A method, product, and apparatus for determining data categories

Abstract

This invention provides a method, model, and apparatus for determining data categories. The method includes: performing feature interaction between the feature information of target data and pre-stored feature information of each sample to obtain interactive feature information, wherein the interactive feature information is used to characterize the similarities and differences between the feature information of the target data and the feature information of each sample; and determining the category of the target data based on the feature information of the target data and the interactive feature information. In determining the category of the target data, multiple preset sample feature information can be referenced, comprehensively considering the similarities and differences between the feature information of the target data and the sample feature information, thereby improving the accuracy of the classification results.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, product, and apparatus for determining data categories. Background Technology

[0002] With the development of computer technology, deep learning network models are widely used in various industries. For example, image classification network models and text classification network models are used for data processing to classify work orders.

[0003] However, both image and text classification network models rely on training data to train the model and then use the trained model parameters for inference to classify work orders. This is essentially memorization-based learning. During testing and inference, often only the trained model parameters are used, without utilizing the training set. For large work order systems with vast amounts of diverse data, existing classification methods, which depend entirely on training data, cannot effectively handle newly added work order data with varying descriptions, resulting in poor classification accuracy.

[0004] It is evident that existing classification methods suffer from poor accuracy. Summary of the Invention

[0005] This invention provides a data category determination method, product, and apparatus to address the problem of poor accuracy in existing classification methods.

[0006] This invention provides a method for determining data categories, the method comprising:

[0007] The feature information of the target data is interacted with the feature information of each pre-stored sample to obtain interactive feature information, which is used to characterize the similarities and differences between the feature information of the target data and the feature information of each sample.

[0008] The category of the target data is determined based on the feature information of the target data and the interaction feature information.

[0009] Optionally, the feature information of the target data includes a first feature matrix and a first label vector, and the feature information of each sample includes a second feature matrix and a second label vector;

[0010] The step of performing feature interaction between the feature information of the target data and the pre-stored feature information of each sample to obtain interactive feature information includes:

[0011] The first feature matrix is ​​interacted with the second feature matrix corresponding to each sample feature information to obtain an interaction matrix. The interaction matrix is ​​concatenated with an interaction label vector, which is a label vector obtained by performing label interaction between the first label vector and the second label vector corresponding to each sample feature information.

[0012] Determining the category of the target data based on the feature information and interaction feature information of the target data includes:

[0013] The interaction matrix and the first feature matrix are combined to obtain a fusion matrix;

[0014] The category of the target data is determined based on the fusion matrix.

[0015] Optionally, determining the category of the target data based on the fusion matrix includes:

[0016] The fusion matrix is ​​pooled to obtain a pooled fusion matrix;

[0017] The category of the target data is determined based on the pooled fusion matrix.

[0018] Optionally, before performing feature interaction between the feature information of the target data and the pre-stored feature information of each sample to obtain interactive feature information, the method further includes:

[0019] Feature extraction is performed on multiple sample data to obtain sample feature information corresponding to the multiple sample data;

[0020] The sample feature information is stored.

[0021] Optionally, after determining the category of the target data based on the feature information and the interaction feature information of the target data, the method further includes:

[0022] Under certain preset conditions, the feature information of the target data is stored as sample feature information, wherein the preset conditions include any one of the following:

[0023] The sample feature information does not include the feature information of the target data;

[0024] Upon re-examination of the target data categories, classification errors were found.

[0025] This invention also provides a data category determination product for implementing the above-described data category determination method, the product comprising:

[0026] The feature extraction module includes a first feature extractor and a first label submodule. The first feature extractor is used to extract features from the target data to obtain a first feature matrix of the target data. The first feature matrix includes a first label vector of the target data. The first label submodule is used to mask the first label vector.

[0027] The memory module stores sample feature information of multiple sample data. The sample feature information includes a second feature matrix and a second label vector. The memory includes a second feature extractor and a second label submodule. The second feature extractor is used to extract features from each sample data to obtain a second feature matrix for each sample data. The second feature matrix includes a corresponding second label vector. The second label submodule is used to obtain the second label vector.

