Tool capability classification method and apparatus, storage medium, and computer program product

Abstract

The present disclosure provides a tool capability classification method and apparatus, a storage medium, and a computer program product. The method comprises: using a multi-scale capsule network module to perform multi-scale feature extraction processing on capability description text information of a tool to obtain a first feature vector; by means of a sparse coding module and on the basis of the first feature vector, obtaining a second feature vector; and by means of a neural network model and on the basis of the second feature vector, obtaining capability classification result information for the tool.

Description

Tool capability classification method, device, storage medium and computer program product

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] The present disclosure is based on and claims priority to an application with CN application number 202410129426.0 and filing date January 30, 2024. The disclosed content of the CN application is hereby incorporated into the present disclosure as a whole. Technical Field

[0003] The present disclosure relates to the field of artificial intelligence technology, and in particular to a tool capability classification method, device, storage medium, and computer program product. Background Art

[0004] With the rapid development of artificial intelligence (AI) technology, classification methods or models can be used for classification processing in many application scenarios. For example, when classifying AI (Artificial Intelligence) tools, classification methods or models are used to determine the classification information of the tools. Classification methods can include feature conditional independence assumptions and decision tree clustering. These methods typically use certain rules to cluster data. However, their disadvantages are high data volatility and inability to consider nonlinear data analysis.

[0005] Summary of the Invention

[0006] According to a first aspect of the present disclosure, a tool capability classification method is provided, comprising: performing multi-scale feature extraction processing on tool capability description text information using a multi-scale capsule network module to obtain a first feature vector; obtaining a second feature vector based on the first feature vector using a sparse coding module; and obtaining capability classification result information for the tool based on the second feature vector using a neural network model.

[0007] In some embodiments, the use of a multi-scale capsule network module to perform multi-scale feature extraction processing on the tool capability description text information to obtain a first feature vector includes: using the multi-scale capsule network module and using convolution kernels of multiple convolution templates to perform multi-scale feature extraction processing on the capability description text information to obtain multiple feature information; and using the capsule network module to encode the multiple feature information to obtain the first feature vector.

[0008] In some embodiments, the multi-scale capsule network module includes: multiple low-level capsules and high-level capsules; using the capsule network module to encode the multiple feature information to obtain the first feature vector includes: using the multiple low-level capsules to process the multiple feature information to obtain multiple low-level capsule feature information; based on the multiple low-level capsule feature information and the corresponding affine matrix, obtaining multiple prediction vectors; performing weighted summation processing on the multiple prediction vectors according to the coupling coefficient to obtain an input vector; using an activation function to compress the input vector, and inputting the compressed processing result into the high-level capsule, so that the high-level capsule outputs the first feature vector based on the compressed processing result; wherein, the coupling coefficient is updated based on the first feature vector and using a dynamic routing algorithm.

[0009] In some embodiments, obtaining the second eigenvector using a sparse coding module and based on the first eigenvector includes: performing low-dimensional spatial feature extraction processing on the first eigenvector using a sparse coding model of the sparse coding module and based on a sparse coding matrix to obtain the second eigenvector.

[0010] In some embodiments, the neural network model includes: a convolution layer, a pooling layer and a softmax classifier; processing the second feature vector through the neural network model to obtain a classification result for the capability data includes: using the convolution layer and the pooling layer to perform feature extraction and fusion processing on the second feature vector, and inputting the processing result into the softmax classifier; and obtaining the capability classification result information based on the result output by the softmax classifier.

[0011] In some embodiments, evaluation index information is determined based on the capability classification result information; the classification result is evaluated according to the evaluation index information; wherein the evaluation index includes at least one of the following: an accuracy index, a precision index, a recall index, and a comprehensive index of precision and recall.

[0012] In some embodiments, the tool includes: an artificial intelligence (AI) tool; the capability description text information of the tool includes: capability description text information of the AI ​​tool.

