Standardized Requirement Retrieval and Matching Method and System Based on User Portrait Feature Fusion

Abstract

The present invention belongs to the field of retrieval and matching, and provides a standardized requirement retrieval and matching method and system based on user portrait feature fusion. First, the standard writing requirements are split into characters, and after splitting, pooling operations and dimension mapping are performed to obtain dimension mapping features. After splitting the dimension mapping features into multiple sub-features, the sub-features are pairwise processed by multi-dimensional symmetric cross-attention to obtain cross-attention splicing features. The cross-attention splicing features are subjected to multiple convolutional splices to achieve feature cross-fusion, and cross-fusion features are obtained. Feature mapping is performed on the cross-fusion features to obtain three different user portrait index mapping vectors. The present invention extracts the standardized writing user portrait features, summarizes and generalizes the user portrait features to form a feature index set, learns the semantic information contained in the text, transforms the feature extraction task into a feature index classification task, and makes the user portrait more accurately depicted.

Description

Technical Field

[0001] The present invention belongs to the technical field of retrieval and matching, and particularly relates to a standardized requirement retrieval and matching method and system based on user portrait feature fusion. Background Art

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Under the background of the rapid development of information technology today, the demand for writing standardized documents is increasing day by day, especially in the fields of knowledge management and content creation. However, when dealing with the description of standardized knowledge by users, the existing technologies often lack effective automated methods to accurately match and provide relevant knowledge. Traditional solutions rely on static classification systems and simple keyword matching, and these methods are unable to cope when faced with complex and changing user needs. In addition, due to the lack of in-depth understanding and analysis of user descriptions, these technologies cannot effectively depict user portraits, resulting in low accuracy and efficiency of knowledge matching. Therefore, it is necessary to develop a new technical solution to learn and understand user descriptions through a deep learning model, so as to achieve precise classification of user needs and portrait characterization, and ultimately achieve the purpose of accurately matching standardized knowledge in the knowledge base. Currently, the research on user portrait feature extraction for standardized writing users and the matching of requirements and knowledge based on user portraits is still in its infancy, and there are no mature rules and methods yet. Summary of the Invention

[0004] To solve the above problems, the present invention proposes a standardized requirement retrieval and matching method and system based on user portrait feature fusion. The present invention extracts the user portrait features of standardized writing users based on a deep learning algorithm of feature cross-stitched fusion, summarizes and generalizes the user portrait features to form a set of feature index, learns the semantic information contained in the text, transforms the feature extraction task into a feature index classification task, and makes the user portrait more accurately depicted.

[0005] According to some embodiments, the first solution of the present invention provides a standardized requirement retrieval and matching method based on user portrait feature fusion, and adopts the following technical solution:

[0006] The standardized requirement retrieval and matching method based on user portrait feature fusion includes:

[0007] Obtain the standard writing requirements of the standard writer, and use a pre-trained feature cross-stitched fusion network model for classification to obtain three different user portrait index mapping vectors;

[0008] Perform semantic retrieval and matching on the knowledge in the knowledge base based on the user portrait index mapping vector;

[0009] Among them, classifying using the pre-trained feature cross-stitched fusion network model to obtain three different user portrait index mapping vectors, specifically:

[0010] First perform character splitting on the standard writing requirements, and after splitting, perform pooling operations and dimensionality mapping to obtain dimensional mapping features;

[0011] After splitting the dimensional mapping features into multiple sub-features, perform multi-dimensional symmetric cross-attention processing on the sub-features in pairs to obtain cross-attention stitching features;

[0012] Perform multiple convolutional stitching on the cross-attention stitching features to achieve feature cross-fusion and obtain cross-fusion features;

[0013] Perform feature mapping on the cross-fusion features to obtain three different user feature mapping vectors.

[0014] Furthermore, the step of first performing character splitting on the standard writing requirements, and after splitting, performing pooling operations and dimensionality mapping to obtain dimensional mapping features is specifically:

[0015] Obtain the standard writing requirement text data for word embedding and preliminary feature extraction to obtain requirement text features;

[0016] Use the text delimiter tags in the requirement text features to construct a mask matrix, and disassemble the requirement text features based on the mask matrix;

[0017] Perform pooling operations on the disassembled features, and perform character relevance matrix mapping on the features after pooling operations to obtain dimensional mapping features.

[0018] Furthermore, the step of using the text delimiter tags in the requirement text features to construct a mask matrix is specifically:

[0019] Divide the requirement text features into n + 1 text fields according to n text delimiter tags;

[0020] Set a one-dimensional pre-mask matrix, and in the order of the fields, define the corresponding elements in the pre-mask matrix as 1 for the number of characters in the text field, and set the remaining elements to 0;

[0021] And so on, finally obtaining n + 1 mask matrices.

