Enterprise demand matching method and system based on big data model

Abstract

The present application discloses a method and system for matching enterprise needs based on a big data model, which relates to the field of big data analysis. The method includes: in response to receiving enterprise demand data, using a big data model to perform feature recognition on the enterprise demand data to obtain a semantic feature set, wherein the enterprise demand data includes image data and / or text data; using a pre-trained language model to convert the semantic feature set into an embedding vector set; calculating the similarity between the embedding vector set and the candidate solution set pre-stored in the electronic device; establishing an interactive attention mechanism between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set; when the similarity is greater than a preset threshold and the matching relationship is a match, outputting an enterprise demand matching solution corresponding to the embedding vector set. The present application can effectively improve the matching rate of enterprise needs.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of big data analysis, and in particular, to a method and system for matching enterprise needs based on a big data model. Background Art

[0002] In order to enhance corporate competitiveness, improve customer satisfaction, and promote cooperation and collaboration, it is very important to accurately capture the needs of corporate users. It can help corporate users achieve sustainable development in a rapidly changing market environment and is a key factor for companies to achieve efficient operations and strategic success.

[0003] Existing enterprise demand matching methods usually use predefined rules and keywords to match the needs of enterprise users. Among them, rules and keywords are manually defined and difficult to dynamically adjust after being predefined. However, enterprise needs are often diverse and complex. Manually defined rules and keywords need to cover all possible demand scenarios, which will cause the rules and keywords to become very complex. In addition, with the increase in needs and scenarios, the number and combination of rules and keywords will increase exponentially, making management and maintenance extremely difficult. Managing a large number of rules and keywords not only requires a lot of time and manpower, but is also prone to errors, resulting in a low enterprise demand matching rate. Therefore, there is an urgent need for an enterprise demand matching method to effectively solve the above problems. Summary of the invention

[0004] The embodiments of the present application provide a method and system for matching enterprise demands based on a big data model, which are used to effectively improve the matching rate of enterprise demands.

[0005] To achieve the above objectives, the embodiments of the present application adopt the following technical solutions:

[0006] In a first aspect, a method for matching enterprise requirements based on a big data model is provided, which is applied to an electronic device, wherein the electronic device is deployed with a big data model, and the method includes:

[0007] In response to receiving the enterprise demand data, the big data model is used to perform feature recognition on the enterprise demand data to obtain a semantic feature set, wherein the enterprise demand data includes image data and / or text data, and when the enterprise demand data is the image data, the semantic feature set is multiple, when the enterprise demand data is the text data, the semantic feature set is one, and when the enterprise demand data is the image data and the text data, the semantic feature set is multiple;

[0008] Using a pre-trained language model, converting the semantic feature set into an embedding vector set;

[0009] Calculating a similarity between the embedding vector set and a set of candidate solutions pre-stored in the electronic device;

[0010] Establishing an interactive attention mechanism between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set;

[0011] When the similarity is greater than a preset threshold and the matching relationship is a match, an enterprise demand matching solution corresponding to the embedded vector set is output.

[0012] In a possible implementation manner of the first aspect, when the enterprise demand data is the image data, the using the big data model to perform feature recognition on the enterprise demand data to obtain a semantic feature set includes:

[0013] In the case where the enterprise demand data is the image data, executing the first step to obtain a plurality of semantic feature sets;

[0014] Wherein, the first step comprises:

[0015] Using the big data model to perform image recognition on the image data to obtain a first semantic feature and a key image feature in the image data;

[0016] Converting the key image feature into a second semantic feature, and merging the second semantic feature with the first semantic feature to obtain a semantic feature set;

[0017] Calculate the first similarity between every two semantic features in the semantic feature set, and classify the semantic feature set according to the first similarity between every two semantic features to obtain multiple similar feature sets, wherein the similar feature set is used to characterize the semantic feature set, and the first similarity between one of the semantic features in each of the similar feature sets and any other semantic feature is greater than a first preset value.

[0018] In another possible implementation manner of the first aspect, when the enterprise demand data is the text data, the adopting the big data model to perform feature recognition on the enterprise demand data to obtain a semantic feature set includes:

[0019] In the case where the enterprise demand data is the text data, executing the second step to obtain a semantic feature set;

[0020] Wherein, the second step comprises:

[0021] The big data model is used to extract key semantic features of the text data to obtain a key feature set, and the key feature set is used to characterize the semantic feature set.

[0022] In another possible implementation manner of the first aspect, when the enterprise demand data is the text data and the image data, the using the big data model to perform feature recognition on the enterprise demand data to obtain a semantic feature set includes:

[0023] Execute the second step to obtain a key feature set;

[0024] Execute the first step to obtain a plurality of similar feature sets;

[0025] respectively calculating a second similarity between each key semantic feature in the key feature set and each of the similar feature sets;

[0026] When the second similarity is maximum, the correspondence between the key semantic feature and the similar feature set is determined, and the key semantic feature and the similar feature set are merged according to the correspondence to obtain an updated feature set, wherein the updated feature set is used to characterize the semantic feature set.

[0027] In another possible implementation manner of the first aspect, the method further includes:

[0028] If there are at least two corresponding relationships between the key semantic feature and the similar feature set when the second similarity is maximum, then executing the third step to obtain a plurality of the semantic feature sets, and there are at least two corresponding relationships between the key semantic feature and the similar feature set for representing that when the second similarity is maximum, there are at least two similar feature sets corresponding to the key semantic feature;

[0029] Wherein, the third step comprises:

[0030] Taking each of the similar feature sets corresponding to the maximum second similarity as a grid, and arranging each of the grids in sequence according to the first order;

[0031] The spatial distance between the key semantic feature and each of the grids is calculated respectively, and the similar feature set corresponding to the grid with the shortest spatial distance is used as the target feature set, and the key semantic feature is merged with the target feature set to obtain an updated feature set, wherein the updated feature set is used to characterize the semantic feature set.

