Resource classification method, device, electronic device and readable storage medium

Abstract

The embodiment of the present invention provides a resource classification method, device, electronic device and readable storage medium, which relates to the field of data processing technology, including: obtaining a resource object set composed of resource objects of multiple data types; extracting features of each resource object in the resource object set to obtain target features corresponding to each resource object; mapping the target features to the target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space; inputting the feature vector set into a target classification model to obtain target categories corresponding to each resource object output by the target classification model. Resource objects in a multimodal resource object set can be classified, and in the classification process, features of resource objects of different data types can be mapped to the same feature space, and batch classification operations can be completed for resource objects in the same feature space, thereby improving classification accuracy while ensuring the efficiency of multimodal resource classification.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and in particular relates to a resource classification method, device, electronic equipment and readable storage medium. Background Art

[0002] All industries are accelerating their digital transformation. In this process, resource objects need to be classified. Resource object classification usually divides data into different categories or groups based on its specific attributes, characteristics or uses, so as to facilitate subsequent management, statistics, research and application.

[0003] In related technologies, resource objects are usually classified manually, which places extremely high demands on the professional level, data literacy, and classification efficiency of the staff. In order to improve the classification efficiency, deep learning technology can also be used to classify specific types of resource objects.

[0004] However, resource objects often include massive, multimodal data, which makes it difficult for related technologies to efficiently and accurately classify multimodal resource objects. Summary of the invention

[0005] The present invention provides a resource classification method, device, electronic device and readable storage medium to solve the technical problem of poor accuracy and efficiency in classifying multi-modal resource objects in related technologies.

[0006] In a first aspect, the present invention provides a resource classification method, the method comprising:

[0007] Get a resource object set consisting of resource objects of multiple data types;

[0008] Extracting features from each resource object in the resource object set to obtain target features corresponding to each resource object;

[0009] Mapping the target feature to a target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space;

[0010] The feature vector set is input into a target classification model to obtain a target category corresponding to each resource object output by the target classification model.

[0011] Optionally, extracting features from each resource object in the resource object set to obtain target features corresponding to each resource object includes:

[0012] Determine the target data type corresponding to the resource object, and obtain the target feature extraction strategy corresponding to the target data type;

[0013] Based on the target feature extraction strategy, feature extraction is performed on the resource object to obtain target features corresponding to the resource object.

[0014] Optionally, the target feature extraction strategy includes a target structuring strategy and a target feature generation strategy, and the step of extracting features from the resource object based on the target feature extraction strategy to obtain the target features corresponding to the resource object includes:

[0015] Performing structured processing on the resource object based on the target structured strategy to obtain target structured data corresponding to the resource object;

[0016] Based on the target feature generation strategy, feature extraction is performed on the target structured data to obtain target features corresponding to the resource object.

[0017] Optionally, the target structuring strategy includes a target segmentation strategy and a target template, and the structuring process of the resource object based on the target structuring strategy to obtain target structured data corresponding to the resource object includes:

[0018] Segmenting the resource object based on the target segmentation strategy to obtain multiple resource fragments;

[0019] Generate fragment description information corresponding to each of the resource fragments;

[0020] generating object description information of the resource object based on the fragment description information;

[0021] The segment description information and the object description information are structured according to the target template to obtain the target structured data.

[0022] Optionally, the target feature generation strategy includes a first feature model corresponding to the data type of the resource object, and a second feature model corresponding to all data types, and the feature extraction of the target structured data based on the target feature generation strategy to obtain the target feature corresponding to the resource object includes:

[0023] Extracting features from the target structured data based on the first feature model to obtain a first feature corresponding to the resource object;

[0024] The first feature is input into a second feature model to obtain a target feature corresponding to the resource object output by the second feature model.

[0025] Optionally, the target classification model includes a first classification sub-model and a second classification sub-model, and the step of inputting the feature vector set into the target classification model to obtain the target category corresponding to each resource object output by the target classification model includes:

[0026] Inputting the feature vector set into the first classification sub-model to obtain a first classification result output by the first classification sub-model;

[0027] Based on the feature vector set and the second classification sub-model, obtaining a second classification result output by the second classification sub-model;

[0028] The target category corresponding to each of the resource objects is determined based on the first classification result and the second classification result.