[0013] According to a second aspect of the present disclosure, a tool capability classification device is provided, including: a first feature mining module, used to use a multi-scale capsule network module to perform multi-scale feature extraction processing on the tool capability description text information to obtain a first feature vector; a second feature mining module, used to use a sparse coding module and based on the first feature vector to obtain the second feature vector; a classification result acquisition module, used to obtain capability classification result information for the tool based on the second feature vector through a neural network model.

[0014] In some embodiments, the first feature mining module includes: a feature extraction unit, configured to utilize the multi-scale capsule network module and convolution kernels of multiple convolution templates to perform multi-scale feature extraction processing on the capability description text information to obtain multiple feature information; and a feature encoding unit, configured to utilize the capsule network module to encode the multiple feature information to obtain the first feature vector.

[0015] In some embodiments, the multi-scale capsule network module includes: multiple low-level capsules and high-level capsules; the feature encoding unit is used to process the multiple feature information using the multiple low-level capsules to obtain multiple low-level capsule feature information; based on the multiple low-level capsule feature information and the corresponding affine matrix, multiple prediction vectors are obtained; the multiple prediction vectors are weighted summed according to the coupling coefficient to obtain an input vector; the input vector is compressed using an activation function, and the compression processing result is input into the high-level capsule, so that the high-level capsule outputs the first feature vector based on the compression processing result; wherein the coupling coefficient is updated based on the first feature vector and using a dynamic routing algorithm.

[0016] In some embodiments, the second feature mining module is used to utilize the sparse coding model of the sparse coding module and based on the sparse coding matrix to perform low-dimensional spatial feature extraction processing on the first feature vector to obtain the second feature vector.

[0017] In some embodiments, the neural network model includes: a convolution layer, a pooling layer and a softmax classifier; the classification result acquisition module is used to use the convolution layer and the pooling layer to perform feature extraction and fusion processing on the second feature vector, and input the processing result into the softmax classifier; according to the result output by the softmax classifier, the ability classification result information is obtained.

[0018] In some embodiments, a classification evaluation module is used to determine evaluation index information based on the capability classification result information; and evaluate the classification result according to the evaluation index information; wherein the evaluation index includes at least one of the following: an accuracy index, a precision index, a recall index, and a comprehensive index of precision and recall.

[0019] According to a third aspect of the present disclosure, a tool capability classification device is provided, comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to execute the method described above based on instructions stored in the memory.

[0020] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the instructions are executed by a processor to perform the method described above.

[0021] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein the computer program implements the steps of the method described above when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or related technologies, a brief introduction to the drawings required for use in the embodiments or related technology descriptions will be given below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor. In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or related technologies, a brief introduction to the drawings required for use in the embodiments or related technology descriptions will be given below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0023] FIG1 is a flow chart of some embodiments of a tool capability classification method according to the present disclosure;

[0024] FIG2 is a schematic diagram of a process for obtaining a first feature vector in some embodiments of the tool capability classification method according to the present disclosure;

[0025] FIG3 is a schematic diagram of an application of some embodiments of the tool capability classification method according to the present disclosure;

[0026] Figure 4 is a schematic diagram of a multi-scale capsule network module;

[0027] FIG5 is a module diagram of some embodiments of a tool capability classification device according to the present disclosure;

[0028] FIG6 is a schematic diagram of modules of other embodiments of a tool capability classification device according to the present disclosure;

[0029] FIG7 is a schematic diagram of a first feature mining module in some embodiments of the tool capability classification device according to the present disclosure;

[0030] FIG8 is a module diagram of still other embodiments of the tool capability classification device according to the present disclosure. DETAILED DESCRIPTION

[0031] Exemplary embodiments of the present disclosure will be described below with reference to the accompanying drawings. For the sake of clarity and conciseness, not all features of the embodiments are described in the specification. However, it should be understood that many implementation-specific settings must be made in the process of implementing the embodiments in order to achieve the developer's specific goals, such as meeting those restrictions related to equipment and services, and these restrictions may vary depending on the implementation. In addition, it should be understood that although the development work may be very complex and time-consuming, it is only a routine task for those skilled in the art who benefit from the contents of this disclosure.