[0022] Furthermore, the step of after splitting the dimensional mapping features into multiple sub-features, performing multi-dimensional symmetric cross-attention processing on the sub-features in pairs to obtain cross-attention stitching features is specifically:

[0023] After splitting the dimensional mapping features into multiple sub-features, input them in pairs into multiple multi-dimensional symmetric cross-attention modules;

[0024] Each multi-dimensional symmetric cross-attention module first performs multi-head attention operation and batch normalization on the combination of two inputs;

[0025] Then, multi-head attention operations are respectively performed on the two inputs and the result after batch normalization to obtain two attention features;

[0026] After respectively splicing the two attention features, the corresponding inputs and the result after batch normalization, batch normalization and convolution operations are performed again. The features after the two convolution operations are cross-spliced as the output of each multi-dimensional symmetric cross-attention module;

[0027] Through the processing of multiple multi-dimensional symmetric cross-attention modules, cross-attention splicing features are obtained.

[0028] Furthermore, the cross-attention splicing features are subjected to multiple convolution splicings to achieve feature cross-fusion, and cross-fusion features are obtained. Specifically:

[0029] The cross-attention splicing features are split into four cross-sub features;

[0030] The four cross-sub features are respectively input into four convolution modules for processing, and the outputs of the four convolution modules are respectively spliced in pairs. After obtaining two spliced outputs, convolution operations are respectively performed;

[0031] The features after the two convolution operations are cross-spliced and then average-pooled. The features after average pooling are respectively cross-spliced and convolved with the features after the two convolution operations to obtain two cross-convolution features;

[0032] The two cross-convolution features are combined and input to perform two different multi-head attention operations respectively, and finally cross-fusion features are obtained.

[0033] Furthermore, the cross-fusion features are subjected to feature mapping to obtain three different user feature mapping vectors. Specifically:

[0034] The cross-fusion features are split into two cross-fusion sub features and then convolution operations are respectively performed;

[0035] Then, the results of the two convolution operations are cross-spliced, convolved and flattened;

[0036] Three fully connected layers are used to output the results of the flattening operation to obtain three different user feature mapping vectors.

[0037] According to some embodiments, the second solution of the present invention provides a standardized requirement retrieval and matching system based on user portrait feature fusion, and adopts the following technical solutions:

[0038] A standardized requirement retrieval and matching system based on user profile feature fusion, comprising:

[0039] A user profile feature mapping module, configured to obtain the standard writing requirements of standard writers, classify them using a pre-trained feature cross-stitched fusion network model, and obtain three different user profile index mapping vectors;

[0040] A semantic retrieval and matching module, configured to perform semantic retrieval and matching based on the knowledge in the user profile index mapping vector knowledge base;

[0041] Among them, the classification using the pre-trained feature cross-stitched fusion network model to obtain three different user profile index mapping vectors is specifically as follows:

[0042] First, perform character splitting on the standard writing requirements, and after splitting, perform pooling operations and dimensional mapping to obtain dimensional mapping features;

[0043] After splitting the dimensional mapping features into multiple sub-features, perform multi-dimensional symmetric cross-attention processing on the sub-features in pairs to obtain cross-attention stitching features;

[0044] Perform multiple convolutional stitchings on the cross-attention stitching features to achieve feature cross-fusion and obtain cross-fusion features;

[0045] Perform feature mapping on the cross-fusion features to obtain three different user feature mapping vectors.

[0046] According to some embodiments, the third aspect of the present invention provides a computer-readable storage medium.

[0047] A computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps in the standardized requirement retrieval and matching method based on user profile feature fusion described in the first aspect above.

[0048] According to some embodiments, the fourth aspect of the present invention provides a computer device.

[0049] A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, it implements the steps in the standardized requirement retrieval and matching method based on user profile feature fusion described in the first aspect above.

[0050] According to some embodiments, the fifth aspect of the present invention provides a computer program product or a computer program.

[0051] The present invention provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the standardized requirement retrieval and matching method based on user portrait feature fusion as described in the first aspect above.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0053] Based on the deep learning algorithm of the feature cross splicing and fusion network model, the present invention extracts the standardized written user portrait features, summarizes and generalizes the user portrait features to form a set of feature indicators, learns the semantic information contained in the text, converts the feature extraction task into a feature indicator classification task, and makes the user portrait more accurate; The user portrait features are extracted and feature fusion is carried out by means of cross splicing, multi-dimensional fusion, etc., which improves the quality of semantic features, improves the accuracy of the final user portrait index classification, and makes the description of the user portrait more reasonable and more in line with the actual application situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0055] Figure 1 is a training flow chart of the standardized requirement retrieval and matching method based on user portrait feature fusion in an embodiment of the present invention;

[0056] Figure 2 is to train a feature cross splicing and fusion network model with a requirement text data set in an embodiment of the present invention;

[0057] Figure 3 is a structure diagram of a character splitting and dimension mapping module in an embodiment of the present invention;

[0058] Figure 4 is a structure diagram of a multi-dimensional symmetric cross attention module in an embodiment of the present invention;

[0059] Figure 5 is a structure diagram of a user portrait feature cross fusion module in an embodiment of the present invention;

[0060] Figure 6 is a structure diagram of a user portrait feature label mapping module in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0062] It should be noted that the following detailed description is illustrative and aims to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention pertains.

