Enterprise digital transformation scheme matching method, system and equipment based on large model

Abstract

The invention provides an enterprise digital transformation scheme matching method, system and device based on a large model, and belongs to the technical field of computers. The method comprises the following steps: acquiring a multi-mode operation panoramic data set from a user terminal; and determining a corresponding first feature vector based on a preset graph neural network and first preset feature data in the multi-modal operation panoramic data set, and inputting the multi-modal operation panoramic data set and the first feature vector into a pre-trained enterprise portrait model to construct an enterprise portrait vector of the corresponding to-be-transformed enterprise. Inputting the enterprise portrait vector into a preset transformation demand large model so as to determine transformation demand information of the to-be-transformed enterprise according to a model output result; and matching one or more candidate transformation schemes in a preset transformation case based on the transformation demand information, determining corresponding transformation assistance values based on the candidate transformation schemes and the multi-modal operation panoramic data set, generating corresponding transformation effect evaluation results, and sending the transformation effect evaluation results to the user terminal.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, system and device for matching enterprise digital transformation solutions based on a large model. Background Technology

[0002] Enterprise digital transformation refers to the process of reconstructing business models through technologies such as artificial intelligence, big data, and cloud computing to achieve data-driven decision-making and innovation.

[0003] Currently, enterprises typically draw on successful case studies for digital transformation, a process that often relies on expert experience or static rule bases. However, expert experience is prone to personal bias and subjective influence, making it highly subjective; static rule bases, on the other hand, cannot dynamically adjust digital transformation solutions, resulting in relatively poor flexibility.

[0004] Therefore, there is an urgent need for a technical solution that can objectively and flexibly match enterprise digital transformation solutions. Summary of the Invention

[0005] This application provides a method, system, and device for matching enterprise digital transformation solutions based on a large model, which addresses the problem of how to provide a technical solution that can objectively and flexibly match enterprise digital transformation solutions.

[0006] In a first aspect, embodiments of this application provide a method for matching enterprise digital transformation solutions based on a large model, the method comprising: Obtain a multimodal operational panorama dataset from user terminals; wherein, the multimodal operational panorama dataset includes structured data and / or unstructured data generated by current enterprise operations; Based on the preset graph neural network and the first preset feature data in the multimodal operation panoramic dataset, the corresponding first feature vector is determined, and the multimodal operation panoramic dataset and the first feature vector are input into the pre-trained enterprise profile model to construct the corresponding enterprise profile vector of the enterprise to be transformed. The enterprise profile vector is input into a preset transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model output results. Based on the transformation demand information, match one or more candidate transformation schemes from the preset transformation cases, and based on each candidate transformation scheme and the multimodal operation panoramic dataset, determine the corresponding transformation assistance value and generate the corresponding transformation effectiveness evaluation result, which is then sent to the user terminal.

[0007] In one implementation of this application, a corresponding first feature vector is determined based on a preset graph neural network and first preset feature data in the multimodal operation panoramic dataset, specifically including: According to preset feature data filtering rules, the first preset feature data is filtered from the multimodal operation panoramic dataset; wherein, the first preset feature data corresponds to the internal and external relationships of the enterprise to be transformed; the internal and external relationships include at least supplier relationships and customer relationships; Based on a preset graph structure modeling algorithm, the nodes in the first preset feature data and the relational attributes corresponding to each node are determined to construct a node attribute matrix. Interaction data between each node corresponding to the relational attributes is extracted from the first preset feature data to construct an adjacency matrix based on the interaction data. The node attribute matrix and the adjacency matrix are input into the preset graph neural network to extract the local and global relationship features of the internal and external relationships, and then concatenate them to obtain the first feature vector.

[0008] In one implementation of this application, the multimodal operation panorama dataset and the first feature vector are input into a pre-trained enterprise profile model to construct an enterprise profile vector for the corresponding enterprise to be transformed, specifically including: The second feature vector corresponding to the structured modality, the third feature vector corresponding to the unstructured modality, and the first feature vector in the multimodal operation panoramic dataset are concatenated to obtain a fused feature matrix; By using a pre-defined multi-head attention mechanism, the modalities in the fusion feature matrix are fused to obtain the corresponding enterprise profile vector.