[0029] Optionally, obtaining a second classification result output by the second classification sub-model based on the feature vector set and the second classification sub-model includes:

[0030] Fusion of the feature vector in the feature vector set with the first classification result corresponding to the feature vector to obtain a fused feature vector set;

[0031] The fused feature vector set is input into the second classification sub-model to obtain a second classification result output by the second classification sub-model.

[0032] In a second aspect, the present invention provides a resource classification device, the device comprising:

[0033] An acquisition module, used to acquire a resource object set consisting of resource objects of multiple data types;

[0034] A feature module, used to extract features from each resource object in the resource object set to obtain target features corresponding to each resource object;

[0035] A mapping module, used to map the target feature to a target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space;

[0036] The classification module is used to input the feature vector set into the target classification model to obtain the target category corresponding to each resource object output by the target classification model.

[0037] Optionally, the feature module includes:

[0038] An extraction strategy submodule, used to determine the target data type corresponding to the resource object, and obtain a target feature extraction strategy corresponding to the target data type;

[0039] The feature extraction submodule is used to extract features from the resource object based on the target feature extraction strategy to obtain target features corresponding to the resource object.

[0040] Optionally, the target feature extraction strategy includes a target structuring strategy and a target feature generation strategy, and the feature extraction submodule includes:

[0041] A structured data unit, used to perform structured processing on the resource object based on the target structuring strategy to obtain target structured data corresponding to the resource object;

[0042] The structured data extraction unit is used to extract features from the target structured data based on the target feature generation strategy to obtain target features corresponding to the resource object.

[0043] Optionally, the target structured strategy includes a target segmentation strategy and a target template, and the structured data unit includes:

[0044] A resource fragment subunit, configured to perform segmentation processing on the resource object based on the target segmentation strategy to obtain a plurality of resource fragments;

[0045] A fragment description information subunit, used to generate fragment description information corresponding to each of the resource fragments;

[0046] An object description information subunit, configured to generate object description information of the resource object based on the fragment description information;

[0047] The structured data generating subunit is used to perform structured processing on the fragment description information and the object description information according to the target template to obtain the target structured data.

[0048] Optionally, the target feature generation strategy includes a first feature model corresponding to a data type of the resource object and a second feature model corresponding to all data types, and the structured data extraction unit includes:

[0049] A first feature subunit, configured to extract features from the target structured data based on the first feature model to obtain a first feature corresponding to the resource object;

[0050] The target feature generation subunit is used to input the first feature into the second feature model to obtain the target feature corresponding to the resource object output by the second feature model.

[0051] Optionally, the target classification model includes a first classification sub-model and a second classification sub-model, and the classification module includes:

[0052] A first classification result submodule, used for inputting the feature vector set into the first classification submodel to obtain a first classification result output by the first classification submodel;

[0053] A second classification result submodule, used for obtaining a second classification result output by the second classification submodel based on the feature vector set and the second classification submodel;

[0054] A target category submodule is used to determine the target category corresponding to each of the resource objects based on the first classification result and the second classification result.

[0055] Optionally, the second classification result submodule comprises:

[0056] A fusion feature vector set unit, used to fuse the feature vector in the feature vector set with the first classification result corresponding to the feature vector to obtain a fused feature vector set;

[0057] A fusion unit is used to input the fused feature vector set into the second classification sub-model to obtain a second classification result output by the second classification sub-model.

[0058] In a third aspect, the present invention provides an electronic device comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned resource classification method when executing the program.

[0059] In a fourth aspect, the present invention provides a readable storage medium, which, when instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to execute the above-mentioned resource classification method.

[0060] In an embodiment of the present invention, it includes: obtaining a resource object set consisting of resource objects of multiple data types; extracting features from each resource object in the resource object set to obtain target features corresponding to each resource object; mapping the target features to the target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space; inputting the feature vector set into a target classification model to obtain target categories corresponding to each resource object output by the target classification model. Resource objects in a multimodal resource object set can be classified, and in the classification process, features of resource objects of different data types can be mapped to the same feature space, so that the features of resource objects have high comparability, and batch classification operations are completed on resource objects in the same feature space, thereby improving classification accuracy while ensuring the efficiency of multimodal resource classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0062] Figure 1 is a flow chart of the steps of a resource classification method provided by an embodiment of the present invention;

[0063] Figure 2 is a flowchart of another resource classification method provided by an embodiment of the present invention;

[0064] Figure 3 is a schematic diagram of structured data processing provided by an embodiment of the present invention;