[0028] An interaction module is used to perform feature interaction between the feature information of the target data and the pre-stored feature information of each sample to obtain interactive feature information. The interactive feature information is used to characterize the similarities and differences between the feature information of the target data and the feature information of each sample.

[0029] The output module is used to determine the category of the target data based on the feature information and the interaction feature information of the target data.

[0030] This invention also provides a data category determination device, comprising:

[0031] An interaction module is used to perform feature interaction between the feature information of the target data and the pre-stored feature information of each sample to obtain interactive feature information. The interactive feature information is used to characterize the similarities and differences between the feature information of the target data and the feature information of each sample.

[0032] The determination module is used to determine the category of the target data based on the feature information and the interaction feature information of the target data.

[0033] Optionally, the feature information of the target data includes a first feature matrix and a first label vector, and the feature information of each sample includes a second feature matrix and a second label vector;

[0034] The interaction module includes:

[0035] The first interaction submodule is used to perform feature interaction between the first feature matrix and the second feature matrix corresponding to each sample feature information to obtain an interaction matrix. The interaction matrix is ​​concatenated with an interaction label vector, which is a label vector obtained by performing label interaction between the first label vector and the second label vector corresponding to each sample feature information.

[0036] The determining module includes:

[0037] The merging submodule is used to merge the interaction matrix with the first feature matrix to obtain a fusion matrix;

[0038] A determination submodule is used to determine the category of the target data based on the fusion matrix.

[0039] Optionally, the determining submodule includes:

[0040] A pooling unit is used to pool the fusion matrix to obtain a pooled fusion matrix;

[0041] The determining unit is used to determine the category of the target data based on the pooled fusion matrix.

[0042] Optionally, the device further includes:

[0043] The extraction module is used to extract features from multiple sample data to obtain sample feature information corresponding to the multiple sample data.

[0044] The first storage module is used to store the sample feature information.

[0045] In this embodiment of the invention, feature interaction is performed between the feature information of the target data and the feature information of each pre-stored sample to capture the similarities and differences between the feature information of the sample and the feature information of the target data. Then, the obtained interactive feature information is combined with the feature information of the target data. In this way, when determining the category of the target data, multiple preset sample feature information can be referenced, and the similarities and differences between the feature information of the target data and the feature information of the sample can be comprehensively considered, thereby improving the accuracy of the classification results. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is one of the flowcharts illustrating the data category determination method provided in this embodiment of the invention;

[0048] Figure 2 This is a schematic diagram of the feature extraction module provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the data stored in the memory module provided in an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of the structure of the memory module provided in an embodiment of the present invention;

[0051] Figure 5 This is a second flowchart illustrating the data category determination method provided in this embodiment of the invention;

[0052] Figure 6 This is a schematic diagram of the interaction method in the interaction module provided in the embodiment of the present invention;

[0053] Figure 7 This is a schematic diagram illustrating the structure of a product determined by data categories according to an embodiment of the present invention;

[0054] Figure 8 This is a schematic diagram of the data category determination device provided in an embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The terms "first," "second," etc., used in the specification and claims of this invention are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such usage can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, the first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0057] Please see Figure 1 , Figure 1 This is one of the flowcharts illustrating the data category determination method provided in this embodiment of the invention. For example... Figure 1 As shown, an embodiment of the present invention provides a data category determination method comprising the following steps:

[0058] Step 101: Perform feature interaction between the feature information of the target data and the feature information of each sample that has been stored in advance to obtain interactive feature information. The interactive feature information is used to characterize the similarities and differences between the feature information of the target data and the feature information of each sample.

[0059] In applications that classify work order data in a work order system, the target data can be text-based work order data, image-based work order data, etc. Feature information of the target data can be obtained using deep learning models (e.g., Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), Transformers, etc.). For Natural Language Processing (NLP), pre-trained language models such as BERT and Roberta are supported; for image data, image classification models such as ResNet, ImageNet, and GoogleNet are supported.