[0009] In one implementation of this application, before inputting the enterprise profile vector into a preset transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model output, the method further includes: The industry knowledge graph corresponding to the enterprise to be transformed is embedded into the preset transformation demand model, and a preset graph attention network is added to the preset transformation demand model. By using a pre-defined contrastive learning strategy, pre-defined positive and negative sample pairs are added to the training samples to train the corresponding pre-defined transformation requirement large model, thereby performing knowledge augmentation processing on the large model.

[0010] In one implementation of this application, the enterprise profile vector is input into a preset transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model output results, specifically including: The enterprise profile vector is input into the preset transformation demand model to determine the transformation demand labels and demand priorities output by the model. Based on the aforementioned priority of requirements, each of the aforementioned transformation requirement tags will be added sequentially to the corresponding transformation requirement sequence; Based on the transformation demand sequence, the multimodal operation panoramic dataset, and the enterprise profile vector, demand analysis is performed through a preset interpretable model to generate a demand analysis report to determine the key influencing factors of each of the transformation demand tags. Add the transformation demand sequence and the key influencing factors to the transformation demand information.

[0011] In one implementation of this application, matching one or more candidate transformation schemes from preset transformation cases based on the transformation requirement information specifically includes: The transformation requirement information is vector-encoded into a first matching vector; The first matching vector is calculated using a collaborative filtering algorithm to determine the matching similarity between the first matching vector and each second matching vector in the preset transformation cases. Based on the matching similarity and the preset time decay factor corresponding to each transformation case, one or more candidate transformation schemes are determined.

[0012] In one implementation of this application, the method further includes: Based on a pre-built reinforcement learning model, the collaborative filtering algorithm is updated and / or each of the candidate transformation schemes is modified. The method further includes: The updated data of the multimodal operation panoramic dataset is obtained in real time through a preset streaming computing framework. If the amount of updated data exceeds a preset threshold, the matching transformation scheme is updated.

[0013] In one implementation of this application, based on each of the candidate transformation schemes and the multimodal operation panoramic dataset, corresponding transformation assistance values ​​are determined and corresponding transformation effectiveness evaluation results are generated, specifically including: The multimodal operation panoramic dataset and the implementation operation panoramic dataset corresponding to each of the candidate transformation schemes are input into a preset two-layer LSTM network to determine the predicted value of the indicator data corresponding to each transformation demand label. Based on the preset set of indicator weights and the predicted values ​​of each indicator data, the transformation assistance value of each indicator is calculated, and a radar chart is constructed with each transformation demand label as the axis label and the predicted values ​​of each indicator data of the candidate transformation scheme as the data points; wherein, the predicted values ​​of each indicator data of the same candidate transformation scheme in the radar chart constitute a data area. The transformation assistance values ​​and the radar chart are used as the transformation effectiveness evaluation results and sent to the user terminal.

[0014] Secondly, embodiments of this application also provide a large-scale enterprise digital transformation solution matching system, the system comprising: The acquisition module is used to acquire a multimodal operational panorama dataset from the user terminal; wherein, the multimodal operational panorama dataset includes structured data and / or unstructured data generated by the current enterprise operation; The input determination module is used to determine the corresponding first feature vector based on the preset graph neural network and the first preset feature data in the multimodal operation panoramic dataset, and input the multimodal operation panoramic dataset and the first feature vector into the pre-trained enterprise profile model to construct the corresponding enterprise profile vector of the enterprise to be transformed. The input module is used to input the enterprise profile vector into a preset transformation demand model, so as to determine the transformation demand information of the enterprise to be transformed based on the model output results. The matching and determination module is used to match one or more candidate transformation schemes from preset transformation cases based on the transformation demand information, and to determine the corresponding transformation assistance values ​​and generate corresponding transformation effectiveness evaluation results based on each candidate transformation scheme and the multimodal operation panoramic dataset, and send them to the user terminal.

[0015] Thirdly, this application also provides a matching device for enterprise digital transformation solutions based on a large model, the device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described enterprise digital transformation solution matching method based on a large model.