[0032] In another possible implementation manner of the first aspect, converting the key image feature into a second semantic feature, and merging the second semantic feature with the first semantic feature to obtain a semantic feature set includes:

[0033] Performing dimensionality reduction on the key image features to obtain a low-dimensional feature vector;

[0034] Inputting the low-dimensional feature vector into a semantic mapping model to generate a corresponding second semantic feature;

[0035] The first semantic feature and the second semantic feature are concatenated to obtain a semantic feature set.

[0036] In another possible implementation manner of the first aspect, the spatial distance is represented by Euclidean distance, and respectively calculating the spatial distance between the key semantic feature and each of the grids includes:

[0037] For each of the grids, a Euclidean distance calculation formula is used to calculate the distance between the key semantic feature and each semantic feature in the grid, and the minimum value of all the distances is calculated as the distance between the key semantic feature and the grid;

[0038] The Euclidean distance calculation formula includes:

[0039]

[0040] Where, is the distance between the key semantic feature k and the grid G, k is the key semantic feature, s is one of the semantic features in the grid, G is a grid, k i With s i are the i-th components of k and s respectively.

[0041] In another possible implementation manner of the first aspect, the electronic device is further deployed with a Transformer model, and establishing an interactive attention mechanism between the embedding vector set and the candidate solution set to capture a matching relationship between the embedding vector set and the candidate solution set includes:

[0042] By using the Transformer model, constructing an interactive attention mechanism between the embedding vector set and the candidate solution set, and using the interactive attention mechanism to calculate an attention score between the embedding vector set and the candidate solution set;

[0043] Calculating an attention weight according to the attention score;

[0044] The attention weights are multiplied by a value matrix generated according to the candidate solutions, and the obtained attention output is used to capture the matching relationship between the embedding vector set and the candidate solution set.

[0045] In another possible implementation of the first aspect, the big data model is a Transformer-based multimodal big data model, the framework of the Transformer-based multimodal big data model is a Transformer model, and the structure of the Transformer-based multimodal big data model includes an embedding module, a graph neural network module, a modal fusion module, and an output layer;

[0046] Among them, the embedding module is used to convert the semantic feature set into an embedding vector set, the graph neural network module is used to perform image recognition on the image data, the modal fusion module is used to establish an interactive attention mechanism between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set, and the output layer is used to output the enterprise demand matching solution corresponding to the embedding vector set.

[0047] In a second aspect, the present application provides an enterprise demand matching system based on a big data model, comprising:

[0048] a memory configured to store instructions; and

[0049] The processor is configured to call the instructions from the memory and implement the above-mentioned enterprise demand matching method based on the big data model when executing the instructions.

[0050] Through the above technical solution, a big data model is used to identify the features of enterprise demand data without relying on manually defined rules and keywords. In addition, a pre-trained language model is used to convert the semantic feature set into an embedding vector set, thereby simplifying the complexity of demand representation and avoiding the exponential growth problem of rules and keywords. By calculating the similarity between the embedding vector set and the candidate solution set, and establishing an interactive attention mechanism, the matching relationship between the embedding vector set and the candidate solution set is accurately captured, which effectively improves the matching rate of enterprise needs. When the similarity is greater than the preset threshold and the matching relationship is a match, the enterprise demand matching solution corresponding to the embedding vector set is automatically output, which reduces human intervention and errors and improves efficiency.

[0051] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1A flowchart of an enterprise demand matching method based on a big data model provided in an embodiment of the present application;

[0053] Figure 2 A structural block diagram of a big data model provided in an embodiment of the present application;

[0054] Figure 3 A structural diagram of an enterprise demand matching system based on a big data model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the specific implementation methods described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0056] It should be noted that if the embodiments of the present application involve directional indications (such as up, down, left, right, front, back...), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0057] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0058] Figure 1 The following schematically shows a flow chart of a method for matching enterprise needs based on a big data model according to an embodiment of the present application. Figure 1 As shown, an embodiment of the present application provides an enterprise demand matching method based on a big data model, which is applied to an electronic device, and the electronic device is deployed with a big data model. The method may include the following steps.

[0059] S110, in response to receiving the enterprise demand data, using a big data model to perform feature recognition on the enterprise demand data to obtain a semantic feature set, wherein the enterprise demand data includes image data and / or text data, and when the enterprise demand data is image data, there are multiple semantic feature sets, when the enterprise demand data is text data, there is one semantic feature set, and when the enterprise demand data is image data and text data, there are multiple semantic feature sets;

[0060] S120, using a pre-trained language model, converting the semantic feature set into an embedding vector set;

[0061] S130, calculating the similarity between the embedding vector set and the candidate solution set pre-stored in the electronic device;

[0062] S140, establishing an interactive attention mechanism between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set;

[0063] S150: When the similarity is greater than a preset threshold and the matching relationship is a match, output an enterprise demand matching solution corresponding to the embedded vector set.

[0064] In response to receiving the enterprise demand data, a big data model is used to perform feature recognition on the enterprise demand data to obtain a set of semantic features. The enterprise demand data may include image data and / or text data. In the case where the enterprise demand data is image data, since the image data contains rich visual information, such as color, shape, texture, etc., it can be extracted and analyzed through computer vision technology. For example, if the enterprise demand data is an image containing multiple electronic devices, the neural network model can be used to identify the different devices in the image and extract the features of each device, such as brand, model, function, etc.

[0065] If the enterprise demand data is text data, since text data mainly contains language information, the semantic feature set is usually one. For example, if the enterprise demand data is a text describing the demand for electronic equipment, the key semantic features in the text data, such as device type, specification requirements, usage scenarios, etc., can be extracted through pre-trained language models (such as BERT, GPT, etc.).

[0066] If the enterprise demand data contains both image data and text data, there are multiple semantic feature sets, and the features in the semantic feature sets are extracted from the image and text respectively to obtain a more comprehensive and accurate demand description.

[0067] A pre-trained language model is used to convert a set of semantic features into a set of embedded vectors. A pre-trained language model is a model that has been trained with a large amount of data and can convert text and semantic features into vector representations with fixed dimensions. The vector can capture the relationship and similarity between semantic features.