[0063] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0064] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0065] Embodiment 1

[0066] This embodiment provides a standardized requirement retrieval and matching method based on user portrait feature fusion. This embodiment takes the application of this method to a server as an example for illustration. It can be understood that this method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, web servers, cloud communications, middleware services, domain name services, security services CDN, and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server can be directly or indirectly connected through wired or wireless communication methods, and this application does not make any restrictions here. In this embodiment, the method includes the following steps:

[0067] Obtain the standard writing requirements of the standard writer, and classify them using a pre-trained feature cross-stitched and fused network model to obtain three different user portrait index mapping vectors;

[0068] Perform semantic retrieval and matching based on the knowledge in the user portrait index mapping vector knowledge base;

[0069] Among them, the use of the pre-trained feature cross-stitched and fused network model for classification to obtain three different user portrait index mapping vectors is specifically as follows:

[0070] First, perform character splitting on the standard writing requirements, and after splitting, perform pooling operations and dimensionality mapping to obtain dimensionality mapping features;

[0071] After splitting the dimensional mapping features into multiple sub - features, pairwise multi - dimensional symmetric cross - attention processing is performed on the sub - features to obtain cross - attention concatenated features;

[0072] Performing multiple convolutional concatenations on the cross - attention concatenated features to achieve feature cross - fusion, obtaining cross - fused features;

[0073] Performing feature mapping on the cross - fused features to obtain three different user feature mapping vectors.

[0074] In order to accurately divide the user portraits of standard - writing users and precisely match the standardized requirement knowledge, the standardized requirement and knowledge matching method based on user portrait feature cross - splicing and fusion disclosed by the present invention, as Figure 1 shown, specifically includes the following steps.

[0075] S1: Obtain the standard - writing requirements of the standard - writer and construct a standard - writing requirement text data set.

[0076] S2: Extract the user portrait feature labels of the standard - writer based on the requirement text data set and label the data.

[0077] S3: Build a feature cross - splicing and fusion network.

[0078] S4: Use the labeled standard - writing requirement text data set to train the feature cross - splicing and fusion network model. Input the data in the data set into the BERT pre - trained model to obtain features . Input the features into the character splitting and dimensional mapping module, and output to obtain features . Input the features into the multi - dimensional symmetric cross - attention module group, and output to obtain features . Input the features into the user portrait feature cross - fusion module, and output to obtain features . Input the features into the user portrait index mapping module, and output to obtain features . Use to map the user portrait feature labels. Calculate the loss using the features , and perform gradient optimization on the feature cross - splicing and fusion network model through an optimizer.

[0079] S5: Construct a user portrait feature label mapping and standardized requirement knowledge matching method.

[0080] S6: Deploy the feature cross-stitched fusion network model and test the system using the requirement text of the new standard writers.

[0081] The following elaborates on each of the above steps in detail.

[0082] I. Step S1: Obtain the standard writing requirements of the standard writers and construct a standard writing requirement text dataset.

[0083] S1-1: Collect the requirements of the standard writers for standardization knowledge during the standard drafting process; the above requirement information is obtained through multiple channels such as interviews and questionnaires, and the obtained user requirement information is stored in text form.

[0084] S1-2: Conduct a preliminary segmentation of the text information, replace the punctuation marks between texts with the [SEP] symbol, and [SEP] is used to identify the demarcation point between the previous text data and the next text data.

[0085] II. Step S2: Extract the feature labels of the standard writer user portraits based on the requirement text dataset and label the data.

[0086] S2-1: Decompose and analyze the collected standard writing requirement text dataset, and extract two-layer user portrait feature labels in combination with the text content and expert opinions. The first layer is divided into three aspects: industry field, standardization document type, and writing purpose. The second layer is an extension of the first layer. In terms of the industry field, it is divided into 21 feature labels such as "Agriculture, Forestry, Animal Husbandry, and Fishery", "Mining", "Manufacturing", "Construction", "Wholesale and Retail", "Transportation", "Warehousing and Postal Services", "Accommodation and Catering", "Information Transmission, Software, and Information Technology Services", "Finance", "Real Estate", "Leasing and Business Services", "Scientific Research and Technical Services", "Water Conservancy, Environment, and Public Facilities Management", "Resident Services, Repair, and Other Services", "Education", "Health and Social Work", "Culture, Sports, and Entertainment", "Public Management", "Social Security", and "Social Organizations"; in terms of the standardization document type, it is divided into 7 feature labels such as national standards, industry standards, local standards, enterprise standards, specification standards, procedure standards, and guide standards; in terms of the writing purpose, it is divided into 5 feature labels such as ensuring quality and safety, promoting efficiency and consistency, supporting technical compatibility and consistency, enhancing international trade and market access, and compliance regulations.

[0087] S2-2: Use 0, 1, and 2 to identify the three aspects of industry field, standardization document type, and writing purpose in the first layer.