[0016] Compared with the prior art, the significant advantages of this application are as follows: Through the above-described approach, this application constructs a multi-dimensional profile of enterprises, thereby accurately identifying transformation needs and dynamically matching suitable digital transformation cases based on a large model. It also predicts the effectiveness of solution implementation in real time to quantify the transformation support value. This provides a reasonable and effective technical solution for objectively and flexibly matching enterprise digital transformation solutions, addressing the problems of existing technologies over-reliance on human experience, subjective factors in matching transformation solutions, and the lack of flexibility in static rule bases. This improves the efficiency of matching enterprise digital transformation solutions. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of a process for matching enterprise digital transformation solutions based on a large model, as described in this application. Figure 2 This is a schematic diagram of the structure of a large-scale enterprise digital transformation solution matching system according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a matching device for an enterprise digital transformation solution based on a large model, as described in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Currently, enterprises typically draw on successful case studies for digital transformation, a process that often relies on expert experience or static rule bases. However, expert experience is prone to personal bias and subjective influence, making it highly subjective; static rule bases, on the other hand, cannot dynamically adjust digital transformation solutions, resulting in relatively poor flexibility.

[0020] Based on this, embodiments of this application provide a method, system, and device for matching enterprise digital transformation solutions based on a large model, in order to solve the problem of how to provide a technical solution that can objectively and flexibly match enterprise digital transformation solutions.

[0021] The various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0022] This application provides a method for matching enterprise digital transformation solutions based on a large model, such as... Figure 1 As shown, the method may include steps S101-S104: S101, the server obtains a multimodal operational panorama dataset from the user terminal.

[0023] The multimodal operation panorama dataset includes structured and / or unstructured data generated from current enterprise operations.

[0024] It should be noted that the server, as the executor of the enterprise digital transformation solution matching method based on the large model, exists only as an example. The executor is not limited to the server, and this application does not make any specific limitation on it.

[0025] User terminals can be understood as terminals connected to the server via wired or wireless means, and may include, but are not limited to, mobile phones, computers, cloud platforms, etc. The multimodal operational panorama dataset contains multi-dimensional enterprise data, such as structured data (financial indicators, profit margins, cash flow), supply chain efficiency (inventory turnover rate, logistics costs), IT system logs (API call frequency, system load), etc.) and unstructured data (enterprise strategy documents, employee feedback texts, industry research reports, etc.). This data can be actively sent to the server by the user, or it can be retrieved by the server from the corresponding storage area of ​​the user terminal after user authorization. The specific method for the server to retrieve the multimodal operational panorama dataset is not specifically limited here.

[0026] In addition, after obtaining the multimodal operation panoramic dataset, the data can be preprocessed before performing subsequent analysis and processing steps, including: cleaning and standardization: using a rule engine to remove outliers and normalizing heterogeneous data (such as text, numerical data, and time series data); feature extraction: using the BERT model to extract text semantic features and using PCA (principal component analysis) technology to reduce the dimensionality of the data.

[0027] S102, the server determines the corresponding first feature vector based on the first preset feature data in the preset graph neural network and the multimodal operation panoramic dataset, and inputs the multimodal operation panoramic dataset and the first feature vector into the pre-trained enterprise profile model to construct the enterprise profile vector of the corresponding enterprise to be transformed.

[0028] In this embodiment of the application, the determination of the corresponding first feature vector based on the first preset feature data in the preset graph neural network and the multimodal operation panoramic dataset specifically includes: According to preset feature data filtering rules, first preset feature data is filtered from the multimodal operation panoramic dataset. This first preset feature data corresponds to the internal and external relationships of the enterprise to be transformed. Internal and external relationships include at least supplier relationships and customer relationships. Based on a preset graph structure modeling algorithm, the nodes in the first preset feature data and the corresponding relationship attributes of each node are determined to construct a node attribute matrix. Interaction data between nodes corresponding to their relationship attributes is extracted from the first preset feature data to construct an adjacency matrix. The node attribute matrix and adjacency matrix are input into a preset graph neural network to extract local and global relationship features of the internal and external relationships, and then concatenated to obtain the first feature vector.