[0068] In the specific implementation process, the semantic features need to be preprocessed first, such as word segmentation, stop word removal, etc., and then the processed features are input into the pre-trained language model to obtain the corresponding embedding vector set. Each vector in the embedding vector set represents a semantic feature, and the embedding vector represents the similarity and correlation between features through the position relationship in the high-dimensional space.

[0069] The similarity between the embedding vector set and the candidate solution set pre-stored in the electronic device is calculated. By calculating the similarity between the embedding vector set and the vectors in the candidate solution set, the matching degree between the enterprise demand and the candidate solution can be determined.

[0070] The similarity calculation method can be cosine similarity, Euclidean distance, Manhattan distance, etc. In the specific implementation process, each solution in the candidate solution set needs to be converted into an embedding vector first, and then the similarity between the embedding vectors in the embedding vector set and the candidate solution vectors is calculated one by one, and a similarity matrix can be obtained. Each element in the matrix represents the similarity between a vector in the embedding vector set and the candidate solution vector. By analyzing the similarity matrix, candidate solutions with high similarity to enterprise needs can be preliminarily screened out.

[0071] An interactive attention mechanism is established between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set. The interactive attention mechanism is a technology used to capture the relationship between different vector sets. By calculating the attention weights between vectors, important matching relationships can be identified. For example, the multi-head attention mechanism in the Transformer model can be used to perform multiple attention calculations on the embedding vector set and the candidate solution set, and each calculation results in an attention weight matrix. By combining multiple attention weight matrices, a more accurate matching relationship can be obtained. In the specific implementation process, the embedding vector set and the candidate solution set need to be input into the multi-head attention mechanism respectively, the attention weights between each vector are calculated, and then the attention weights are combined with the similarity matrix to obtain a weighted similarity matrix. By analyzing the weighted similarity matrix, the important matching relationship between the embedding vector set and the candidate solution set can be identified. In summary, the matching relationship between enterprise needs and candidate solutions can be captured more accurately, providing a basis for the final matching results.

[0072] When the similarity is greater than the preset threshold and the matching relationship is a match, the enterprise demand matching solution corresponding to the embedded vector set is output. The preset threshold is a standard for screening matching results with higher similarity. By setting a reasonable threshold, matching results with lower similarity can be filtered out to ensure that the output matching solution has higher accuracy and relevance.

[0073] In the specific implementation process, it is first necessary to screen out the matching results with similarities greater than the preset threshold according to the similarity matrix and the weighted similarity matrix, and then further verify the matching relationship of the matching results, and determine whether the matching relationship is a valid match by analyzing the attention weight in the interactive attention mechanism. If the matching relationship is a valid match, the corresponding enterprise demand matching solution is output. If the matching relationship is invalid, other matching results will continue to be screened until a valid enterprise demand matching solution is found or there are no more matching results. In this way, it can be ensured that the output enterprise demand matching solution has high accuracy and relevance to meet the needs of the enterprise.

[0074] This embodiment can achieve precise matching of enterprise demand data and provide efficient and accurate solutions through steps such as feature recognition, embedding vector conversion, similarity calculation, establishment of interactive attention mechanism and matching solution output. It can not only process image data and text data, but also comprehensively analyze multiple data types, effectively improving the accuracy and efficiency of enterprise demand matching.

[0075] In one implementation of this embodiment, when the enterprise demand data is image data, a big data model is used to perform feature recognition on the enterprise demand data to obtain a semantic feature set, including the following steps:

[0076] S210, when the enterprise demand data is image data, executing the first step to obtain multiple semantic feature sets;

[0077] The first step includes:

[0078] S220, performing image recognition on the image data using a big data model to obtain a first semantic feature and a key image feature in the image data;

[0079] S230, converting the key image feature into a second semantic feature, and merging the second semantic feature with the first semantic feature to obtain a semantic feature set;

[0080] S240, calculating the first similarity between every two semantic features in the semantic feature set, and classifying the semantic feature set according to the first similarity between every two semantic features to obtain multiple similar feature sets, wherein the similar feature set is used to characterize the semantic feature set, and the first similarity between one semantic feature in each similar feature set and any other semantic feature is greater than a first preset value.

[0081] When the enterprise demand data is image data, a big data model is used to perform feature recognition on the image data to obtain a set of semantic features. First, the first step is performed to obtain multiple sets of semantic features. Specifically, it includes performing image recognition on the image data and extracting the first semantic features and key image features in the image. Image recognition usually uses a deep learning model, which can identify different devices in the image and the text data in the image, and extract the first semantic features and key image features of each device, such as brand, model, color, etc.

[0082] The key image features are converted into second semantic features, and the second semantic features are merged with the first semantic features to obtain a semantic feature set. Key image features refer to the most representative parts of an image, such as appearance features, logos, etc. The key image features can be converted into higher-level semantic features through a preset conversion algorithm. For example, low-level features such as color and shape in an image can be converted into vectors representing high-level features such as device type and function, namely, second semantic features. Then, the second semantic features are merged with the first semantic features extracted previously to form a complete semantic feature set.

[0083] The first similarity between every two semantic features in the semantic feature set is calculated, and the semantic feature set is classified according to the first similarity between every two semantic features to obtain multiple similar feature sets. Similarity calculation is achieved by comparing the distance or similarity between two feature vectors. In this embodiment, the similarity is calculated by cosine similarity. For example, cosine similarity represents the degree of similarity by calculating the cosine value of the angle between two vectors. The closer the cosine value is to 1, the more similar the two vectors are.

[0084] In the specific implementation process, each feature vector in the semantic feature set needs to be standardized first, and then the similarity between each two feature vectors is calculated one by one to obtain a similarity matrix. By analyzing the similarity matrix, similar features can be preliminarily determined.

[0085] According to the data in the similarity matrix, the semantic feature set is classified to obtain multiple similar feature sets. The purpose of classification is to classify features with high similarity into one category to form a similar feature set.