[0032] It should be noted that the relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

[0033] Those skilled in the art will understand that the terms "first" and "second" in the embodiments of the present disclosure are only used to distinguish different steps, devices or modules, and do not represent any specific technical meanings, nor do they indicate a necessary logical order between them.

[0034] It should also be understood that in the embodiments of the present disclosure, “a plurality of” may refer to two or more than two, and “at least one” may refer to one, two, or more than two.

[0035] It should also be understood that any component, data or structure mentioned in the embodiments of the present disclosure can generally be understood as one or more, unless explicitly limited or otherwise indicated in the context.

[0036] In addition, the term "and / or" in this disclosure is merely a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this disclosure generally indicates that the related objects are in an "or" relationship.

[0037] It should also be understood that the description of the various embodiments in this disclosure focuses on the differences between the various embodiments, and the same or similar aspects thereof can be referenced with each other. For the sake of brevity, they will not be described one by one.

[0038] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0039] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0040] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0041] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0042] In addition, in order to avoid obscuring the present disclosure with unnecessary details, only the processing steps and / or device structures that are closely related to at least the solution according to the present disclosure are shown in the drawings, while other details that are not closely related to the present disclosure are omitted. It should also be noted that similar reference numerals and letters in the drawings indicate similar items, and therefore once an item is defined in one drawing, it does not need to be discussed again for subsequent drawings.

[0043] In the related art known to the inventors, classification models typically use neural network models, which are capable of extracting feature information from massive amounts of data. These neural network models are typically convolutional neural network (CNN) models, which lack the ability to deeply extract feature data. Consequently, these classification methods or models often have low accuracy.

[0044] In view of this, a technical problem to be solved by the present disclosure is to provide a tool capability classification method, which uses a multi-scale capsule network module to perform multi-scale feature extraction processing, and uses a sparse coding module to obtain feature vectors, and obtains tool capability classification result information based on the feature vectors through a neural network model; by introducing sparse coding rules into the capsule network to classify tool capabilities, the training iteration time can be reduced, the ability to extract local and global features of text data is improved, and the accuracy of classification is improved.

[0045] FIG1 is a flow chart of some embodiments of the tool capability classification method according to the present disclosure, as shown in FIG1 :

[0046] Step 101: Use a multi-scale capsule network module to perform multi-scale feature extraction processing on the tool capability description text information to obtain a first feature vector.

[0047] In some embodiments, the tool can be a variety of tools, such as AI tools, etc.; AI tools are software tools that use artificial intelligence technology, etc., and AI tools can be used for intelligent processing such as demand assessment, resource allocation, and electricity bill management.

[0048] Capsule Networks (CapsNets) can better model the hierarchical relationships within neural networks. They use vectors to encode information and output the probability of data presence, thereby predicting the probability of data features. The multi-scale Capsule Network module can extract information from different temporal and spatial domains. This module uses multi-scale feature extraction to process textual descriptions of tool capabilities. By using this method to mine multi-dimensional feature relationships in data, the number of parameters can be reduced while increasing feature expressiveness.

[0049] Step 102: Obtain a second eigenvector based on the first eigenvector using a sparse coding module.

[0050] In some embodiments, the sparse coding algorithm is an unsupervised learning method. The sparse coding module may apply a related sparse coding algorithm to process the first feature vector and output a second feature vector, where the second feature vector is a sparse vector.

[0051] Step 103: Obtain capability classification result information of the tool through the neural network model based on the second eigenvector.

[0052] In some embodiments, multiple neural network models can be used to process the second eigenvector to obtain tool capability classification information. The classification information of the AI ​​tool can be used to indicate the category to which the AI ​​tool belongs. The categories can include demand assessment, resource allocation, electricity cost management, and other categories in multiple scenarios. The multiple scenarios can include cloud network planning, service provisioning, network and service quality optimization, and other scenarios.

[0053] Based on the capability classification results of AI tools, an intelligent asset warehouse can be established to provide a unified AI capability view, which can quickly match the needs of AI tool capabilities, solve the fragmentation problem of AI tool capabilities, improve the reuse rate of AI tool capabilities, and reduce development costs.