[0088] Use 0a, 0b, 0c, 0d, 0e, 0f, 0g, 0h, 0i, 0j, 0k, 0l, 0m, 0n, 0o, 0p, 0q, 0r, 0s, 0t, 0u to identify agriculture, forestry, animal husbandry, and fishery, mining, manufacturing, construction, wholesale and retail, transportation, warehousing and postal services, accommodation and catering, information transmission, software and information technology services, finance, real estate, leasing and business services, scientific research and technical services, water conservancy, environment and public facilities management, residential services, repair and other services, education, health and social work, culture, sports and entertainment, public management, social security and social organizations in the industry field;

[0089] Use 1a, 1b, 1c, 1d, 1e, 1f, and 1g to identify national standards, industry standards, local standards, enterprise standards, specification standards, procedure standards, and guide standards in terms of standardization document types, and use 2a, 2b, 2c, 2d, and 2e to identify ensuring quality and safety, promoting efficiency and consistency, supporting technical compatibility and consistency, enhancing international trade and market access, and compliance specifications in terms of writing purposes.

[0090] S2-3: Annotate the data in the standard writing requirement text dataset with reference to expert opinions and user portrait feature tags. Three examples after annotation: ① , ② , ③ .

[0091] S2-4: Divide the standard writing requirement text dataset into a training set, a validation set, and a test set according to the ratio of 6:2:2.

[0092] III. Step S3: Build a feature cross-joining and fusion network.

[0093] S3-1: As Figure 2 shown, the feature cross-joining and fusion network model includes a character splitting and dimension mapping module, a multi-dimensional symmetric cross-attention module group, a user portrait feature cross-fusion module, and a user portrait feature tag mapping module.

[0094] S3-2: Specifically, input the data in the standard writing requirement text dataset into the BERT pre-trained model to perform word embedding and preliminary feature extraction on the characters, and output the requirement text features .

[0095] For the standard writing requirements, first perform character splitting, then perform pooling operations and dimension mapping to obtain dimension mapping features, specifically:

[0096] Obtain the standard writing requirement text data for word embedding and preliminary feature extraction to get the requirement text features;

[0097] Construct a mask matrix using the text delimiter tags in the requirement text features, and decompose the requirement text features based on the mask matrix;

[0098] Perform a pooling operation on the decomposed features, and map the features after the pooling operation to a character correlation matrix to obtain dimension-mapped features.

[0099] The construction of the mask matrix using the text delimiter tags in the requirement text features is specifically as follows:

[0100] Divide the requirement text features into n + 1 text fields according to n text delimiter tags;

[0101] Set a one-dimensional pre-mask matrix. In the order of the fields, define the elements in the corresponding order in the pre-mask matrix as 1 for the number of characters in the text field, and set the remaining elements to 0;

[0102] And so on, finally obtaining n + 1 mask matrices.

[0103] The character splitting and dimension mapping module is as Figure 3 shown. In the splitting module, a mask matrix is constructed using the [SEP] tags in the text data. When a text data contains n [SEP] - text delimiter tags, the number of mask matrices constructed based on the [SEP] tags is n + 1. Use the mask matrix to decompose the requirement text features .

[0104] Specifically, the mask matrix is specifically: set a pre-mask matrix, denoted as as the number of characters. First, set the in to 0. Use the [SEP] identifier to construct a mask. The mask matrix of the first field is denoted as , where the number of 1s is the number of characters before the first [SEP] identifier, and the number of 0s is the number of other characters. The mask matrix of the second field is denoted as , where the number of 1s is the number of characters between the first [SEP] identifier and the second [SEP] identifier, and the subsequent mask construction is the same.

[0105] Apply to the LSE pooling operation to obtain the pooled output feature . The LSE pooling operation can be expressed by the formula:

[0106] (1);

[0107] Among them, is a hyperparameter that can be optimized by training with a deep learning model; is the total number of all points in the feature matrix, is the th row and th column feature point in the input and output features; is the output feature point after the LSE pooling operation.

[0108] Map the pooled output feature to a character correlation matrix, use the character correlation mapping function to perform matrix mapping on the features, and obtain the mapped output feature .

[0109] The character correlation mapping function can be expressed as the formula:

[0110] (2);

[0111] Among them, is the input feature matrix, is the learnable parameter matrix, is the learnable bias term, is the output feature matrix. When the dimension of the input feature matrix is , the dimension of the learnable parameter matrix is to ensure that the output feature matrix is a square matrix. For the convenience of feature transmission, group the mapped output feature set, denoted as the dimension-mapped feature .

[0112] S3-4: After splitting the dimension-mapped feature into multiple sub-features, perform multi-dimensional symmetric cross-attention processing on the sub-features in pairs to obtain the cross-attention concatenated feature, specifically:

[0113] After splitting the dimension-mapped feature into multiple sub-features, input them in pairs into multiple multi-dimensional symmetric cross-attention modules;

[0114] Each multi-dimensional symmetric cross-attention module first performs multi-head attention operation and batch normalization on the combination of the two inputs;

[0115] Then perform multi-head attention operations on the two inputs and the result after batch normalization respectively to obtain two attention features;

[0116] After concatenating the two attention features, the corresponding inputs, and the result after batch normalization respectively, perform batch normalization and convolution operations, and cross-concatenate the features after the two convolution operations as the output of each multi-dimensional symmetric cross-attention module;

[0117] After being processed by multiple multi-dimensional symmetric cross-attention modules, cross-attention splicing features are obtained.