[0065] Figure 4 is a schematic diagram of unstructured data processing provided by an embodiment of the present invention;

[0066] Figure 5 is a schematic diagram of text object processing provided by an embodiment of the present invention;

[0067] Figure 6 is a schematic diagram of image object processing provided by an embodiment of the present invention;

[0068] Figure 7 is a schematic diagram of audio object processing provided by an embodiment of the present invention;

[0069] Figure 8 is a schematic diagram of video object processing provided by an embodiment of the present invention;

[0070] Fig. 9 is an architecture diagram of a resource classification system provided by an embodiment of the present invention;

[0071] Fig.10 is a structural diagram of a resource classification device provided by an embodiment of the present invention;

[0072] Fig.11 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0073] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0074] Reference Figure 1 , Figure 1 1 is a flowchart of a resource classification method provided by an embodiment of the present invention. Figure 1 As shown, the method may include:

[0075] Step 101: Acquire a resource object set consisting of resource objects of multiple data types.

[0076] In the embodiment of the present invention, the resource objects to be classified can be collected, collated, preprocessed, etc. to form a multimodal resource object set. The data types of these resource objects may include but are not limited to structured types (such as MySQL data), unstructured types (such as NoSQL data), image types, text types, audio types, video types, etc., which are not specifically limited in the embodiment of the present invention.

[0077] For example, Company A needs to archive and classify all resources for digital transformation. It can first collect all resource objects. These resource objects of different data types can form a resource object set.

[0078] It should be noted that the resource object set may represent all resource objects of a storage space, all resource objects of a subject, all resource objects of a folder, a set of all resource objects with associated relationships, etc. The resource object set may be specific, such as all resource objects in the same folder, or virtual, such as all resource objects generated by a company this month, which is not specifically limited in the embodiments of the present invention.

[0079] Step 102: extracting features from each resource object in the resource object set to obtain target features corresponding to each resource object.

[0080] Extracting features from resource objects means extracting key information from the original data of resource objects and capturing useful characteristics or patterns in the data for subsequent tasks. The form of target features may include but is not limited to various types of feature data such as feature text, feature vector, embedding representation, feature matrix, etc., which are not specifically limited in the embodiments of the present invention.

[0081] In an embodiment of the present invention, feature extraction can be performed on each resource object in the resource object set, so as to obtain target features corresponding to each resource object in the resource object set. Specifically, feature extraction can be performed on each resource object through a feature extraction model to obtain corresponding target features. Among them, the feature extraction model can be a neural network model, such as a convolutional neural network (CNN), a recurrent neural network (RNN), an autoencoder, etc., and the feature extraction model can also be a mathematical model, such as a frequency domain analysis model, a feature statistical model, etc., which is not specifically limited in the embodiment of the present invention.

[0082] It should be noted that since the resource objects can be of different data types, a multimodal feature extraction model can be used, such as CLIP (Contrastive Language–Image Pretraining), ViLT (Vision-and-Language Transformer), a multimodal Transformer model, etc., which is not specifically limited in the embodiments of the present invention. A multimodal feature extraction model is a deep learning model that can extract features from different types of data modalities (such as images, text, audio, video, etc.).

[0083] Step 103: Map the target features to a target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space.

[0084] Feature space is an abstract vector space. Mapping the target features to the feature space enables them to be operated and compared in the space in the form of vectors. In an embodiment of the present invention, the extracted feature vectors of resource objects of different modalities can be transformed and mapped into a common target feature space through linear and nonlinear methods such as principal component analysis (PCA), linear discriminant analysis (LDA), kernel principal component analysis (KPCA), and autoencoder, so that the features of resource objects of different modalities have high correlation and comparability in this common feature space. Among them, the target feature space after completing the above-mentioned feature mapping includes the feature vectors of each resource object in the resource object set, which together form a high-dimensional feature vector space.

[0085] In addition, the target features corresponding to each resource object can be mapped to the target feature space through the feature mapping model to obtain the feature vectors corresponding to each resource object in the target feature space, and these feature vectors can constitute the feature vector set of the resource object set in the target feature space. Different feature mapping models can map the target features to different feature spaces.