[0054] Related convolutional neural networks lack the ability to deeply extract and enhance feature data, and features cannot be reused at multiple levels. Using a sparse coding module can use "super-complete" basis vectors to represent the original data, increasing the ability to capture features and making the encoded data present a super-Gaussian distribution. The multi-scale capsule network module can reduce the number of parameters and increase the expressiveness of features. The tool capability classification method disclosed in the present invention can use a multi-scale capsule network module to perform multi-scale feature extraction processing on the tool capability description text information and use a sparse coding module to obtain feature vectors. By introducing sparse coding rules into the capsule network to classify tool capabilities, the training iteration time can be reduced, the ability to extract local and global features of text data is improved, and the accuracy of classification is improved.

[0055] FIG2 is a schematic diagram of a process for obtaining a first feature vector in some embodiments of the tool capability classification method according to the present disclosure, as shown in FIG2 :

[0056] In step 201 , a multi-scale capsule network module is used and convolution kernels of multiple convolution templates are used to perform multi-scale feature extraction processing on the capability description text information to obtain multiple feature information.

[0057] In some embodiments, as shown in FIG3 , the multi-scale capsule network module and the sparse coding module constitute the feature fine classification part; the neural network model is the feature coarse classification part, and the neural network model further captures local text semantic information and global text semantic information.

[0058] A convolution kernel of a single scale can only extract feature information with a single score. The capability description text is input into the multi-scale capsule network module, which includes a convolutional layer (not shown). The convolutional layer presets multiple convolution templates with different convolution scales. By using convolution kernels from multiple convolution templates in the convolutional layer, multi-scale feature extraction can be performed on the capability description text, obtaining multiple feature information, namely, multiple feature vectors.

[0059] Step 202: Encode multiple feature information using a capsule network module to obtain a first feature vector.

[0060] In some embodiments, as shown in FIG4 , the multi-scale capsule network module includes multiple low-level capsules 41 and high-level capsules 42 . Multiple feature information is processed using the multiple low-level capsules 41 to obtain multiple low-level capsule feature information. Multiple prediction vectors are obtained based on the multiple low-level capsule feature information and the corresponding affine matrices.

[0061] The weighted summation of multiple prediction vectors is performed based on the coupling coefficient to obtain an input vector. The input vector is compressed using an activation function, and the compressed result is input to the high-level capsule 42. The high-level capsule 42 outputs a first eigenvector. That is, the high-level capsule 42 processes the compressed result to obtain the first eigenvector. The coupling coefficient can be updated based on the first eigenvector using a dynamic routing algorithm.

[0062] For example, the capability information of AI tools in various provinces of a company is summarized to generate textual descriptions of the capability of the AI ​​tools. The textual descriptions of the capability include: AI tool capability descriptions, indicators and other information.

[0063] The capability description text information is input into the multi-scale capsule network module. The convolutional layer of the multi-scale capsule network module uses convolution kernels from multiple convolution templates to perform multi-scale feature extraction on the capability description text information, obtaining multiple feature information. Multiple lower-level capsules 41 process the multiple feature information to obtain multiple lower-level capsule feature information u. The multiple lower-level capsule feature information is multiple lower-level capsule feature vectors.

[0064] The multi-scale capsule network module converts all neurons into directional vector outputs. The length of the vector represents the probability of the target's presence, and the direction of the vector represents the target's characteristics. Finally, the information is encapsulated and output in the form of capsules. During the conversion of low-level capsules to high-level capsules, scalar nonlinear processing can be performed to establish correspondences between different capsules. During this process, weights are updated using a dynamic routing algorithm.

[0065] Multiply multiple low-level capsule feature information and the corresponding affine matrix to obtain multiple prediction vectors. The calculation formula is as follows: j|i =W ij u i (1);

[0066] Among them, u i Characterize the low-level capsule feature information output by the low-level capsule, i = 1, 2, ..., n, n is the number of low-level capsules; W ij Affine matrix representing positional relationship; u j|i Represents the prediction vector of the output.