[0029] In other words, this application can analyze the internal and external relationships of enterprises undergoing transformation (enterprises that have pre-input a multimodal operational panorama dataset and generated such a dataset). These internal and external relationships can at least include supplier relationships and customer relationships. In practical applications, users can further expand the relational objects corresponding to these internal and external relationships; no specific limitations are made here. The preset feature data filtering rules can be a pre-set set of data extraction labels that can be compared and filtered to obtain the first preset feature data corresponding to the internal and external relationships within the multimodal operational panorama dataset. Alternatively, the preset feature data filtering rules can also invoke a lightweight model to filter the multimodal operational panorama dataset, thereby obtaining the first preset feature data; no specific limitations are made here.

[0030] Secondly, this application also includes a preset graph structure modeling algorithm, capable of extracting nodes from the first preset feature data, i.e., entities predefined by the preset graph structure modeling algorithm, such as the enterprise to be transformed "AA", supplier "BB", customer "CC", etc., and simultaneously acquiring the relationship attributes between nodes. Relationship attributes can be understood as the relationships between nodes, such as transaction matching, etc. The preset graph structure modeling algorithm constructs a node attribute matrix with nodes as rows and node attributes as columns. Furthermore, it extracts specific interaction data between nodes according to node attributes from the first preset feature data. A specific interaction data may have multiple data sub-dimensions. This application can pre-acquire multi-dimensional weighted reorganization from an expert system, calculate the weighted sum of the multiple data sub-dimensions corresponding to the interaction data, obtain the unique edge weight corresponding to the corresponding interaction data, and construct an edge weight association matrix between each node as an adjacency matrix.

[0031] Subsequently, this application employs a pre-defined graph neural network to further extract features from the node attribute matrix and adjacency matrix. Before feature extraction, the pre-defined graph neural network (such as a GNN or GCN model) can first standardize the node attribute matrix and adjacency matrix, such as through matrix normalization. Then, features between each node and its direct neighbors (i.e., 1-hop neighbors) are extracted and aggregated to obtain local relation features. Based on the extracted local relation features, features between 2-hop or multi-hop neighbors are extracted and aggregated to obtain global relation features. The extracted features (matrices) are then concatenated and compressed to obtain the first feature vector.

[0032] The above scheme yields a feature vector that integrates the internal and external relationships of the enterprise to be transformed. This feature vector is personalized for the enterprise and plays a relatively important role in the digital transformation process. This application extracts the feature vector separately through the above steps, which can ensure that the impact of this information on the matching of the scheme cannot be ignored during the subsequent digital transformation, thereby improving the personalization of the matching scheme.

[0033] In addition, users can also specify and filter other preset feature data to generate feature vectors separately through the above embodiments. This application does not make specific limitations in this regard.

[0034] In this embodiment of the application, the above-mentioned inputting the multimodal operation panoramic dataset and the first feature vector into the pre-trained enterprise profile model to construct the corresponding enterprise profile vector of the enterprise to be transformed specifically includes: The second feature vector corresponding to the structured modality, the third feature vector corresponding to the unstructured modality, and the first feature vector in the multimodal operation panorama dataset are concatenated to obtain a fused feature matrix. A pre-defined multi-head attention mechanism is then used to fuse the modalities within the fused feature matrix to obtain the corresponding enterprise profile vector.

[0035] In other words, this application performs vector encoding on the data in the multimodal operation panorama dataset according to modality type. The modality type includes structured modalities, unstructured modules, and relational modalities obtained in the above steps. The encoded second feature vector, third feature vector, and the aforementioned first feature vector are concatenated to construct a fusion feature matrix. Simultaneously, a preset multi-head attention mechanism is used to combine the various modalities in the fusion feature matrix. Each attention head calculates its attention weight, and then the output features of each attention head are concatenated. Semantic integration is performed using linear transformation, and after residual connection and layer normalization, an enterprise profile vector is generated.

[0036] By integrating data from multiple modalities to construct enterprise profile vectors, a comprehensive positioning of the enterprise can be achieved, enabling a full and reasonable match with the enterprise's digital transformation solutions.

[0037] S103, the server inputs the enterprise profile vector into the preset transformation demand model, and determines the transformation demand information of the enterprise to be transformed based on the model output results.