[0086] In the specific implementation process, a clustering algorithm can be used to cluster features with higher similarity based on the similarity between feature vectors. For example, when using the K-means clustering algorithm, it is first necessary to determine the number of clusters K, and then randomly select K initial cluster centers, assign each feature vector to the cluster center that is most similar to it, calculate the center point of each cluster, and iterate and update until the clustering result is stable. In this way, multiple similar feature sets can be obtained, and the similarity between the features in each similar feature set is greater than the first preset value. Similar feature sets can effectively represent different categories in the semantic feature set.

[0087] In the case of image data, this implementation can use a big data model to perform feature recognition on enterprise demand data to obtain multiple semantic feature sets. Specifically, through steps such as image recognition, feature conversion, similarity calculation and feature classification, it is possible to effectively extract and classify important features in the image to form an accurate semantic feature set. The semantic feature set can accurately represent enterprise needs, facilitate subsequent demand matching, and effectively improve the accuracy and efficiency of demand matching.

[0088] In one implementation of this embodiment, when the enterprise demand data is text data, a big data model is used to perform feature recognition on the enterprise demand data to obtain a semantic feature set, including the following steps:

[0089] S310, when the enterprise demand data is text data, executing the second step to obtain a semantic feature set;

[0090] The second step includes:

[0091] S320. Use a big data model to extract key semantic features of the text data to obtain a key feature set, which is used to represent the semantic feature set.

[0092] In the case where the enterprise demand data is text data, a big data model is used to perform feature recognition on the text data to obtain a semantic feature set. First, the second step is performed to obtain a semantic feature set. Specifically, it includes extracting key semantic features of the text data using a big data model to obtain a key feature set, and the key feature set is used to represent the semantic feature set.

[0093] In the process of text data processing, the text data needs to be preprocessed first. Preprocessing includes steps such as text cleaning, word segmentation, stop word removal and stem extraction.

[0094] Text cleaning refers to removing noise data in the text, such as punctuation marks, special characters, etc. Word segmentation is to split the text into individual words. Stop-word removal means removing high-frequency words that do not contribute much to the semantics of the text, such as "de", "shi", "zai", etc. Stemming is to reduce the diversity of words by reducing them to their root forms. For example, when processing text describing enterprise requirements, the preprocessing step cleans the text to contain only meaningful words and performs stemming on these words to form a list of words.

[0095] Next, a big data model is used to extract the key semantic features of the text data. The words in the text data can be converted into high-dimensional vectors to capture the semantic relationships between the words. Specifically, the big data model can convert each word into a vector, and the vector represents the semantic information of the word in the high-dimensional space. By aggregating the vectors, an overall semantic representation of the text can be obtained. For more complex text data, the big data model can be a BERT model, which captures context information through bidirectional encoding of the text and extracts more accurate semantic features.

[0096] In the specific implementation process, first, the preprocessed words are input into the big data model to obtain the vector representation of each word, and then these vectors are aggregated to obtain the key semantic feature vector of the text.

[0097] The extracted key semantic feature vectors form a key feature set. The key feature set is used to represent the semantic features of the entire text data. The key feature set includes not only the semantic features of individual words but also the overall semantic representation of the text. For example, when processing a text describing an enterprise's demand for a certain product, the key feature set will include all the important features related to the product, such as product type, specifications, uses, etc. The important features are represented by vectors and can accurately reflect the semantic information of the text in the high-dimensional space.

[0098] Finally, by analyzing the key feature set, a semantic feature set of the text data can be obtained. The semantic feature set is a comprehensive representation of all the important information in the text data and can accurately reflect the core content of the enterprise's requirements.

[0099] In the above way, complex text data can be transformed into a structured semantic feature set, providing a basis for subsequent demand matching and analysis. For example, in enterprise demand analysis, the semantic feature set can be used to match similar demands, find the historical demand records most similar to the current demand, and thus provide a more accurate solution.

[0100] In the case of text data, this implementation can use a big data model to perform feature recognition on enterprise demand data to obtain a semantic feature set. Specifically, through the steps of text preprocessing, semantic feature extraction, and feature set formation, it is possible to effectively extract and represent important information in the text and form an accurate semantic feature set. The semantic feature set can accurately represent enterprise needs, facilitate subsequent demand matching and analysis, and effectively improve the accuracy and efficiency of demand matching.

[0101] In one implementation of this embodiment, when the enterprise demand data is text data and image data, a big data model is used to perform feature recognition on the enterprise demand data to obtain a semantic feature set, including the following steps:

[0102] S410, executing the second step to obtain a key feature set;

[0103] S420, executing the first step to obtain multiple similar feature sets;

[0104] S430, respectively calculating the second similarity between each key semantic feature in the key feature set and each similar feature set;

[0105] S440. When the second similarity is maximum, determine the correspondence between the key semantic feature and the similar feature set, and merge the key semantic feature with the similar feature set according to the correspondence to obtain an updated feature set, wherein the updated feature set is used to represent the semantic feature set.

[0106] When the enterprise demand data is text data and image data, the steps of using the big data model to perform feature recognition on the enterprise demand data to obtain a semantic feature set are as follows:

[0107] First, execute the second step to obtain a key feature set. For detailed instructions on the second step, please refer to S310-S320, and this application does not limit this.

[0108] Secondly, execute the first step to obtain multiple similar feature sets. Specifically, it includes performing image recognition on the image data to extract the first semantic feature and the key image feature in the image. Next, the key image feature is converted into a second semantic feature, and the second semantic feature is merged with the first semantic feature to obtain a semantic feature set. Then, the first similarity between every two semantic features in the semantic feature set is calculated, and the semantic feature set is classified according to the first similarity between every two semantic features to obtain multiple similar feature sets. For the relevant description of the first step, please see S210-S240, which will not be repeated in this application.