[0067] Perform weighted summation on multiple prediction vectors according to the coupling coefficient to obtain the input vector S j , the calculation formula is as follows: j =Σ i c ij u j|i (2);

[0068] Among them, cij is the coupling coefficient, satisfying Σ i c ij =1;c ij By updating the factor b ij Obtained, c ij The update formula is as follows:

[0069] Use the activation function to compress the input vector and dynamically compress the input vector to the range of (0, 1). The activation function can be a nonlinear activation function such as the squash function. The compression result is input into the high-level capsule, and the high-level capsule outputs the first eigenvector. The calculation formula is as follows:

[0070] Among them, v j is the first eigenvector, j is the first eigenvector and the number of high-level capsules.

[0071] Based on the first eigenvector and using the dynamic routing algorithm to update the coupling coefficient. For example, using the dynamic routing algorithm to update b ij with c ij , the formula is as follows:

[0072] b ij =b ij +v j u j|i (5);

[0073] Among them, b ij Represents the logarithmic prior probability, and the initial value is set to 0.

[0074] In some embodiments, a sparse coding model is provided within the sparse coding module. The sparse coding model of the sparse coding module is used to perform low-dimensional spatial feature extraction processing on the first eigenvector based on the sparse coding matrix to obtain the second eigenvector. The sparse coding matrix contains fewer non-zero entries, which has higher computational efficiency and better classification effect. The sparse coding algorithm applied by the sparse coding model is:

[0075] Where X is the first eigenvector of the input; A i is the convolution feature corresponding to the i-th filter; λ is the regularization parameter; D is the circulant matrix, that is, the sparse coding matrix.

[0076] The sparse coding module can be used to express the information of low-dimensional space in the form of coding, and the sparsity of features can be used to represent the potential low-dimensional space of the target text, so as to extract more representative text features.

[0077] In some embodiments, as shown in FIG3 , the neural network model includes a convolutional layer, a pooling layer, and a softmax classifier. The convolutional and pooling layers are used to extract and fuse the second feature vector, and the processing results are input into the softmax classifier. Based on the output of the softmax classifier, capability classification result information is obtained. This capability classification result information can be used to map the content of the capability description text of an AI tool, etc., to multiple capability categories.

[0078] In some embodiments, various methods can be used to train the multi-scale capsule network module, sparse coding module, and neural network model. For example, training text describing the capabilities of a tool can be obtained, annotated, a training set constructed, and a loss function constructed. Iterative training is then performed, using the multi-scale capsule network module to extract multi-scale features from the training text describing the capabilities in the training set to obtain a first feature vector. This first feature vector is then input into the sparse coding module to obtain a second feature vector, which is then processed using the neural network model to obtain capability classification results.

[0079] Based on the ability classification result information and the annotation information of the ability description training text information in the training set, the value of the loss function is determined, and the parameters of the multi-scale capsule network module, sparse coding module and neural network model are adjusted based on the value of the loss function; when the value of the loss function is less than the preset loss threshold, the training is determined to be completed, and the parameters of the multi-scale capsule network module, sparse coding module and neural network model are determined.

[0080] In some embodiments, evaluation index information is determined based on the capability classification result information, and the classification result is evaluated according to the evaluation index information. The evaluation index includes at least one of the following: accuracy index, precision index, recall index, comprehensive index of precision and recall, and other indicators.

[0081] For example, evaluation metrics include: Accuracy (A), Precision (P), Recall (R), and F1 value, where F1 value represents a comprehensive index of accuracy and recall. The formulas for calculating the values ​​of accuracy, precision, recall, and the combined index of accuracy and recall are as follows:

[0082] Among them, TP represents the number of samples that are both predicted and actual as positive, TN represents the number of samples that are both predicted and actual as negative; FN represents the number of samples that are actually positive and predicted as negative, and FP represents the number of samples that are actually negative and predicted as positive.