[0060] Feature extraction is performed on the target data to obtain its feature information. This feature information can include a first label vector, which can be concatenated onto a mask label. The mask label is a learnable embedding that facilitates subsequent learning and generalization. Sample feature information can include the label embedding of cluster centers or randomly selected label embeddings from different types. After obtaining the feature information of the target data, SelfAttention is used to perform feature interaction between the target data feature information and the sample feature information to determine the interaction feature information. Based on the interaction feature information, the similarities and differences between the target data feature information and the sample feature information can be determined.

[0061] Further, according to step 102, the features of the target data and the features after the interaction between the target data and the sample data are combined to improve the accuracy of classifying the target data.

[0062] Step 102: Determine the category of the target data based on the feature information and the interaction feature information of the target data.

[0063] Based on the mapping relationship between the features of the target data and the features of each sample data, the similarities and differences between the target data and the sample data are determined, i.e., the interaction feature information is obtained. Then, the feature information of the target data is combined with the interaction feature information. In this way, for a work order system with a large amount of data and many types, after extracting features from the work order data to be classified (i.e., the target data), the classification results are improved by referring to the pre-stored feature information of each sample when determining the category of the work order data based on the feature information of the work order data.

[0064] In this embodiment of the invention, feature interaction is performed between the feature information of the target data and the feature information of each pre-stored sample to capture the similarities and differences between the feature information of the sample and the feature information of the target data. Then, the obtained interactive feature information is combined with the feature information of the target data. In this way, when determining the category of the target data, multiple preset sample feature information can be referenced, and the similarities and differences between the feature information of the target data and the feature information of the sample can be comprehensively considered, thereby improving the accuracy of the classification results.

[0065] The target data can be one or more datasets, which improves the efficiency of data classification and is better suited for situations where the work order system has a large amount of data and many types.

[0066] In some optional implementations, the feature information of the target data includes a first feature matrix and a first label vector, and the feature information of each sample includes a second feature matrix and a second label vector;

[0067] Step 101 involves performing feature interaction between the feature information of the target data and the pre-stored feature information of each sample to obtain interactive feature information, including:

[0068] The first feature matrix is ​​interacted with the second feature matrix corresponding to each sample feature information to obtain an interaction matrix. The interaction matrix is ​​concatenated with an interaction label vector, which is a label vector obtained by performing label interaction between the first label vector and the second label vector corresponding to each sample feature information.

[0069] Step 102, determining the category of the target data based on the feature information and interaction feature information of the target data, includes:

[0070] The interaction matrix and the first feature matrix are combined to obtain a fusion matrix;

[0071] The category of the target data is determined based on the fusion matrix.

[0072] Specifically, the feature extraction module can extract features from the target data to obtain a first feature matrix and a first label vector. The feature extraction module may include a first feature extractor and a first label submodule, such as... Figure 2As shown, the first feature extractor is used to extract features from the target data, obtaining a first feature matrix of the target data. The first feature matrix includes the first label vector of the target data. The first feature extractor can support mainstream deep learning models, such as Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and Transformers. Preferably, the first feature extractor can be a Transformer structure.

[0073] The first feature matrix can have dimensions L*M. M can be the same number as the feature vectors of the target data; for example, if the word vectors of the target data in the BERT model for text processing are 768, then M can be 768. L can be the total number of categories plus 1; for example, if there are 7 labels for the work order categories to be identified, then L can be 8. The extra bit in L is reserved as a mask label. The first label vector of the target data can be concatenated onto the mask label. During feature interaction, the first label vector can be masked through the first label submodule, so that it has no effect during parameter updates.

[0074] In one example, before step 101, which describes performing feature interaction between the feature information of the target data and the pre-stored feature information of each sample to obtain the interactive feature information, the method further includes:

[0075] Feature extraction is performed on multiple sample data to obtain sample feature information corresponding to the multiple sample data;

[0076] The sample feature information is stored.