[0118] The multi-dimensional symmetric cross-attention module group is composed of multiple multi-dimensional symmetric cross-attention modules. The multi-dimensional symmetric cross-attention module is as Figure 4 shown. The multi-dimensional symmetric cross-attention module group obtains the output features of the character splitting and dimension mapping module - dimension mapping features and disassembles them to form dimension mapping sub-features, denoted as . is input into multiple multi-dimensional symmetric cross-attention module groups.

[0119] The output of the multi-dimensional symmetric cross-attention module group is 4 feature matrices. Therefore, the number of multi-dimensional symmetric cross-attention modules is related to the number of dimension mapping sub-features . Assuming the input is dimension mapping sub-features, the first layer requires multi-dimensional symmetric cross-attention modules to form, and outputs features; the second layer requires multi-dimensional symmetric cross-attention modules to form, and outputs features, and so on until the number of output features is 4. The total number of multi-dimensional symmetric cross-attention modules required can be calculated by the formula .

[0120] The multi-dimensional symmetric cross-attention module obtains two inputs, denoted as input one and input two respectively. and are input into the first multi-head attention of the multi-dimensional symmetric cross-attention module to obtain the output feature . The feature is input into the first batch normalization module of the multi-dimensional symmetric cross-attention module to obtain the output feature .

[0121] and the feature are input into the second multi-head attention of the multi-dimensional symmetric cross-attention module to obtain the output attention feature . and the feature are input into the third multi-head attention of the multi-dimensional symmetric cross-attention module to obtain the output attention feature .

[0122] 、feature and feature ​​Add to obtain features . Add , feature and feature to obtain feature .

[0123] Input feature into the second batch normalization module of the multi-dimensional symmetric cross-attention module, and output to obtain feature . Input feature into the third batch normalization module of the multi-dimensional symmetric cross-attention module, and output to obtain feature .

[0124] Input feature into the first convolution module of the multi-dimensional symmetric cross-attention module, and output to obtain feature . Input feature into the second convolution module of the multi-dimensional symmetric cross-attention module, and output to obtain feature .

[0125] Input feature and feature into the cross splicing module of the multi-dimensional symmetric cross-attention module, and output to obtain feature . The four feature matrices finally output by the above multi-dimensional symmetric cross-attention module are denoted as . For the convenience of feature transmission, group the feature set, denoted as cross-attention splicing feature .

[0126] S3-5: The above-mentioned cross-attention splicing feature is subjected to multiple convolution splicings to achieve feature cross-fusion, and the cross-fusion feature is obtained. Specifically:

[0127] Split the cross-attention splicing feature into four cross-sub features;

[0128] Input the four cross-sub features into four convolution modules for processing respectively, and splice the outputs of the four convolution modules in pairs respectively. After obtaining two splicing outputs, perform convolution operations respectively;

[0129] Cross-splice the features after the two convolution operations and then perform average pooling. The features after average pooling are cross-spliced and convolved with the features after the two convolution operations respectively to obtain two cross-convolution features;

[0130] Merge and input the two cross-convolution features to perform two different multi-head attention operations respectively, and finally obtain the cross-fusion feature.

[0131] The user portrait feature cross-fusion module is as Figure 5 shown. Input feature Decompose into 4 cross sub - features . Input the feature into the first convolutional module in the user portrait feature cross - fusion module, and output the feature . Input the feature into the second convolutional module in the user portrait feature cross - fusion module, and output the feature . Input the feature into the third convolutional module in the user portrait feature cross - fusion module, and output the feature . Input the feature into the fourth convolutional module in the user portrait feature cross - fusion module, and output the feature .

[0132] Input the feature and into the first cross - splicing module in the user portrait feature cross - fusion module, and output the feature . Input the feature and into the second cross - splicing module in the user portrait feature cross - fusion module, and output the feature .

[0133] Input the feature into the fifth convolutional module in the user portrait feature cross - fusion module, and output the feature . Input the feature into the sixth convolutional module in the user portrait feature cross - fusion module, and output the feature . The feature and are input into the third cross - splicing module in the user portrait feature cross - fusion module, and output the feature . Input the feature into the average pooling module in the user portrait feature cross - fusion module, and output the feature .

[0134] Input the feature and into the second cross - splicing module in the user portrait feature cross - fusion module, and output the feature . Input the feature and into the second cross - splicing module in the user portrait feature cross - fusion module, and output the feature . Input the feature into the seventh convolutional module in the user portrait feature cross - fusion module, and output the cross - convolutional feature . Input the feature In the eighth convolution module of the input user portrait feature cross-fusion module, cross-convolution features are output. .

[0135] The feature and are input into the first multi-head attention module of the input user portrait feature cross-fusion module, and the feature is output. The feature and are input into the second multi-head attention module of the input user portrait feature cross-fusion module, and the feature is output. To facilitate feature transmission, the feature and the feature are grouped, denoted as the feature .