[0086] Specifically, a feature mapping model capable of mapping features of objects of different data types to the same feature space can be trained in advance through sample features of sample objects of multiple data types. First, an initial model is set, and then the sample features of the sample objects are input into the initial model to obtain the mapped feature vectors of the sample objects output by the initial model. The predicted distances between the sample objects are calculated based on the feature vectors of the sample objects, and the model loss is calculated based on the difference between the predicted distances and the label distances between the sample objects. The parameters of the initial model are adjusted based on the model loss until the training end condition of the initial model is reached to obtain the above-mentioned feature mapping model.

[0087] By mapping the target features to the target feature space and obtaining the feature vector set corresponding to the resource object set in the target feature space, resource objects of different data types can be made more comparable in the target feature space, thereby improving the accuracy of subsequent resource classification.

[0088] Step 104: input the feature vector set into a target classification model to obtain a target category corresponding to each resource object output by the target classification model.

[0089] In the embodiment of the present invention, the feature vector set may be input into the target classification model to perform batch classification operations on resource objects to obtain target categories corresponding to various resource objects output by the target classification model.

[0090] In one embodiment, a target classification model can be obtained in advance by training sample objects with category labels, so that the target classification model can continuously optimize the cluster centers corresponding to each preset category. The target classification model can determine the target category of the resource object corresponding to each feature vector based on the above cluster center and the feature distance between each feature vector in the feature vector set in the target feature space. The target classification model can also use the above cluster center to perform clustering operations on each feature vector in the feature vector set in the target feature space, that is, the target category of the resource object corresponding to all feature vectors in the clustering result is determined as the preset category of the cluster center corresponding to the clustering result. Among them, the above preset category can be a multi-level category. Exemplarily, the preset category can include a first-level organizational category (such as a subsidiary, department, project team, etc.) and a second-level content category (such as news, financial reports, conference materials, etc.).

[0091] In another embodiment, the feature vector set can be input into a clustering model, and the feature vectors in the feature vector set can be clustered by the clustering model to obtain the clustering results of the feature vectors in the target feature space, each clustering result (class) can include multiple feature vectors, and then the target category corresponding to the class is determined based on the resource object corresponding to at least one feature vector contained in each class, and the resource objects corresponding to all feature vectors in the class can be determined as the target category corresponding to the class. Exemplarily, the above resource objects and preset categories can be input into a large language model, and the target category corresponding to the resource object can be determined from the preset categories by the large language model.

[0092] In summary, a resource classification method provided by an embodiment of the present invention includes: obtaining a resource object set consisting of resource objects of multiple data types; extracting features from each resource object in the resource object set to obtain target features corresponding to each resource object; mapping the target features to the target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space; inputting the feature vector set into a target classification model to obtain target categories corresponding to each resource object output by the target classification model. The resource objects in a multimodal resource object set can be classified, and in the classification process, the features of resource objects of different data types can be mapped to the same feature space, so that the features of the resource objects have high comparability, and batch classification operations are completed on the resource objects in the same feature space, thereby improving the classification accuracy while ensuring the efficiency of multimodal resource classification.

[0093] Reference Figure 2 , Figure 2 FIG. 1 is a flowchart of another resource classification method provided by an embodiment of the present invention. Figure 2 As shown, the method may include:

[0094] Step 201: Acquire a resource object set consisting of resource objects of multiple data types.

[0095] This step can refer to the above-mentioned step 101, and will not be described in detail in this embodiment of the present invention.

[0096] Step 202: determine the target data type corresponding to the resource object, and obtain the target feature extraction strategy corresponding to the target data type.

[0097] In an embodiment of the present invention, the resource object may be different types of data. In order to improve the accuracy of subsequent feature comparison and feature clustering, different methods may be used to extract features for different resource objects. The target data type corresponding to the resource object may be determined first. The data type may include, but is not limited to, structured types (e.g., MySQL data), unstructured types (e.g., NoSQL data), image types, text types, audio types, video types, etc., which are not specifically limited in the embodiment of the present invention.

[0098] Specifically, the target data type of the resource object can be determined according to the file suffix of the resource object, or the file header of the resource object can be detected; the corresponding target data type can be matched according to the file header; the resource object can also be analyzed for content to determine the corresponding target data type, for example, the content of a JSON file usually starts with { or [, and the content of an XML file usually starts with <? xml; the target data type of the resource object can also be determined by other methods, which are not specifically limited in the embodiments of the present invention.

[0099] In the embodiment of the present invention, different feature extraction strategies can be set for different data types, so that after obtaining the target data type of the resource object, the target feature extraction strategy corresponding to the target data type can be matched, and the target feature of the resource object can be extracted through the target feature extraction strategy.