[0083] The datasets used for evaluation are shown in Table 1 below:

[0084] Table 1 - Datasets used for evaluation

[0085] In some embodiments, the tool capability classification method disclosed herein can be compared with several other algorithms to verify the classification accuracy of the tool capability description text information of the tool capability classification method disclosed herein. Several other algorithms include: Bert (Bidirectional Encoder Representations from Transformers, a bidirectional encoding representation algorithm based on the Transformer algorithm) algorithm, TextCNN (Text Convolutional Neural Networks, text convolutional neural network) algorithm, LSTM (Long Short-Term Memory, long short-term memory) algorithm, etc.; among them, the Bert algorithm has huge parameters and is pre-trained on large-scale semantic text, and its learning ability is insufficient; the TextCNN algorithm has weak model interpretability, and the convolution operation can only capture a single convolution kernel size, and its classification effect is poor; the LSTM algorithm has poor parallel data processing effect and is limited to processing longer sequences of data.

[0086] Based on the data set in Table 1 above, tests were performed using the Bert algorithm, TextCNN algorithm, LSTM algorithm, and the tool capability classification method disclosed in this disclosure. The experimental results are shown in Table 2 below.

[0087] Table 2 - Comparison of verification results of various classification methods

[0088] As can be seen from Table 2 above, the tool capability classification method disclosed in the present invention has the highest accuracy. Compared with algorithms such as BERT, TextCNN, and LSTM, the classification accuracy is improved by 5.8%, 2.6%, and 1.8%, respectively. This shows that the tool capability classification method disclosed in the present invention has higher classification accuracy for tool capability description text information.

[0089] In some embodiments, as shown in FIG5 , the present disclosure provides a tool capability classification device 50, comprising a first feature mining module 51, a second feature mining module 52, and a classification result acquisition module 53. The first feature mining module 51 utilizes a multi-scale capsule network module to perform multi-scale feature extraction on the tool capability description text to obtain a first feature vector. The second feature mining module 52 utilizes a sparse coding module and obtains a second feature vector based on the first feature vector. The classification result acquisition module 53 utilizes a neural network model and obtains the tool capability classification result information based on the second feature vector.

[0090] In some embodiments, the second feature mining module 52 utilizes the sparse coding model of the sparse coding module and performs low-dimensional spatial feature extraction processing on the first feature vector based on the sparse coding matrix to obtain the second feature vector.

[0091] The neural network model includes a convolution layer, a pooling layer, and a softmax classifier, etc.; the classification result acquisition module 53 uses the convolution layer and the pooling layer to extract and fuse the second feature vector, and inputs the processing result into the softmax classifier; the classification result acquisition module 53 obtains the ability classification result information based on the result output by the softmax classifier.

[0092] As shown in FIG6 , the tool capability classification device 50 includes a classification evaluation module 54 . The classification evaluation module 54 determines evaluation index information based on capability classification result information and evaluates the classification result according to the evaluation index information.

[0093] As shown in Figure 7, the first feature mining module 51 includes a feature extraction unit 511 and a feature encoding unit 512. The feature extraction unit 511 utilizes a multi-scale capsule network module and convolution kernels from multiple convolution templates to perform multi-scale feature extraction on the capability description text information, obtaining multiple feature information. The feature encoding unit 512 utilizes the capsule network module to encode the multiple feature information to obtain a first feature vector.

[0094] The multi-scale capsule network module includes multiple low-level capsules and high-level capsules; the feature encoding unit 512 uses the multiple low-level capsules to process multiple feature information to obtain multiple low-level capsule feature information; the feature encoding unit 512 obtains multiple prediction vectors based on the multiple low-level capsule feature information and the corresponding affine matrix.

[0095] The feature encoding unit 512 performs weighted summation processing on multiple prediction vectors according to the coupling coefficient to obtain an input vector; the feature encoding unit 512 uses an activation function to compress the input vector, and inputs the compression processing result into a high-level capsule, so that the high-level capsule outputs a first feature vector based on the compression processing result, wherein the feature encoding unit 512 updates the coupling coefficient based on the first feature vector and using a dynamic routing algorithm.