[0038] In this embodiment of the application, before inputting the enterprise profile vector into a preset transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model output, the method further includes: An industry knowledge graph corresponding to the enterprise to be transformed is embedded into a pre-defined transformation demand model, and a pre-defined graph attention network is added to the pre-defined transformation demand model. Through a pre-defined contrastive learning strategy, pre-defined positive and negative sample pairs are added to the training samples to train the corresponding pre-defined transformation demand model, thereby performing knowledge enhancement processing on the model.

[0039] In other words, this application performs knowledge enhancement processing on the large model of pre-defined transformation needs, specifically by embedding industry knowledge graphs (such as manufacturing process chains and retail consumption trends) into the model, enhancing contextual understanding through graph attention networks, and adopting a contrastive learning strategy to introduce positive and negative sample pairs (such as success cases and failure cases) during training to improve the model's ability to generalize to industry differences.

[0040] Furthermore, in this embodiment of the application, the enterprise profile vector is input into a preset transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model output results, specifically including: The enterprise profile vector is input into a pre-defined transformation demand model to determine the transformation demand labels and priorities output by the model. Based on the demand priorities, each transformation demand label is added sequentially to its corresponding transformation demand sequence. Based on the transformation demand sequence, the multimodal operation panorama dataset, and the enterprise profile vector, demand analysis is performed using a pre-defined interpretable model to generate a demand analysis report to determine the key influencing factors for each transformation demand label. The transformation demand sequence and key influencing factors are then added to the transformation demand information.

[0041] In other words, by using a pre-built large-scale transformation demand model, the enterprise profile vector is processed to obtain transformation demand labels and priorities from the model output. Transformation demand labels can be understood as pain point labels for enterprises facing digital transformation, such as "supply chain response delays" and "IT system silos." The large-scale transformation demand model can specifically employ a pre-trained model based on the Transformer-XL architecture, which supports long-sequence input, such as inputting three consecutive years of operational data from the enterprise. According to the priority of the transformation demand labels, each label is arranged sequentially to obtain a transformation demand sequence. Then, a demand analysis report is generated using a pre-set interpretability module (such as LIME), displaying the key influencing factors in the demand analysis report. Key influencing factors can be understood as the reasons that affect the corresponding demand priorities of each transformation demand label. Combining the transformation demand sequence and key influencing factors yields transformation demand information. This solution provides interpretable matching logic, improving enterprise decision-making efficiency.

[0042] S104, the server matches one or more candidate transformation schemes from the preset transformation cases based on the transformation demand information, and determines the corresponding transformation assistance values ​​based on each candidate transformation scheme and the multimodal operation panoramic dataset, and generates the corresponding transformation effectiveness evaluation results, which are then sent to the user terminal.

[0043] Matching candidate transformation solutions can be understood as having a pre-built dynamic solution matching module in this application to complete solution matching by executing the following embodiments. In the embodiments of this application, the above-mentioned matching of one or more candidate transformation solutions from preset transformation cases based on transformation requirement information specifically includes: The transformation requirement information is vectorized into a first matching vector. A collaborative filtering algorithm is used to calculate the similarity between the first matching vector and each second matching vector in the preset transformation cases. Based on each similarity and a preset time decay factor corresponding to each transformation case, one or more candidate transformation schemes are determined.

[0044] In other words, this application can use collaborative filtering to perform similarity matching between the first matching vector corresponding to the transformation demand information and each of the second matching vectors in the preset transformation cases, so as to filter out similar transformation cases based on the above-mentioned enterprise profile. At the same time, a time decay factor is introduced to prioritize the recommendation of recent successful cases. That is, for example, the more recent the time, the larger the time decay factor. The product of the time decay factor and the corresponding matching similarity is used as the filtering index value. Transformation cases with larger filtering index values ​​or values ​​greater than the predetermined value are selected as candidate transformation solutions. The specific filtering method can be set by the user according to the actual use scenario, and is not specifically limited here.

[0045] Furthermore, in one embodiment of this application, the application further includes: Based on a pre-built reinforcement learning model, the collaborative filtering algorithm is updated and / or each candidate transformation scheme is modified.