[0100] Exemplarily, for structured data types, the corresponding feature extraction strategy can analyze the file content, extract the metadata and data ontology of the data object, and use the metadata and data ontology as its target features, or generate target features based on the metadata and data ontology. Figure 3 , Figure 3 A schematic diagram of structured data processing provided by an embodiment of the present invention is shown, such as Figure 3 As shown, metadata can be extracted from the structured data object first, and then the corresponding data ontology can be extracted based on the metadata.

[0101] For example, for unstructured data objects (such as NoSQL data), the corresponding feature extraction strategy can analyze the file content, convert it into structured data first, then extract the metadata and data ontology of the unstructured data object, and use the metadata and data ontology as its target features, or generate target features based on the metadata and data ontology. Figure 4 , Figure 4 A schematic diagram of unstructured data processing provided by an embodiment of the present invention is shown, Figure 4 As shown, the unstructured data object can be structured first to obtain the corresponding structured data, then the metadata of the structured data can be extracted, and then the corresponding data ontology can be extracted based on the metadata.

[0102] For example, refer to Figure 5 , Figure 5 A schematic diagram of text object processing provided by an embodiment of the present invention is shown. Figure 5 As shown, for resource objects of text data types, the corresponding feature extraction strategy can generate feature information based on at least one result of document analysis, document sentence segmentation, document word segmentation, part-of-speech tagging, stop word processing, high-frequency / low-frequency word processing, entity extraction, etc., to extract target features of the document data object. Among them, document analysis can identify the document structure and extract structural elements and their relationships through structural analysis algorithms, document classification models, document analysis models, etc.; document sentence segmentation can identify the sentence boundaries of the document content through the sentence segmentation model, thereby separating the document into sentences; document word segmentation can analyze documents or sentences through word segmentation libraries, word segmenters, and word segmentation models to obtain word segmentation results such as words, phrases, and phrases; stop word processing can remove stop words from the word segmentation results through a predefined stop word list; high-frequency / low-frequency word processing can statistically analyze the frequency of occurrence of each word in the text, and filter the high-frequency / low-frequency words in the word segmentation results according to the preset high-frequency and low-frequency word thresholds; part-of-speech tagging can use the part-of-speech tagging model to tag the word segmentation results with corresponding parts of speech or word categories.

[0103] For example, refer to Figure 6 , Figure 6 A schematic diagram of image object processing provided by an embodiment of the present invention is shown. Figure 6 As shown, for resource objects of image data types, the corresponding feature extraction strategy can cut the target area in the image through an optical character recognition (OCR) algorithm based on features, templates, and deep learning, and locate and detect the target content from the target area, perform structure extraction based on the target area to obtain image structure information, perform content extraction based on the target content to obtain image content information, generate feature information based on the image structure information and image content information, and obtain the corresponding target features. In addition, if the image contains a document structure (such as a scanned copy, PDF document, etc.), the document content in the image can also be processed through algorithms such as entity objects, headers and footers, table extraction, text block recognition, and key field extraction to extract the structured information and content therein for generating target features.

[0104] For example, refer to Figure 7 , Figure 7 A schematic diagram of audio object processing provided by an embodiment of the present invention is shown. Figure 7As shown, for resource objects of audio data types, the corresponding feature extraction strategy can first process the audio through digital signal processing (DSP) methods such as filters, spectral subtraction, and time domain smoothing to enhance, reduce noise, remove, truncate, and normalize the audio to obtain processed audio, and then use short-time Fourier transform and audio feature algorithms to extract the Mel Frequency Cepstral Coefficients (MFCC), Linear Predictive Coding (LPC), Zero Crossing Rate (ZCR), Energy (Energy) and other audio features of the processed audio, and map the audio features to at least one model such as Hidden Markov Model (HMM), Convolutional Neural Network (CNN), n-gram model, etc. to obtain the audio features of the audio resource object output by the model.

[0105] Exemplarily, for resource objects of video data type, the corresponding feature extraction strategy can be based on theoretical methods of motion trajectory (such as optical flow method, motion trajectory clustering, etc.), suppress irrelevant information such as static background in the video, and identify and track moving targets in the video, extract key information of the moving targets and perform classification judgment, so as to analyze, identify and understand the targets and dynamic information in the video content, and effectively judge the category and key information of the targets of interest in the video resource objects.