[0136] S3-6: The cross-fusion features are feature-mapped to obtain three different user feature mapping vectors, specifically:

[0137] The cross-fusion features are split into two cross-fusion sub-features and then convolutional operations are performed separately;

[0138] Then, the results of the two convolutional operations are cross-stitched, convolved, and flattened;

[0139] The results of the flattening operation are output using three fully connected layers to obtain three different user feature mapping vectors.

[0140] The user portrait feature label mapping module is as Figure 6 shown. The feature is disassembled into 2 cross-fusion sub-features . The feature is input into the first convolutional group of the user portrait feature label mapping module, and the feature is output. The feature is input into the second convolutional group of the user portrait feature label mapping module, and the feature is output.

[0141] The feature and the feature are input into the cross-stitching module of the user portrait feature label mapping module, and the feature is output. The feature is input into the third convolutional group of the user portrait feature label mapping module, and the feature is output. The feature is input into the flattening operation module of the user portrait feature label mapping module, and the feature is output.

[0142] The feature In the first fully connected layer group in the input user portrait feature label mapping module, the user portrait index mapping vector one is output. The feature is input into the second fully connected layer group in the user portrait feature label mapping module, and the user portrait index mapping vector two is output. The feature is input into the third fully connected layer group in the user portrait feature label mapping module, and the user portrait index mapping vector three is output. .

[0143] S3-7: The cross-stitched fusion described in steps S3-4, S3-5, and S3-6 is specifically as follows:

[0144] The cross-stitched fusion module receives two inputs of the same size, denoted as and . The two inputs and are cross-arranged, which can be expressed as

[0145] (3);

[0146] Use this method for cross-stitched fusion of features. When the size of the input feature is , the size of the output feature after passing through the cross-stitched fusion module becomes of the input feature size.

[0147] IV. Step S4: Use the dataset of the written requirement text to obtain a deployable feature cross-stitched fusion network model.

[0148] S4-1: The loss function of the feature cross-stitched fusion network model is obtained by summing the classification losses of the user portrait index mapping vector one, the classification losses of the user portrait index mapping vector two, and the classification losses of the user portrait index mapping vector three. Both loss functions are cross-entropy loss functions.

[0149] Among them, the cross-entropy loss function is defined as follows:

[0150] (4);

[0151] Among them, is the loss value, represents the sample serial number, represents the number of samples, represents the embedded representation of the label true value, represents the predicted value of the user portrait label mapping.

[0152] S4-2: During the training and validation processes, the SGD optimizer is used to optimize the gradients of the feature cross-stitched fusion network model.

[0153] V. Step S5: Construct a user profile feature label mapping and standardized requirement knowledge matching method.

[0154] S5-1: Normalize the index mapping vectors. Input the data into the deployed feature cross-stitched fusion network model to obtain three label classification outputs of the feature cross-stitched fusion network model, namely the user profile index mapping vector one, the user profile index mapping vector two, and the user profile index mapping vector three.

[0155] Normalize the three user profile index mapping vectors and standardize the index calculation scores to the range of [0, 1].

[0156] Use the normalization formula to perform normalization calculation on the index calculation scores. Among them, is the th element in the one-dimensional feature, is the minimum value of all elements in the one-dimensional feature, is the maximum value of all elements in the one-dimensional feature. The output of this step is three normalized user profile index mapping vectors, denoted as , and .

[0157] S5-2: User profile feature mapping. Read the element values of the normalized user profile index mapping vectors , and , and extract the elements in the mapping vectors whose element values are greater than or equal to 0.75.

[0158] Suppose has 24 elements as follows:

[0159] [0.47, 0.21, 0.29, 0.56, 0.23, 0.17, 0.64, 0.35, 0.72, 0.09, 0.51, 0.27, 0.89, 0.26, 0.83, 0.41, 0.25, 0.86, 0.28, 0.24, 0.58]. Among them, the numbers greater than or equal to 0.75 are 0.89, 0.83, 0.86, and the element indices corresponding to these three numbers are: 13, 16, 18.

[0160] Use the element index to extract the industry fields corresponding to this user demand data from the index vector ["Agriculture, Forestry, Animal Husbandry and Fishery", "Mining", "Manufacturing", "Construction", "Wholesale and Retail", "Transportation", "Warehousing and Postal Services", "Accommodation and Catering Services", "Information Transmission, Software and Information Technology Services", "Finance", "Real Estate", "Leasing and Business Services", "Scientific Research and Technical Services", "Water Conservancy, Environment and Public Facilities Management", "Resident Services, Repair and Other Services", "Education", "Health and Social Work", "Culture, Sports and Entertainment", "Public Administration", "Social Security", "Social Organizations"], that is, the industry fields corresponding to 13, 16, 18 ["Scientific Research and Technical Services", "Education", "Culture, Sports and Entertainment"]. and the user portrait feature mapping method is the same as and use the extracted user portrait features to represent the user portrait.