[0096] In some embodiments, as shown in FIG8 , the present disclosure provides a tool capability classification apparatus, which may include a memory 62, a processor 61, a communication interface 63, and a bus 64. The memory 62 is used to store instructions, and the processor 61 is coupled to the memory 62. The processor 61 is configured to execute the aforementioned tool capability classification method based on the instructions stored in the memory 62.

[0097] Memory 62 can be high-speed RAM, non-volatile memory, or a memory array. Memory 62 can also be divided into blocks, and the blocks can be combined into virtual volumes according to certain rules. Processor 61 can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the tool capability classification method disclosed herein.

[0098] In some embodiments, the present disclosure provides a computer-readable storage medium storing computer instructions. When the instructions are executed by a processor, the method in any of the above embodiments is implemented.

[0099] Computer readable storage media can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive enumeration) of readable storage media can include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0100] The embodiments of the present disclosure may also be a computer program product, which includes computer program instructions. When the computer program instructions are executed by a processor, the processor executes the steps of the method according to various embodiments of the present disclosure described in the above “Exemplary Method” section of this specification.

[0101] The tool capability classification method, device, storage medium and computer program product in the above-mentioned embodiments use a multi-scale capsule network module to perform multi-scale feature extraction processing, and use a sparse coding module to obtain feature vectors, and obtain tool capability classification result information based on the feature vectors through a neural network model; by introducing sparse coding rules into the capsule network to classify tool capabilities, the training iteration time can be reduced, the ability to extract local and global features of text data is improved, the classification accuracy is improved, and the user experience can be improved.

[0102] The basic principles of the present disclosure have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this disclosure are merely illustrative and not restrictive, and should not be construed as necessarily possessed by each embodiment of the present disclosure. Furthermore, the specific details disclosed above are provided for illustrative purposes and to facilitate understanding, rather than as limitations. These details do not limit the present disclosure to necessarily being implemented using these specific details.

[0103] Each embodiment in this specification is described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. References to the same or similar parts between the various embodiments are sufficient. For system embodiments, since they largely correspond to method embodiments, their description is relatively simple. For relevant parts, references to the description of the method embodiments are sufficient.

[0104] The block diagrams of the devices, devices, equipment, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "include," "comprise," "have," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0105] It should also be noted that in the apparatus, device, and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0106] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0107] The above description has been provided for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, it will be understood by those skilled in the art that the above embodiments are merely illustrative and do not limit the scope of the present disclosure. It will be understood by those skilled in the art that the above embodiments may be combined, modified, or replaced without departing from the scope and essence of the present disclosure.

Claims

1. A tool capability classification method, comprising: Use the multi-scale capsule network module to perform multi-scale feature extraction on the tool capability description text information to obtain the first feature vector; Obtaining the second eigenvector based on the first eigenvector using a sparse coding module; Capability classification result information of the tool is obtained through a neural network model and based on the second feature vector.

2. The tool capability classification method according to claim 1, wherein: The method of performing multi-scale feature extraction processing on the tool capability description text information using the multi-scale capsule network module to obtain the first feature vector includes: Utilizing the multi-scale capsule network module and using convolution kernels of multiple convolution templates, performing multi-scale feature extraction processing on the capability description text information to obtain multiple feature information; The capsule network module is used to encode the plurality of feature information to obtain the first feature vector.

3. The tool capability classification method according to claim 2, wherein: The multi-scale capsule network module includes: a plurality of low-level capsules and a plurality of high-level capsules; and encoding the plurality of feature information using the capsule network module to obtain the first feature vector includes: Processing the plurality of feature information using the plurality of lower-level capsules to obtain a plurality of lower-level capsule feature information; Obtaining a plurality of prediction vectors based on the plurality of low-level capsule feature information and corresponding affine matrices; Performing weighted summation processing on the multiple prediction vectors according to the coupling coefficient to obtain an input vector; compressing the input vector using an activation function, and inputting the compression result into the high-level capsule, so that the high-level capsule outputs the first feature vector based on the compression result; The coupling coefficient is updated based on the first eigenvector and using a dynamic routing algorithm.