[0077] Because the work order system contains a wide variety of work order data types and differs in dialects across regions, resulting in diverse text descriptions, sample feature information from multiple sample data sets can be pre-stored in the memory module. Specifically, the number of labels is determined based on the required classification precision. K data points are randomly selected as samples within each label. The extracted feature information for each sample data set is stored in the memory module, pre-storing multiple sample feature information sets and increasing the diversity of sample feature information. Each type of sample feature information can be labeled. The number of sample data sets in each label can be consistent or inconsistent. For example, with numerous classifications, the number of labels can be increased, or the number of feature information sets within the same label can be increased. The sample labels corresponding to each type of sample feature information are as follows: Figure 3As shown. Sample feature information may include a second feature matrix and a second label vector. The features of the sample data can be expressed in the form of the second feature matrix, and the second label vector can be expressed in the form of a vector of sample labels, so as to facilitate subsequent feature interaction based on the first feature matrix and the second feature matrix.

[0078] Among them, such as Figure 4 As shown, the memory module may include a second feature extractor and a second label submodule. The second label submodule obtains a second label vector. The second feature extractor extracts features from each sample data to obtain a second feature matrix for each sample data. The second feature matrix includes the corresponding second label vector. The feature extraction method in the second feature extractor can be the same as that in the feature extraction module; that is, the second feature extractor can also support mainstream deep learning models, such as Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and Transformers. Preferably, the second feature extractor can be a Transformer structure. By sharing model parameters, the feature representation (i.e., sample feature information) of each sample data is obtained.

[0079] In this embodiment, such as Figure 5 As shown, after feature extraction from the target data by the feature extraction module, a first feature matrix of the target data is obtained. The dimension of the first feature matrix can be L*M. M can be the same number as the number of feature vectors of the target data, and L can be the total number of categories + 1. The extra bit in L is reserved as a mask label. The first label vector of the target data can be concatenated onto the mask label. During feature interaction, the first label vector can be masked by the first label submodule so that it has no effect when updating parameters. At the same time, the memory module pre-stores sample feature information of multiple sample data (including a second feature matrix and a second label vector). The dimension of the second feature matrix can be the same as the dimension of the first feature matrix, and the second label vector can be a label embedding vector, which is concatenated onto the second feature matrix.

[0080] Then, in the interaction module, the first feature matrix is ​​interacted with the second feature matrix corresponding to each sample feature information to obtain an interaction matrix. The interaction matrix is ​​concatenated with an interaction label vector, which can be a label vector obtained by interacting the first label vector with the second label vector corresponding to each sample feature information. The interaction matrix captures the similarities and differences between the feature information of the target data and the target data. In the feature merging module, a fusion matrix is ​​generated based on the similarities and differences in the first feature matrix and the interaction matrix. During the interaction between the first and second feature matrices, the first feature vector and the second feature vector also simultaneously interact with each other. That is, the first feature vector learns the similarities and differences between the feature information of the target data and the sample feature information during the interaction process. The generated interaction labels include these similarities and differences. Therefore, when merging the interaction matrix and the first feature matrix to obtain the fusion matrix, the similarities and differences between the feature information of the target data and the sample feature information are comprehensively considered. The category of the target data is determined based on the fusion matrix, improving the accuracy of the classification results.

[0081] The interaction module can adopt the structure of the encoder in Transformer. A single Transformer Block mainly consists of two parts: multi-head attention mechanism and feedforward neural network.

[0082] The formula for Multi-Head Attention is as follows:

[0083]

[0084]

[0085] For each Self-Attention instance, there are three independently maintained linear mapping matrices W. i V W i K and W i Q By multiplying the input matrix X with three mapping matrices, we obtain the three inputs of Self-Attetnion: Queries, Keys, and Values.

[0086] Where V, Q, and K represent the Queries, Keys, and Values ​​matrices of the input sentence, respectively. Each row of the matrix represents the Query, Key, and Value vectors corresponding to each word. kThis represents the vector length. Therefore, the Transformer has efficient parallel computing capabilities. Through the output of Self-Attention, the first feature matrix and the second feature matrix are interacted, and this is done through a mapping matrix W. O Multiplying this matrix by the first feature matrix (to compress the output matrix) yields the entire Multi-HeadAttention output, which is the interaction matrix. This interaction matrix is ​​then combined with the first feature matrix to obtain a fusion matrix. The target data category is then determined by referencing multiple pre-defined sample feature information and comprehensively considering the similarities between the target data's feature information and the sample feature information, thereby improving the accuracy of the classification results.