[0046] That is, reinforcement learning is used to optimize the model in the above steps. Reinforcement learning includes a state space: real-time enterprise data (such as inventory changes and market fluctuations); an action space: scheme adjustment strategies (such as replacing modules and increasing or decreasing resource allocation); and a reward function: dynamically updating the model weights based on the changes in the assist value after implementation to maximize long-term benefits.

[0047] Furthermore, in another embodiment of this application, the method further includes: By using a pre-defined streaming computing framework, updated data from a multimodal operational panorama dataset is acquired in real time. When the volume of updated data exceeds a preset threshold, a matching transformation plan is updated.

[0048] In other words, this application can specifically use a streaming computing framework such as Apache Flink to process enterprise data updates. When enterprise data is updated or the amount of updated data exceeds a preset threshold, a scheme re-matching is triggered. The preset threshold can be set by the user or based on expert experience, and is not specifically limited here.

[0049] In this embodiment, the application further includes an effectiveness evaluation module, which is specifically capable of determining the corresponding transformation assistance values ​​and generating corresponding transformation effectiveness evaluation results based on each candidate transformation scheme and the multimodal operation panoramic dataset. The steps include: The multimodal operation panorama dataset and the corresponding implementation operation panorama datasets for each candidate transformation scheme are input into a pre-defined two-layer Long Short-Term Memory (LSTM) network to determine the predicted values ​​of the indicator data corresponding to each transformation requirement label. Based on the pre-defined indicator weight set and the predicted values ​​of each indicator data, each transformation contribution value is calculated. A radar chart is constructed with each transformation requirement label as the axis label and the predicted values ​​of each indicator data for the candidate transformation scheme as the data points. In the radar chart, the predicted values ​​of each indicator data for the same candidate transformation scheme constitute a data region. The transformation contribution values ​​and the radar chart are sent to the user terminal as the transformation effectiveness evaluation results.

[0050] In other words, this application uses a two-layer LSTM network, taking as input the sequence of relevant enterprise indicators from a multimodal operational panorama dataset before and after the implementation of candidate transformation plans for the enterprise to be transformed, and predicting key indicator values ​​for a predetermined time, such as the next 6 months. This prediction process can also incorporate adversarial training to enhance the model's robustness to noisy data. Subsequently, this application can also use methods such as the Analytic Hierarchy Process (AHP) to obtain an indicator weight set, or directly use a preset indicator weight set from user terminals or other expert systems or externally connected devices to perform weighted calculations on the predicted values ​​of each indicator data, obtaining the transformation assistance value corresponding to each predicted indicator data value. The calculation formula is as follows: Transformation Assist Value = ,in, The total number of tags representing transformation needs. Indicates the first The weight values ​​of indicators for each transformation demand label. Indicates the first This application provides predicted values ​​for key performance indicators (KPIs) for each transformation requirement tag. Furthermore, it can generate radar charts to display the improvement effects across various dimensions (transformation requirement tags), enabling comparison of multiple solutions.

[0051] The aforementioned implementation and operation panoramic dataset can be generated by experts or obtained from the implementation prediction model pre-set by the server of this application. The model can be a pre-trained neural network model or other prediction models, which are not specifically limited here.

[0052] Through the above-described approach, this application constructs a multi-dimensional profile of enterprises, thereby accurately identifying transformation needs and dynamically matching suitable digital transformation cases based on a large model. It also predicts the effectiveness of solution implementation in real time to quantify the transformation support value. This provides a reasonable and effective technical solution for objectively and flexibly matching enterprise digital transformation solutions, addressing the problems of existing technologies over-reliance on human experience, subjective factors in matching transformation solutions, and the lack of flexibility in static rule bases. This improves the efficiency of matching enterprise digital transformation solutions.

[0053] Figure 2 A schematic diagram of a large-model-based enterprise digital transformation solution matching system provided in this application embodiment is shown below. Figure 2 As shown, the enterprise digital transformation solution matching system 200 based on a large model includes: The acquisition module 201 is used to acquire a multimodal operational panorama dataset from the user terminal. This dataset includes structured and / or unstructured data generated by the current enterprise operations. The input determination module 202 is used to determine a corresponding first feature vector based on a preset graph neural network and first preset feature data in the multimodal operational panorama dataset. The multimodal operational panorama dataset and the first feature vector are then input into a pre-trained enterprise profile model to construct an enterprise profile vector for the enterprise to be transformed. The input module 203 is used to input the enterprise profile vector into a preset transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model's output. The matching determination module 204 is used to match one or more candidate transformation schemes from preset transformation cases based on the transformation demand information. Based on each candidate transformation scheme and the multimodal operational panorama dataset, it determines corresponding transformation assistance values ​​and generates corresponding transformation effectiveness evaluation results, which are then sent to the user terminal.