[0109] In addition, the second similarity between each key semantic feature in the key feature set and each similar feature set is calculated respectively. The second similarity calculation is achieved by comparing the similarity between the feature vector in the key feature set and the feature vector in the similar feature set.

[0110] The second similarity of this embodiment can be obtained by calculating the cosine similarity. Specifically, the second similarity is represented by calculating the cosine value of the angle between the two vectors. The closer the cosine value is to 1, the more similar the two vectors are. In the specific implementation process, it is first necessary to standardize each feature vector in the key feature set and the similar feature set, and then calculate the similarity between each key feature vector and each similar feature vector one by one to obtain a similarity matrix. By analyzing the similarity matrix, similar features can be preliminarily judged.

[0111] Next, when the second similarity is the largest, the correspondence between the key semantic features and the similar feature set is determined, and the key semantic features and the similar feature set are merged according to the correspondence to obtain an updated feature set. Feature merging is used to classify features with high similarity into one category to form an updated feature set.

[0112] In the specific implementation process, a clustering algorithm can be used to cluster features with high similarity based on the similarity between feature vectors. For example, when using the K-means clustering algorithm, it is necessary to first determine the number of clusters K, then randomly select K initial cluster centers, assign each feature vector to the cluster center that is most similar to it, calculate the center point of each cluster, and iterate and update until the clustering result is stable.

[0113] By the above method, an updated feature set can be obtained, and the similarity between features in each updated feature set is greater than the second preset value. The updated feature set can effectively represent different categories in the semantic feature set, providing accurate feature classification for subsequent demand matching.

[0114] In the case of text data and image data, this implementation can use a big data model to perform feature recognition on enterprise demand data to obtain a semantic feature set. Specifically, through steps such as text preprocessing, semantic feature extraction, image recognition, feature conversion, similarity calculation and feature merging, it can effectively extract and represent important information in text and images to form an accurate semantic feature set. The semantic feature set can accurately represent enterprise needs, which is conducive to subsequent demand matching and analysis, and can effectively improve the accuracy and efficiency of demand matching.

[0115] In one implementation of this embodiment, the following steps are also included:

[0116] S510, if there are at least two corresponding relationships between the key semantic feature and the similar feature set when the second similarity is maximum, then execute the third step to obtain multiple semantic feature sets, and there are at least two corresponding relationships between the key semantic feature and the similar feature set for representing that when the second similarity is maximum, there are at least two similar feature sets corresponding to the key semantic feature;

[0117] The third step includes:

[0118] S520, taking each similar feature set corresponding to the second maximum similarity as a grid, and arranging each grid in sequence according to the first order;

[0119] S530, respectively calculating the spatial distance between the key semantic feature and each grid, and taking the similar feature set corresponding to the grid with the shortest spatial distance as the target feature set, merging the key semantic feature with the target feature set to obtain an updated feature set, wherein the updated feature set is used to characterize the semantic feature set.

[0120] In the process of processing enterprise demand data, if there are at least two corresponding relationships between the key semantic feature and the similar feature set when calculating the second similarity, it is necessary to perform the third step to obtain multiple semantic feature sets. In this case, there are at least two corresponding relationships between the key semantic feature and the similar feature set, which is used to represent that when the second similarity is maximum, there are at least two similar feature sets corresponding to the key semantic feature.

[0121] First, each similar feature set corresponding to the maximum second similarity is taken as a grid, and each grid is arranged in sequence according to the first order. Specifically, assuming that there are three similar feature sets with the maximum second similarity to a key semantic feature, then these three similar feature sets will be respectively regarded as three independent grids. The first order can be based on the order of the timestamps of the similar feature sets, so that the grids can be processed in an orderly manner, thereby ensuring the accuracy and consistency of feature merging.

[0122] Next, the spatial distance between the key semantic feature and each grid is calculated respectively, and the similar feature set corresponding to the grid with the shortest spatial distance is used as the target feature set. In this embodiment, the Euclidean distance is used to represent the spatial distance.

[0123] By calculating the spatial distance between the key semantic feature and each grid, the grid with the shortest distance can be found, indicating that the similar feature set of the grid is closest to the key semantic feature in the feature space. The similar feature set corresponding to the grid is used as the target feature set to ensure the accuracy of feature merging.

[0124] Finally, the key semantic features are merged with the target feature set to obtain an updated feature set. Specifically, the key semantic features can be merged with the target feature set by putting the key semantic features into the target feature set to obtain a merged feature set, i.e., an updated feature set. The updated feature set can more accurately reflect the characteristics of the enterprise demand data.

[0125] It should be noted that the spatial distance in this embodiment is unique, that is, in practical applications, it is extremely rare for the spatial distance to be completely consistent. The reasons are as follows: in high-dimensional space, if the distance between data points is measured using the Euclidean distance (Euclidean distance), since the Euclidean distance has a stronger ability to distinguish in high-dimensional space, compared with low-dimensional space, the probability of the spatial distance in high-dimensional space being completely consistent is almost zero; in this embodiment, the grid division is fine-grained, which further ensures the uniqueness of the spatial distance.

[0126] This implementation can determine the optimal feature merging method when there are multiple corresponding relationships between key semantic features and similar feature sets. First, by treating each similar feature set as a grid and arranging them in order, the systematic and orderly nature of the processing is ensured. Secondly, by calculating the spatial distance between the key semantic feature and each grid, the target feature set closest to the key semantic feature is found, ensuring the accuracy of feature merging. Finally, through feature merging, an updated feature set is obtained, which can more accurately characterize the characteristics of enterprise demand data. The accuracy of feature recognition is effectively improved, which is conducive to subsequent demand matching and analysis.

[0127] In one implementation of this embodiment, converting the key image feature into a second semantic feature, and merging the second semantic feature with the first semantic feature to obtain a semantic feature set includes the following steps:

[0128] S610, reducing the dimension of key image features to obtain a low-dimensional feature vector;

[0129] S620, inputting the low-dimensional feature vector into a semantic mapping model to generate a corresponding second semantic feature;

[0130] S630: Concatenate the first semantic feature and the second semantic feature to obtain a semantic feature set.