[0087] It should be noted that in the data category determination method provided by this invention, Self-Attention is a vertical interaction method, enabling the first feature matrix corresponding to the feature information of the target data and the second feature matrix corresponding to the feature information of each sample to interact, thus integrating the features of all samples in memory. For example... Figure 6 As shown, distance is illustrated using text data. Existing Self-Attention is a horizontal interaction method, which is the interaction between different words in a text, while this invention is a vertical interaction method, which is the interaction between a text and its other texts.

[0088] Example 1: You are a good student;

[0089] Memory 1: I like programming;

[0090] Memory 2: The weather is nice today;

[0091] Traditional approach: a1, a2, a3 are different words in the same text, representing "you," "is," and "one," respectively. Self-Attention across different texts: In a1, b1, b2, a1 and b1 are words in the same relative position in different texts, representing "you" and "I," or b1 and b2, representing "you" and "today," respectively. a1 and b2 can be words in different positions in different texts, representing "you" and "like," or b1 and b2, representing "you" and "weather," respectively. This allows the feature information of the target data to interact with the feature information of each sample.

[0092] In some optional implementations, determining the category of the target data based on the fusion matrix and the interaction tags includes:

[0093] The fusion matrix is ​​pooled to obtain a pooled fusion matrix;

[0094] The category of the target data is determined based on the pooled fusion matrix.

[0095] In one example, such as Figure 5 As shown, the second feature matrix of the sample feature information is interacted with the first feature matrix of the target data feature information to obtain the interaction matrix. Then, the first feature matrix corresponding to the target data feature information and the interaction matrix determined by comparing the target data feature information with the sample feature information in the memory module are further merged through the feature merging module. A fully connected layer is connected to the output layer, and the model is trained using the following loss function: mean pooling or max pooling can be chosen for the matrix connection. Pooling is then applied to the fused matrix. In this way, when determining the category of the target data, multiple preset sample feature information are referenced, comprehensively considering the similarities and differences between the target data feature information and the sample feature information, thereby improving the accuracy of the classification results. The loss function can be found below:

[0096]

[0097] Where N is the number of samples and K is the number of labels.

[0098] In some optional implementations, after determining the category of the target data based on the feature information and the interaction feature information of the target data, the method further includes:

[0099] Under certain preset conditions, the feature information of the target data is stored as sample feature information, wherein the preset conditions include any one of the following:

[0100] The sample feature information does not include the feature information of the target data;

[0101] Upon re-examination of the target data categories, classification errors were found.

[0102] In one example, the memory module pre-stores the sample feature information of each sample data in multiple sample data, and each sample data has a corresponding sample feature information and sample label stored in the memory module.

[0103] The memory module can also actively acquire feature information from sample data and store it as sample feature information. For example, if a classification error is found during a re-examination of the target data's category, the target data can be used as sample data, and its feature information can be extracted and stored as sample feature information to improve the accuracy of subsequent classifications. Alternatively, when the sample feature information does not include the target data's feature information, the target data's feature information can also be stored as sample feature information. This enriches the types of sample feature information and enhances the model's learning and generalization abilities.

[0104] Please see Figure 7 , Figure 7 This is a structural diagram illustrating the data category determination of a product provided in an embodiment of the present invention. For example... Figure 7 As shown in the figure, an embodiment of the present invention provides a data category determination product for implementing the above-described data category determination method. The product includes:

[0105] The feature extraction module includes a first feature extractor and a first label submodule. The first feature extractor is used to extract features from the target data to obtain a first feature matrix of the target data. The first feature matrix includes a first label vector of the target data. The first label submodule is used to mask the first label vector.

[0106] The memory module stores sample feature information of multiple sample data. The sample feature information includes a second feature matrix and a second label vector. The memory includes a second feature extractor and a second label submodule. The second feature extractor is used to extract features from each sample data to obtain a second feature matrix for each sample data. The second feature matrix includes a corresponding second label vector. The second label submodule is used to obtain the second label vector.