[0054] Figure 3 A schematic diagram of the structure of a matching device for an enterprise digital transformation solution based on a large model, as provided in this application embodiment, is shown below. Figure 3 As shown, the device includes: At least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Acquire a multimodal operational panorama dataset from the user terminal. This dataset includes structured and / or unstructured data generated from current enterprise operations. Based on a pre-defined graph neural network and first pre-defined feature data from the multimodal operational panorama dataset, determine the corresponding first feature vector. Input the multimodal operational panorama dataset and the first feature vector into a pre-trained enterprise profile model to construct the corresponding enterprise profile vector for the enterprise to be transformed. Input the enterprise profile vector into a pre-defined transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model's output. Match one or more candidate transformation schemes from pre-defined transformation cases based on the transformation demand information. Based on each candidate transformation scheme and the multimodal operational panorama dataset, determine the corresponding transformation assistance values ​​and generate corresponding transformation effectiveness evaluation results, which are then sent to the user terminal.

[0055] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system and device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0056] The systems, devices, and methods provided in this application are one-to-one correspondences. Therefore, the systems and devices also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and devices will not be repeated here.

[0057] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0058] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for matching enterprise digital transformation solutions based on a large model, characterized in that, The method includes: Obtain a multimodal operational panorama dataset from user terminals; wherein, the multimodal operational panorama dataset includes structured data and / or unstructured data generated by current enterprise operations; Based on the preset graph neural network and the first preset feature data in the multimodal operation panoramic dataset, the corresponding first feature vector is determined, and the multimodal operation panoramic dataset and the first feature vector are input into the pre-trained enterprise profile model to construct the corresponding enterprise profile vector of the enterprise to be transformed. The enterprise profile vector is input into a preset transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model output results. Based on the transformation demand information, match one or more candidate transformation schemes from the preset transformation cases, and based on each candidate transformation scheme and the multimodal operation panoramic dataset, determine the corresponding transformation assistance value and generate the corresponding transformation effectiveness evaluation result, which is then sent to the user terminal.

2. The enterprise digital transformation solution matching method based on a large model according to claim 1, characterized in that, Based on the preset graph neural network and the first preset feature data in the multimodal operation panoramic dataset, a corresponding first feature vector is determined, specifically including: According to preset feature data filtering rules, the first preset feature data is filtered from the multimodal operation panoramic dataset; wherein, the first preset feature data corresponds to the internal and external relationships of the enterprise to be transformed; the internal and external relationships include at least supplier relationships and customer relationships; Based on a preset graph structure modeling algorithm, the nodes in the first preset feature data and the relational attributes corresponding to each node are determined to construct a node attribute matrix. Interaction data between each node corresponding to the relational attributes is extracted from the first preset feature data to construct an adjacency matrix based on the interaction data. The node attribute matrix and the adjacency matrix are input into the preset graph neural network to extract the local and global relationship features of the internal and external relationships, and then concatenate them to obtain the first feature vector.

3. The enterprise digital transformation solution matching method based on a large model according to claim 2, characterized in that, The multimodal operation panoramic dataset and the first feature vector are input into a pre-trained enterprise profile model to construct the corresponding enterprise profile vector for the enterprise to be transformed, specifically including: The second feature vector corresponding to the structured modality, the third feature vector corresponding to the unstructured modality, and the first feature vector in the multimodal operation panoramic dataset are concatenated to obtain a fused feature matrix; By using a pre-defined multi-head attention mechanism, the modalities in the fusion feature matrix are fused to obtain the corresponding enterprise profile vector.