[0131] Reduce the dimension of key image features to obtain low-dimensional feature vectors. Images usually have high-dimensional feature space. For example, a 224x224 pixel RGB image has a feature dimension of 224x224x3, or 150,528 dimensions. High-dimensional image features bring great computational burden and storage pressure when processing and analyzing, so dimensionality reduction is required. Principal component analysis (PCA) can be used for dimensionality reduction. For example, PCA can reduce the dimension by finding the principal components of the data while trying to retain the main information of the data. Assuming that the original feature matrix is ​​X, its dimension is n×d (where n is the number of samples and d is the feature dimension), it can be converted into a low-dimensional feature matrix Y with a dimension of n×k (where k is the feature dimension after dimensionality reduction, usually k<<d) through PCA. The feature vector after dimensionality reduction not only reduces the computational complexity, but also removes redundant information and retains the most important feature information.

[0132] Next, the low-dimensional feature vector is input into the semantic mapping model to generate the corresponding second semantic feature. The semantic mapping model is a pre-trained model that can map the low-dimensional feature vector into the semantic space to generate a feature representation with semantic meaning. Common semantic mapping models include convolutional neural networks (CNN), recurrent neural networks (RNN), etc. Taking CNN as an example, the low-dimensional feature vector is processed by several convolutional layers, pooling layers, and fully connected layers, and finally outputs a semantic feature vector. Assuming that the input low-dimensional feature vector is z, after being processed by the semantic mapping model f(·), the generated second semantic feature is s2=f(z). Through the layer-by-layer abstraction of the deep learning model, high-level features with semantic meaning are extracted. The generated second semantic feature not only contains the key information of the original image, but also has been enhanced at the semantic level.

[0133] Finally, the first semantic feature is concatenated with the second semantic feature to obtain a semantic feature set. Assuming that the first semantic feature vector is s1 and the second semantic feature vector is s2, the concatenation operation can connect the two vectors to obtain a new feature vector s = [s1; s2].

[0134] Specifically, we can use vector concatenation, that is, arranging the elements of two vectors in sequence to form a new longer vector. The concatenated semantic feature set not only contains the semantic information of the original image, but also integrates the enhanced features after dimensionality reduction and semantic mapping, which can more comprehensively characterize the features of the image.

[0135] This implementation can convert high-dimensional key image features into low-dimensional semantic features, and effectively integrate them with existing semantic features to obtain a more comprehensive and accurate set of semantic features. First, through dimensionality reduction processing, the feature dimension is reduced, the computational complexity is reduced, and the most important feature information is retained. Secondly, through the semantic mapping model, the corresponding second semantic features are generated, and through feature splicing, the first semantic features from different sources are spliced ​​with the second semantic features, which effectively improves the richness of feature representation.

[0136] In one implementation of this embodiment, the spatial distance is represented by the Euclidean distance, and the spatial distance between the key semantic feature and each grid is calculated respectively, including the following steps:

[0137] S710, for each grid, using the Euclidean distance calculation formula to calculate the distance between the key semantic feature and each semantic feature in the grid, and calculating the minimum value of all distances as the distance between the key semantic feature and the grid;

[0138] Among them, the Euclidean distance calculation formula includes:

[0139]

[0140] Where, is the distance between the key semantic feature k and the grid G, k is the key semantic feature, s is one of the semantic features in the grid, G is a grid, k i With s i are the i-th components of k and s respectively.

[0141] When calculating the spatial distance between the key semantic feature and each grid, the Euclidean distance is used to represent the distance, and the distance between the key semantic feature and each semantic feature in the grid is calculated separately. Finally, the minimum value of all distances is calculated as the distance between the key semantic feature and the grid. The similarity between the key semantic feature and the semantic feature in each grid can be quantified. The smaller the distance, the more similar the two features are.

[0142] This embodiment selects the minimum distance as the distance between the key semantic feature and the grid, which can ensure that the semantic feature closest to the key semantic feature is found among multiple semantic features, thereby more accurately representing the similarity between the key semantic feature and the entire grid.

[0143] In a specific implementation, a loop and conditional judgment may be used to calculate the distance between the key semantic features and the grid one by one and find the minimum value.

[0144] This implementation can accurately calculate the spatial distance between the key semantic feature and each grid, and select the minimum distance as the distance between the key semantic feature and the grid. First, the similarity between the key semantic feature and each semantic feature in the grid is quantified by the Euclidean distance formula. Secondly, by calculating the minimum distance, it is ensured that the semantic feature closest to the key semantic feature is found among multiple semantic features, thereby more accurately representing the similarity between the key semantic feature and the entire grid, effectively improving the accuracy of feature matching.

[0145] In one implementation of this embodiment, the electronic device is further deployed with a Transformer model, and an interactive attention mechanism is established between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set, including the following steps:

[0146] S810, constructing an interactive attention mechanism between the embedding vector set and the candidate solution set through the Transformer model, and using the interactive attention mechanism to calculate the attention score between the embedding vector set and the candidate solution set;

[0147] S810, calculating the attention weight according to the attention score;

[0148] S810, multiplying the attention weight by the value matrix generated according to the candidate solution, and the obtained attention output is used to capture the matching relationship between the embedding vector set and the candidate solution set.

[0149] Through the Transformer model, an interactive attention mechanism is constructed between the embedding vector set and the candidate solution set. The Transformer model is a deep learning model based on the attention mechanism, which is used to capture the relationship between input data through the attention mechanism. In this embodiment, it is assumed that the embedding vector set is represented by E = {e1, e2, …, em}, and the candidate solution set is represented by C = {c1, c2, …, cn}. Through the interactive attention mechanism, the attention score between each vector in the embedding vector set and each vector in the candidate solution set can be calculated. Specifically, the interactive attention mechanism obtains the attention score by calculating the dot product between the query vector, the key vector, and the value vector. Assuming that the query vector is Q, the key vector is K, and the value vector is V, the calculation formula of the attention score is:

[0150]

[0151] Among them, d k is the dimension of the key vector, and the softmax function is used to normalize the scores into a probability distribution.