[0107] An interaction module is used to perform feature interaction between the feature information of the target data and the pre-stored feature information of each sample to obtain interactive feature information. The interactive feature information is used to characterize the similarities and differences between the feature information of the target data and the feature information of each sample.

[0108] The output module is used to determine the category of the target data based on the feature information and the interaction feature information of the target data.

[0109] The data category determination product provided in this embodiment of the invention can achieve Figure 1 The various processes implemented in the method embodiments shown are capable of achieving the same beneficial effects, and will not be described again here to avoid repetition.

[0110] Please see Figure 8 , Figure 8 This is a schematic diagram of the data category determination device provided in an embodiment of the present invention. Figure 8 As shown, an embodiment of the present invention also provides a data category determination device 800, comprising:

[0111] The interaction module 801 is used to perform feature interaction between the feature information of the target data and the pre-stored feature information of each sample to obtain interactive feature information. The interactive feature information is used to characterize the similarities and differences between the feature information of the target data and the feature information of each sample.

[0112] The determining module 802 is used to determine the category of the target data based on the feature information of the target data and the interaction feature information.

[0113] Optionally, the feature information of the target data includes a first feature matrix and a first label vector, and the feature information of each sample includes a second feature matrix and a second label vector;

[0114] The interaction module 801 includes:

[0115] The first interaction submodule is used to perform feature interaction between the first feature matrix and the second feature matrix corresponding to each sample feature information to obtain an interaction matrix. The interaction matrix is ​​concatenated with an interaction label vector, which is a label vector obtained by performing label interaction between the first label vector and the second label vector corresponding to each sample feature information.

[0116] The determining module 802 includes:

[0117] The merging submodule is used to merge the interaction matrix with the first feature matrix to obtain a fusion matrix;

[0118] A determination submodule is used to determine the category of the target data based on the fusion matrix.

[0119] Optionally, the determining submodule includes:

[0120] A pooling unit is used to pool the fusion matrix to obtain a pooled fusion matrix;

[0121] The determining unit is used to determine the category of the target data based on the pooled fusion matrix.

[0122] Optionally, the data category determination device 800 further includes:

[0123] The extraction module is used to extract features from multiple sample data to obtain sample feature information corresponding to the multiple sample data.

[0124] The first storage module is used to store the sample feature information.

[0125] Optionally, the data category determination device 800 further includes:

[0126] The second storage module is used to store the feature information of the target data as sample feature information under preset conditions, wherein the preset conditions include any one of the following:

[0127] The sample feature information does not include the feature information of the target data;

[0128] Upon re-examination of the target data categories, classification errors were found.

[0129] The data category determination device provided in this embodiment of the invention can achieve Figure 1 The various processes implemented in the method embodiments shown are capable of achieving the same beneficial effects, and will not be described again here to avoid repetition.

[0130] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of the present invention is not limited to performing functions in the order discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0131] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0132] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A data category determination method characterized by comprising: The method comprises: characteristic information of target data is interacted with each sample characteristic information stored in advance to obtain interaction characteristic information, the interaction characteristic information is used to represent the same points and different points between the characteristic information of the target data and the each sample characteristic information; the category of the target data is determined according to the characteristic information of the target data and the interaction characteristic information; the characteristic information of the target data comprises a first feature matrix and a first label vector, and each sample characteristic information comprises a second feature matrix and a second label vector; the characteristic information of the target data is interacted with each sample characteristic information stored in advance to obtain interaction characteristic information, which comprises: the first feature matrix is interacted with the second feature matrix corresponding to each sample characteristic information to obtain an interaction matrix, the interaction matrix is spliced with an interaction label vector, and the interaction label vector is a label vector obtained by label interaction according to the first label vector and the second label vector corresponding to each sample characteristic information; the category of the target data is determined according to the characteristic information of the target data and the interaction characteristic information, which comprises: the interaction matrix is combined with the first feature matrix to obtain a fusion matrix; the category of the target data is determined according to the fusion matrix.