4. The tool capability classification method according to any one of claims 1 to 3, wherein: The obtaining the second eigenvector by using the sparse coding module and based on the first eigenvector includes: The sparse coding model of the sparse coding module is used and based on a sparse coding matrix, low-dimensional spatial feature extraction processing is performed on the first eigenvector to obtain the second eigenvector.

5. The tool capability classification method according to any one of claims 1 to 4, wherein: The neural network model includes: a convolution layer, a pooling layer and a softmax classifier; the processing of the second feature vector by the neural network model to obtain a classification result for the capability data includes: The convolution layer and the pooling layer are used to extract and fuse the second feature vector. The processing result is input into the softmax classifier; According to the result output by the softmax classifier, the capability classification result information is obtained.

6. The tool capability classification method according to any one of claims 1 to 5, further comprising: Determining evaluation indicator information based on the capability classification result information; Evaluate the classification results according to the evaluation index information; The evaluation index includes at least one of the following: an accuracy index, a precision index, a recall index, and a comprehensive index of precision and recall.

7. The tool capability classification method according to any one of claims 1 to 6, wherein: Said tools include: artificial intelligence AI tools; The capability description text information of the tool includes: capability description text information of the AI tool.

8. A tool capability classification device, comprising: A first feature mining module is used to perform multi-scale feature extraction processing on the tool capability description text information using a multi-scale capsule network module to obtain a first feature vector; A second feature mining module, configured to obtain the second feature vector based on the first feature vector using a sparse coding module; The classification result obtaining module is used to obtain capability classification result information of the tool based on the second feature vector through a neural network model.

9. The tool capability classification device according to claim 8, wherein: The first feature mining module includes: A feature extraction unit is configured to perform multi-scale feature extraction processing on the capability description text information using the multi-scale capsule network module and convolution kernels of multiple convolution templates to obtain multiple feature information; A feature encoding unit is used to encode the multiple feature information using the capsule network module to obtain the first feature vector.

10. The tool capability classification device according to claim 9, wherein: The multi-scale capsule network module includes: multiple low-level capsules and high-level capsules; The feature encoding unit is configured to process the plurality of feature information using the plurality of low-level capsules to obtain a plurality of low-level capsule feature information; obtain a plurality of prediction vectors based on the plurality of low-level capsule feature information and corresponding affine matrices; perform weighted summation processing on the plurality of prediction vectors according to the coupling coefficient to obtain an input vector; compress the input vector using an activation function, and input the compressed processing result into the high-level capsule, so that the high-level capsule outputs the first feature vector based on the compressed processing result; wherein, based on the first feature vector and using The coupling coefficient is updated using a dynamic routing algorithm.

11. The tool capability classification device according to any one of claims 8 to 10, wherein: The second feature mining module is used to utilize the sparse coding model of the sparse coding module and based on the sparse coding matrix to perform low-dimensional spatial feature extraction processing on the first feature vector to obtain the second feature vector.

12. The tool capability classification device according to any one of claims 8 to 11, wherein: The neural network model includes: a convolutional layer, a pooling layer and a softmax classifier; The classification result acquisition module is used to use the convolution layer and the pooling layer to perform feature extraction and fusion processing on the second feature vector, and input the processing result into the softmax classifier; according to the result output by the softmax classifier, the ability classification result information is obtained.

13. The tool capability classification device according to any one of claims 8 to 12, further comprising: A classification evaluation module is used to determine evaluation index information based on the capability classification result information; and evaluate the classification result according to the evaluation index information; wherein the evaluation index includes at least one of the following: an accuracy index, a precision index, a recall index, and a comprehensive index of precision and recall.

14. A tool capability classification device, comprising: Memory; and a processor coupled to the memory, wherein the processor is configured to execute the method according to any one of claims 1 to 7 based on instructions stored in the memory. 15 . A computer-readable storage medium storing computer instructions, wherein the instructions are executed by a processor to execute the method according to claim 1 .

16. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images