[0152] Next, the attention weights are calculated based on the attention scores. The attention score matrix represents the similarity between each vector in the embedding vector set and each vector in the candidate solution set. In order to obtain the attention weights between each embedding vector and the candidate solution, the attention scores need to be normalized. Specifically, the attention scores can be converted to weights through the softmax function. Assuming that the attention score matrix is ​​A, the calculation formula of the attention weight matrix W is:

[0153] W = softmax(A);

[0154] Among them, the softmax function converts the attention score of each row into a probability distribution so that the sum of the weights of each row is 1. In this way, the attention weight of each embedded vector in the candidate solution set can be obtained, which quantifies the degree of match between the embedded vector and the candidate solution. The larger the weight, the higher the similarity between the two vectors.

[0155] Finally, the attention weights are multiplied by the candidate solutions, and the obtained attention output is used to capture the matching relationship between the embedding vector set and the candidate solution set.

[0156] It should be noted that, in this embodiment, the query matrix Q is generated by the embedding vector set; the key matrix K is generated by the candidate solution set; and the value matrix V is generated by the candidate solution set.

[0157] The specific formula is: Q = EW Q ; K = CW K ; V = CW V .

[0158] Among them, W Q , W K , W V is a learnable weight matrix.

[0159] This implementation first calculates the similarity between the embedding vector and the candidate solution through the interactive attention mechanism of the Transformer model. Secondly, the attention score is converted into a weight through the softmax function to quantify the degree of match between the embedding vector and the candidate solution. Finally, the attention output is obtained, which can capture the matching relationship between the embedding vector and the candidate solution. In summary, an interactive attention mechanism can be effectively established between the embedding vector set and the candidate solution set, and the matching relationship between the two can be accurately captured through the attention score, attention weight and attention output, thereby improving the accuracy of feature matching.

[0160] In one implementation of this embodiment, refer to Figure 2, the big data model is a Transformer-based multimodal big data model, the framework of the Transformer-based multimodal big data model is a Transformer model, and the structure of the Transformer-based multimodal big data model includes an embedding module, a graph neural network module, a modal fusion module and an output layer;

[0161] Among them, the embedding module is used to convert the semantic feature set into an embedding vector set, the graph neural network module is used to perform image recognition on image data, the modal fusion module is used to establish an interactive attention mechanism between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set, and the output layer is used to output the enterprise demand matching solution corresponding to the embedding vector set.

[0162] Specifically, in the Transformer-based multimodal big data model, the model framework is composed of the Transformer model, and its structure includes an input layer, an embedding module, a graph neural network module, a modal fusion module, and an output layer. This structural design is designed to effectively process and fuse multimodal data to achieve efficient matching of enterprise needs. The following is a detailed description of each module:

[0163] First, the embedding module is responsible for converting the semantic feature set into an embedding vector set. The semantic feature set can include text data, label information, etc. The data needs to be converted into an embedding vector that the model can process. The embedding module usually uses word embedding technology to convert text data into a high-dimensional vector representation. The embedding vector retains the semantic information of the original data, allowing the model to understand and process the text data. For example, suppose there is a set of enterprise demand description texts. Through the embedding module, these texts can be converted into an embedding vector set with a dimension of 768.

[0164] Next, the graph neural network module is used to perform image recognition on image data. Graph neural network (GNN) is a neural network model used to process graph structured data, which can capture the relationship between nodes and the global structure of the graph. In the multimodal big data model, GNN can be used to process image data and convert image features into embedded vectors. Assuming there is a set of image data, after extracting image features through convolutional neural network (CNN), the image features can be input into GNN for processing.

[0165] Then, the modal fusion module is used to establish an interactive attention mechanism between the embedding vector set and the candidate solution set to capture the matching relationship between the two. The core of the modal fusion module is the multi-head attention mechanism of the Transformer model, which captures the similarity and matching relationship between the two by calculating the attention score between the embedding vector set and the candidate solution set. Specifically, assuming there is an embedding vector set and a candidate solution set, the attention score matrix between the two can be calculated through the multi-head attention mechanism, and converted into attention weights to finally obtain the attention output.

[0166] Finally, the output layer is used to output the enterprise demand matching solution corresponding to the embedded vector set. After being processed by the modality fusion module, the obtained attention output represents the matching relationship between the embedded vector set and the candidate solution set. The output layer converts the matching relationship into a specific enterprise demand matching solution. Specifically, the attention output can be mapped to the final output space through the fully connected layer to obtain a solution that matches the enterprise demand.

[0167] In this embodiment, the Transformer-based multimodal big data model realizes efficient processing and fusion of multimodal data through the collaborative work of the embedding module, the graph neural network module, the modal fusion module and the output layer. The embedding module converts the semantic feature set into an embedding vector set, so that the model can understand and process text data. The graph neural network module performs image recognition on image data and converts image features into embedding vectors. The modal fusion module captures the matching relationship between the embedding vector set and the candidate solution set through the interactive attention mechanism. The output layer converts these matching relationships into specific enterprise demand matching solutions. In the above manner, enterprise needs and solutions can be accurately matched, the accuracy and efficiency of feature matching are improved, and intelligent solution recommendation services are provided for enterprises.

[0168] The embodiment of the present application also provides an enterprise demand matching system based on a big data model, including:

[0169] a memory configured to store instructions; and

[0170] The processor is configured to call instructions from the memory and implement the above-mentioned enterprise demand matching method based on the big data model when executing the instructions.

[0171] An embodiment of the present application also provides a machine-readable storage medium, on which instructions are stored, and the instructions are used to enable a machine to execute the above-mentioned enterprise demand matching method based on a big data model.