[0161] S5-3: Match the user demand label features with the knowledge. Use the extracted user portrait features to retrieve and match the knowledge in the knowledge base. Use the keyword semantic retrieval optimized based on cosine similarity to retrieve and match the content in the knowledge base. Map the text in the knowledge base and the extracted user portrait feature label text into vectors using the TF-IDF vectorizer, and then use the sum of the cosine similarity algorithm and the Tanimoto coefficient algorithm as the standard calculation method to measure the similarity between the two vectors, calculate the matching score between the two, and perform sorting according to the matching score and then push the standard domain knowledge for the user portrait. For a t-dimensional vector, the algorithm formula is as follows:

[0162] (5);

[0163] where is the vector formed by mapping the text in the knowledge base through the TF-IDF vectorizer, is the vector formed by mapping the extracted user portrait feature label text through the TF-IDF vectorizer, is the dimension of the vector.

[0164] VI. Step S6: Deploy the feature cross-stitched fusion network model and test the system using the demand text of the new standard writer.

[0165] Package the above models and methods into a standardized requirement knowledge matching system and deploy it. Test the system using the requirement texts of new standard writers. Input the new user requirement texts into the standardized requirement knowledge matching system according to this method, and output the user portrait index mapping vectors. Describe the user portrait through the user portrait feature tags mapped by the vectors. Use the obtained user portrait tags to perform matching retrieval of the knowledge in the knowledge base using keyword retrieval optimized by semantic search.

[0166] The role of the user portrait index mapping vector is to calculate the matching degree between the currently input requirement and the knowledge in the knowledge base. For example, assume that a text is input, and the results obtained from three index mapping vectors are "Field: Agriculture, Forestry, Animal Husbandry and Fishery, Standardized Document Type: National Standard, Compilation Purpose: Promote Efficiency and Consistency".

[0167] Use a matching algorithm to match the knowledge in the knowledge base with the tag content, that is, take the one with the highest matching similarity as the final matching result.

[0168] Embodiment 2

[0169] This embodiment provides a standardized requirement retrieval and matching system based on user portrait feature fusion, including:

[0170] A user portrait feature mapping module, configured to obtain the standard writing requirements of standard writers, perform classification using a pre-trained feature cross-stitched fusion network model, and obtain three different user portrait index mapping vectors;

[0171] A semantic retrieval and matching module, configured to perform semantic retrieval and matching on the knowledge in the knowledge base based on the user portrait index mapping vectors;

[0172] Among them, the process of using the pre-trained feature cross-stitched fusion network model to perform classification and obtain three different user portrait index mapping vectors is specifically as follows:

[0173] First, perform character splitting on the standard writing requirements, and then perform pooling operations and dimensionality mapping after splitting to obtain dimensionality mapping features;

[0174] After splitting the dimensionality mapping features into multiple sub-features, perform multi-dimensional symmetric cross-attention processing on the sub-features in pairs to obtain cross-attention stitching features;

[0175] Perform multiple convolutional stitchings on the cross-attention stitching features to achieve feature cross-fusion and obtain cross-fusion features;

[0176] Perform feature mapping on the cross-fusion features to obtain three different user feature mapping vectors.

[0177] The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the first embodiment above. It should be noted that the above modules, as part of a system, can be executed in a computer system such as a set of computer-executable instructions.

[0178] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0179] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the above modules is only a logical function division. In actual implementation, there can be other division methods. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0180] Embodiment Three

[0181] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in the method for standardized requirement retrieval and matching based on user portrait feature fusion as described in the first embodiment above.

[0182] Embodiment Four

[0183] This embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for standardized requirement retrieval and matching based on user portrait feature fusion as described in the first embodiment above.

[0184] Embodiment Five

[0185] This embodiment provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the method for standardized requirement retrieval and matching based on user portrait feature fusion as described in the first embodiment above.

[0186] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories and optical memories, etc.) containing computer-usable program code.

[0187] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0188] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0189] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0190] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0191] Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation to the protection scope of the present invention. Those skilled in the art should understand that based on the technical solutions of the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present invention.

Claims

1. A standardized demand retrieval and matching method based on user portrait feature fusion, characterized in that: include: Obtain the standard writing requirements of the standard writer, use the pre-trained feature cross-stitching fusion network model for classification, and obtain three different user portrait indicator mapping vectors; Perform semantic retrieval and matching based on the knowledge in the user portrait index mapping vector knowledge base; Among them, the pre-trained feature cross-stitching fusion network model is used for classification to obtain three different user portrait indicator mapping vectors, specifically: For standard writing requirements, character splitting is first performed, followed by pooling and dimension mapping to obtain dimension mapping features. After splitting the dimension mapping feature into multiple sub-features, the sub-features are processed with multi-dimensional symmetrical cross attention in pairs to obtain the cross attention splicing feature; The cross-attention splicing features are convolved multiple times to achieve feature cross-fusion and obtain cross-fusion features; Perform feature mapping on the cross-fusion features to obtain three different user feature mapping vectors; The cross-attention splicing features are convolved multiple times to achieve feature cross-fusion, and obtain cross-fusion features, specifically: Split the cross-attention concatenation feature into four cross-sub-features; The four cross sub-features are input into four convolution modules for processing respectively, and the outputs of the four convolution modules are spliced ​​in pairs to obtain two spliced ​​outputs and then convolution operations are performed respectively; The features after the two convolution operations are cross-joined and then average-pooled. The features after average-pooling are cross-joined and convolved with the features after the two convolution operations to obtain two cross-convolution features. The two cross-convolution features are combined and inputted to perform two different multi-head attention operations, and finally the cross-fusion features are obtained.