2. The method of claim 1, wherein, the category of the target data is determined according to the fusion matrix, which comprises: the fusion matrix is pooled to obtain a pooled fusion matrix; the category of the target data is determined according to the pooled fusion matrix.

3. The method of claim 1, wherein, Before the characteristic information of the target data is interacted with each sample characteristic information stored in advance to obtain interaction characteristic information, the method further comprises: sample characteristic information corresponding to a plurality of sample data is obtained by performing feature extraction on the plurality of sample data; the sample characteristic information is stored.

4. The method of claim 1, wherein, After the category of the target data is determined according to the characteristic information of the target data and the interaction characteristic information, the method further comprises: in the case that a preset condition is met, the characteristic information of the target data is stored as sample characteristic information, wherein the preset condition comprises any one of the following: the sample characteristic information does not include the characteristic information of the target data; the category of the target data is rechecked, and there is a classification error.

5. A data class determination product, characterized by The product for implementing the data category determination method in any one of claims 1 to 4 comprises: a feature extraction module comprising a first feature extractor and a first label submodule, the first feature extractor is used for performing feature extraction on the target data to obtain a first feature matrix of the target data, the first feature matrix comprises a first label vector of the target data, and the first label submodule is used for masking the first label vector; A memory module stores sample feature information of a plurality of sample data, the sample feature information comprising a second feature matrix and a second label vector, wherein the memory comprises a second feature extractor and a second label submodule, the second feature extractor is configured to perform feature extraction on each sample data to obtain a second feature matrix of each sample data, the second feature matrix comprises a corresponding second label vector, and the second label submodule is configured to obtain the second label vector; An interaction module is configured to perform feature interaction between feature information of target data and each sample feature information stored in advance to obtain interaction feature information, the interaction feature information being used to represent same points and different points between the feature information of the target data and the each sample feature information; An output module is configured to determine a category of the target data according to the feature information of the target data and the interaction feature information; The feature interaction between the feature information of the target data and each sample feature information stored in advance to obtain the interaction feature information comprises: performing feature interaction between the first feature matrix and a second feature matrix corresponding to each sample feature information to obtain an interaction matrix, the interaction matrix being spliced with an interaction label vector, and the interaction label vector being a label vector obtained by performing label interaction between the first label vector and a second label vector corresponding to each sample feature information; The determination of the category of the target data according to the feature information of the target data and the interaction feature information comprises: performing feature merging between the interaction matrix and the first feature matrix to obtain a fusion matrix; and determining the category of the target data according to the fusion matrix.

6. A data class determination apparatus characterized by comprising: Comprise: An interaction module is configured to perform feature interaction between feature information of target data and each sample feature information stored in advance to obtain interaction feature information, the interaction feature information being used to represent same points and different points between the feature information of the target data and the each sample feature information; An output module is configured to determine a category of the target data according to the feature information of the target data and the interaction feature information; The feature information of the target data comprises a first feature matrix and a first label vector, and each sample feature information comprises a second feature matrix and a second label vector; The interaction module comprises: A first interaction submodule is configured to perform feature interaction between the first feature matrix and a second feature matrix corresponding to each sample feature information to obtain an interaction matrix, the interaction matrix being spliced with an interaction label vector, and the interaction label vector being a label vector obtained by performing label interaction between the first label vector and a second label vector corresponding to each sample feature information; The determination module comprises: A merging submodule is configured to perform feature merging between the interaction matrix and the first feature matrix to obtain a fusion matrix; and A determination submodule is configured to determine the category of the target data according to the fusion matrix.

7. The apparatus of claim 6, wherein, The determination submodule comprises: A pooling unit is configured to perform pooling on the fusion matrix to obtain a pooled fusion matrix. A determination unit is configured to determine a category of the target data according to the pooled fusion matrix.

8. The apparatus of claim 6, wherein, The device further includes: An extraction module is configured to perform feature extraction on a plurality of sample data to obtain sample feature information corresponding to the plurality of sample data. A first storage module is configured to store the sample feature information.

Images