4. The enterprise digital transformation solution matching method based on a large model according to claim 1, characterized in that, Before inputting the enterprise profile vector into a preset transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model output, the method further includes: The industry knowledge graph corresponding to the enterprise to be transformed is embedded into the preset transformation demand model, and a preset graph attention network is added to the preset transformation demand model. By using a pre-defined contrastive learning strategy, pre-defined positive and negative sample pairs are added to the training samples to train the corresponding pre-defined transformation requirement large model, thereby performing knowledge augmentation processing on the large model.

5. The enterprise digital transformation solution matching method based on a large model according to claim 4, characterized in that, The enterprise profile vector is input into a preset transformation demand model to determine the transformation demand information of the enterprise to be transformed based on the model output, specifically including: The enterprise profile vector is input into the preset transformation demand model to determine the transformation demand labels and demand priorities output by the model. Based on the aforementioned priority of requirements, each of the aforementioned transformation requirement tags will be added sequentially to the corresponding transformation requirement sequence; Based on the transformation demand sequence, the multimodal operation panoramic dataset, and the enterprise profile vector, demand analysis is performed through a preset interpretable model to generate a demand analysis report to determine the key influencing factors of each of the transformation demand tags. Add the transformation demand sequence and the key influencing factors to the transformation demand information.

6. The enterprise digital transformation solution matching method based on a large model according to claim 1, characterized in that, Based on the transformation demand information, one or more candidate transformation solutions are matched from preset transformation cases, specifically including: The transformation requirement information is vector-encoded into a first matching vector; The first matching vector is calculated using a collaborative filtering algorithm to determine the matching similarity between the first matching vector and each second matching vector in the preset transformation cases. Based on the matching similarity and the preset time decay factor corresponding to each transformation case, one or more candidate transformation schemes are determined.

7. The enterprise digital transformation solution matching method based on a large model according to claim 6, characterized in that, The method further includes: Based on a pre-built reinforcement learning model, the collaborative filtering algorithm is updated and / or each of the candidate transformation schemes is modified. The method further includes: The updated data of the multimodal operation panoramic dataset is obtained in real time through a preset streaming computing framework. If the amount of updated data exceeds a preset threshold, the matching transformation scheme is updated.

8. The method for matching enterprise digital transformation solutions based on a large model according to claim 1, characterized in that, Based on the candidate transformation schemes and the multimodal operation panoramic dataset, the corresponding transformation assistance values ​​are determined and the corresponding transformation effectiveness evaluation results are generated, specifically including: The multimodal operation panoramic dataset and the implementation operation panoramic dataset corresponding to each of the candidate transformation schemes are input into a preset two-layer LSTM network to determine the predicted value of the indicator data corresponding to each transformation demand label. Based on the preset set of indicator weights and the predicted values ​​of each indicator data, the transformation assistance value of each indicator is calculated, and a radar chart is constructed with each transformation demand label as the axis label and the predicted values ​​of each indicator data of the candidate transformation scheme as the data points; wherein, the predicted values ​​of each indicator data of the same candidate transformation scheme in the radar chart constitute a data area. The transformation assistance values ​​and the radar chart are used as the transformation effectiveness evaluation results and sent to the user terminal.

9. A matching system for enterprise digital transformation solutions based on a large model, characterized in that, The system includes: The acquisition module is used to acquire a multimodal operational panorama dataset from the user terminal; wherein, the multimodal operational panorama dataset includes structured data and / or unstructured data generated by the current enterprise operation; The input determination module is used to determine the corresponding first feature vector based on the preset graph neural network and the first preset feature data in the multimodal operation panoramic dataset, and input the multimodal operation panoramic dataset and the first feature vector into the pre-trained enterprise profile model to construct the corresponding enterprise profile vector of the enterprise to be transformed. The input module is used to input the enterprise profile vector into a preset transformation demand model, so as to determine the transformation demand information of the enterprise to be transformed based on the model output results. The matching and determination module is used to match one or more candidate transformation schemes from preset transformation cases based on the transformation demand information, and to determine the corresponding transformation assistance values ​​and generate corresponding transformation effectiveness evaluation results based on each candidate transformation scheme and the multimodal operation panoramic dataset, and send them to the user terminal.

10. A matching device for enterprise digital transformation solutions based on a large model, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the enterprise digital transformation solution matching method based on a large model as described in any one of claims 1-8.

Images