2. The standardized demand retrieval and matching method based on user portrait feature fusion according to claim 1 is characterized in that: The standard writing requirements are first split into characters, and then pooling and dimension mapping are performed to obtain dimension mapping features, which are specifically: Obtain standard writing requirement text data for word embedding and preliminary feature extraction to obtain requirement text features; The text boundary labels in the requirement text features are used to construct a mask matrix, and the requirement text features are decomposed based on the mask matrix; The decomposed features are pooled, and the features after the pooling operation are mapped into a character correlation matrix to obtain dimension mapping features.

3. The standardized demand retrieval and matching method based on user portrait feature fusion according to claim 2 is characterized in that: The mask matrix is ​​constructed by using the text boundary labels in the demand text features, specifically: The requirement text features are divided into n+1 text fields according to n text boundary labels; Set a one-dimensional pre-mask matrix. According to the order of the fields, define the elements in the pre-mask matrix with the corresponding order as 1, which is the number of characters in the text field, and set the remaining elements to 0; And so on, finally we get n+1 mask matrices.

4. The standardized demand retrieval and matching method based on user portrait feature fusion according to claim 1 is characterized in that: After the dimensional mapping feature is split into multiple sub-features, the sub-features are subjected to multi-dimensional symmetrical cross-attention processing in pairs to obtain a cross-attention splicing feature, specifically: After splitting the dimension mapping feature into multiple sub-features, they are input into multiple multi-dimensional symmetric cross attention modules in pairs; Each multi-dimensional symmetric cross attention module first merges the two inputs to perform multi-head attention operation and batch normalization; Then perform multi-head attention operation on the two inputs and the batch normalized results to obtain two attention features; After concatenating the two attention features and the corresponding input with the batch normalized results, batch normalization and convolution operations are performed, and the features after the two convolution operations are cross-concatenated as the output of each multi-dimensional symmetric cross attention module; After being processed by multiple multi-dimensional symmetric cross-attention modules, the cross-attention splicing features are obtained.

5. The standardized demand retrieval and matching method based on user portrait feature fusion according to claim 1 is characterized in that: The cross-fusion features are subjected to feature mapping to obtain three different user feature mapping vectors, specifically: Split the cross-fusion feature into two cross-fusion sub-features and perform convolution operations on each of them; The results of the two convolution operations are then cross-joined, convolved, and flattened; The results of the flattening operation are output using three fully connected layers to obtain three different user feature mapping vectors.

6. A standardized demand retrieval and matching system based on user portrait feature fusion, characterized in that: include: The user portrait feature mapping module is configured to obtain the standard writing requirements of the standard writer, and use the pre-trained feature cross-stitching fusion network model for classification to obtain three different user portrait indicator mapping vectors; A semantic search and matching module, configured to perform semantic search and matching based on knowledge in a user portrait indicator mapping vector knowledge base; Among them, the pre-trained feature cross-stitching fusion network model is used for classification to obtain three different user portrait indicator mapping vectors, specifically: For standard writing requirements, character splitting is first performed, followed by pooling and dimension mapping to obtain dimension mapping features. After splitting the dimension mapping feature into multiple sub-features, the sub-features are processed with multi-dimensional symmetrical cross attention in pairs to obtain the cross attention splicing feature; The cross-attention splicing features are convolved multiple times to achieve feature cross-fusion and obtain cross-fusion features; Perform feature mapping on the cross-fusion features to obtain three different user feature mapping vectors; The cross-attention splicing features are convolved multiple times to achieve feature cross-fusion, and obtain cross-fusion features, specifically: Split the cross-attention concatenation feature into four cross-sub-features; The four cross sub-features are input into four convolution modules for processing respectively, and the outputs of the four convolution modules are spliced ​​in pairs to obtain two spliced ​​outputs and then convolution operations are performed respectively; The features after the two convolution operations are cross-joined and then average-pooled. The features after average-pooling are cross-joined and convolved with the features after the two convolution operations to obtain two cross-convolution features. The two cross-convolution features are combined and inputted to perform two different multi-head attention operations, and finally the cross-fusion features are obtained.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the standardized demand retrieval and matching method based on user portrait feature fusion as described in any one of claims 1 to 5 are implemented.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the standardized demand retrieval and matching method based on user portrait feature fusion as described in any one of claims 1 to 5 are implemented.

9. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, the steps in the standardized demand retrieval and matching method based on user portrait feature fusion as described in any one of claims 1 to 5 are implemented.

Images