[0172] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0173] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0174] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0175] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0176] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0177] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0178] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0179] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0180] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for matching enterprise needs based on a big data model, characterized in that: Applied to an electronic device, the electronic device is deployed with a big data model, and the method includes: In response to receiving the enterprise demand data, the big data model is used to perform feature recognition on the enterprise demand data to obtain a semantic feature set, wherein the enterprise demand data includes image data and / or text data, and when the enterprise demand data is the image data, the semantic feature set is multiple, when the enterprise demand data is the text data, the semantic feature set is one, and when the enterprise demand data is the image data and the text data, the semantic feature set is multiple; Using a pre-trained language model, converting the semantic feature set into an embedding vector set; Calculating a similarity between the embedding vector set and a set of candidate solutions pre-stored in the electronic device; Establishing an interactive attention mechanism between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set; When the similarity is greater than a preset threshold and the matching relationship is a match, outputting an enterprise demand matching solution corresponding to the embedded vector set; In the case where the enterprise demand data is the image data, the adopting the big data model to perform feature recognition on the enterprise demand data to obtain a semantic feature set includes: In the case where the enterprise demand data is the image data, executing the first step to obtain a plurality of semantic feature sets; Wherein, the first step comprises: Using the big data model to perform image recognition on the image data to obtain a first semantic feature and a key image feature in the image data; Converting the key image feature into a second semantic feature, and merging the second semantic feature with the first semantic feature to obtain a semantic feature set; Calculate the first similarity between every two semantic features in the semantic feature set, and classify the semantic feature set according to the first similarity between every two semantic features to obtain multiple similar feature sets, wherein the similar feature set is used to characterize the semantic feature set, and the first similarity between one of the semantic features in each of the similar feature sets and any other semantic feature is greater than a first preset value.

2. The method according to claim 1, characterized in that In the case where the enterprise demand data is the text data, the adopting the big data model to perform feature recognition on the enterprise demand data to obtain a semantic feature set includes: In the case where the enterprise demand data is the text data, executing the second step to obtain a semantic feature set; Wherein, the second step comprises: The big data model is used to extract key semantic features of the text data to obtain a key feature set, and the key feature set is used to characterize the semantic feature set.

3. The method according to claim 2, characterized in that In the case where the enterprise demand data is the text data and the image data, the adopting the big data model to perform feature recognition on the enterprise demand data to obtain a semantic feature set includes: Execute the second step to obtain a key feature set; Execute the first step to obtain a plurality of similar feature sets; respectively calculating a second similarity between each key semantic feature in the key feature set and each of the similar feature sets; When the second similarity is maximum, the correspondence between the key semantic feature and the similar feature set is determined, and the key semantic feature and the similar feature set are merged according to the correspondence to obtain an updated feature set, wherein the updated feature set is used to characterize the semantic feature set.

4. The method according to claim 3, characterized in that The method further comprises: If there are at least two corresponding relationships between the key semantic feature and the similar feature set when the second similarity is maximum, then executing the third step to obtain a plurality of the semantic feature sets, and there are at least two corresponding relationships between the key semantic feature and the similar feature set for representing that when the second similarity is maximum, there are at least two similar feature sets corresponding to the key semantic feature; Wherein, the third step comprises: Taking each of the similar feature sets corresponding to the maximum second similarity as a grid, and arranging each of the grids in sequence according to the first order; The spatial distance between the key semantic feature and each of the grids is calculated respectively, and the similar feature set corresponding to the grid with the shortest spatial distance is used as the target feature set, and the key semantic feature is merged with the target feature set to obtain an updated feature set, wherein the updated feature set is used to characterize the semantic feature set.

5. The method according to claim 1, characterized in that The step of converting the key image feature into a second semantic feature and merging the second semantic feature with the first semantic feature to obtain a semantic feature set includes: Performing dimensionality reduction on the key image features to obtain a low-dimensional feature vector; Inputting the low-dimensional feature vector into a semantic mapping model to generate a corresponding second semantic feature; The first semantic feature and the second semantic feature are concatenated to obtain a semantic feature set.

6. The method according to claim 4, characterized in that The spatial distance is represented by the Euclidean distance, and the respectively calculating the spatial distance between the key semantic feature and each of the grids includes: For each of the grids, a Euclidean distance calculation formula is used to calculate the distance between the key semantic feature and each semantic feature in the grid, and the minimum value of all the distances is calculated as the distance between the key semantic feature and the grid; The Euclidean distance calculation formula includes: ； In the formula, is the distance between the key semantic feature k and the grid G, k is the key semantic feature, s is one of the semantic features in the grid, G is a grid, k i With s i are the i-th components of k and s respectively.

7. The method according to claim 1, characterized in that The electronic device is also deployed with a Transformer model, and the establishing of an interactive attention mechanism between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set includes: By using the Transformer model, constructing an interactive attention mechanism between the embedding vector set and the candidate solution set, and using the interactive attention mechanism to calculate an attention score between the embedding vector set and the candidate solution set; Calculating an attention weight according to the attention score; The attention weights are multiplied by a value matrix generated according to the candidate solutions, and the obtained attention output is used to capture the matching relationship between the embedding vector set and the candidate solution set.

8. The method according to claim 7, characterized in that The big data model is a Transformer-based multimodal big data model, the framework of the Transformer-based multimodal big data model is a Transformer model, and the structure of the Transformer-based multimodal big data model includes an embedding module, a graph neural network module, a modal fusion module and an output layer; Among them, the embedding module is used to convert the semantic feature set into an embedding vector set, the graph neural network module is used to perform image recognition on the image data, the modal fusion module is used to establish an interactive attention mechanism between the embedding vector set and the candidate solution set to capture the matching relationship between the embedding vector set and the candidate solution set, and the output layer is used to output the enterprise demand matching solution corresponding to the embedding vector set.

9. An enterprise demand matching system based on a big data model, characterized in that: The enterprise demand matching method based on a big data model applied to any one of claims 1 to 8, the system comprising: a memory configured to store instructions; and A processor is configured to call the instructions from the memory and to implement the enterprise demand matching method based on a big data model according to any one of claims 1 to 8 when executing the instructions.

Images