy based on the feature attributes and the risk assessment type; The assessment module is used to perform risk assessment processing on the fusion features according to the target processing strategy through a risk assessment model, and generate risk assessment results corresponding to the target enterprise. The risk assessment model includes multiple target sub-models with different risk assessment types, and the resource allocation weight of each target sub-model in the risk assessment model is determined by the target processing strategy.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the enterprise data processing method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the enterprise data processing method according to any one of claims 1 to 7.

Images