[0106] Reference Figure 8 , Figure 8 A schematic diagram of video object processing provided by an embodiment of the present invention is shown. Figure 8As shown in the figure, the core operations for feature extraction of video resource objects may include target detection, target tracking, action recognition, event detection, and key frame extraction; among them, target detection may use algorithms such as YOLO (You Only Look Once), Faster R-CNN (Faster Region-based Convolutional Neural Network), and SSD (Single Shot MultiBox Detector) to detect and locate targets in each frame of video sequence or key frame; target tracking may use Simple Online and Realtime Tracking (SORT), Deep Simple Online and Realtime Tracking (DESRT) to detect and locate targets in each frame of video sequence or key frame. Tracking, DeepSORT) and other algorithms are used to initialize the target tracker, and update and predict the target position in subsequent frames, so as to track the motion trajectory of the target in the video frame in the continuous video frames; action recognition and event detection can use models such as convolutional neural network (CNN) and recurrent neural network (RNN) to extract video frame features or inter-frame features, and input these frame features and inter-frame features into pre-trained classifiers or recurrent neural networks to classify and recognize actions and events in video frames; key frame extraction can select the key frame that best represents the entire video content from the video sequence through algorithms based on image quality assessment, inter-frame difference and visual feature extraction, and then generate feature information based on the results of the above target detection, target tracking, action recognition and event detection, and key frame extraction to obtain the corresponding target features.

[0107] Step 203: extract features from the resource object based on the target feature extraction strategy to obtain target features corresponding to the resource object.

[0108] In the embodiment of the present invention, a target feature extraction strategy corresponding to a resource object may be used to extract features from the resource object to obtain target features corresponding to the resource object.

[0109] By determining the target data type corresponding to the resource object; extracting the target feature strategy based on the target data type; and extracting the features of the resource object based on the target feature extraction strategy, the target features corresponding to the resource object are obtained, and the features of different types of resource objects can be accurately extracted, which helps to improve the accuracy of subsequent classification of resource objects.

[0110] Optionally, step 203 may include:

[0111] Sub-step 2031: structuring the resource object based on the target structuring strategy to obtain target structured data corresponding to the resource object.

[0112] In an embodiment of the present invention, the target feature extraction strategy may include a target structuring strategy and a target feature generation strategy, wherein the target structuring strategy is used to perform structured processing on the resource object to obtain the structured data of the resource object; the target feature generation strategy is used to generate its target features based on the structured data of the resource object.

[0113] Different structuring strategies can be set for different data types according to their data characteristics.

[0114] In one embodiment, the target structured data obtained by resource objects of different data types may have the same data structure. Exemplarily, the target structured data may be obtained by concatenating the file name, data type characteristics, and content characteristics of the resource object, or by concatenating the information and characteristics contained in resource objects of other data types. This embodiment of the present invention does not make any specific limitation.

[0115] In another embodiment, the target structured data of resource objects of different data types may have different data structures, so that the target structured data can more specifically reflect the actual content of the resource object and improve the accuracy of obtaining target features based on structured data. For example, the target structured data of image-type resource objects can be obtained by splicing the file name, image format information, target features contained in the image, and image size information; the target structured data of audio-type resource objects can be obtained by splicing the file name, audio format information, voice recognition content, and sound features.

[0116] It should be noted that even when the target structured data corresponding to each data type has the same data structure, the structuring strategies corresponding to each data type may be different due to the different data processing methods required for different data types.

[0117] Optionally, sub-step 2032 may include:

[0118] Sub-step 20321, segmenting the resource object based on the target segmentation strategy to obtain multiple resource fragments.

[0119] In the embodiment of the present invention, the target structuring strategy may include a target segmentation strategy and a target template, wherein the target segmentation strategy is used to segment resource objects to obtain target segments for feature extraction, and the target template is used to fix the format of structured data.

[0120] Corresponding segmentation strategies can be set for different data types. Specifically, for NoSQL data, the metadata corresponding to each data entry and the data ontology corresponding to the metadata can be extracted as a segment. For example, if each data entry of a NoSQL resource object includes three metadata: project name, project members, and project description, a resource segment can be obtained based on the above metadata and the data ontology corresponding to the metadata, that is, the specific data content combination.

[0121] For resource objects of the document type, the corresponding target segmentation strategy includes, but is not limited to, segmentation by paragraph, segmentation by sentence, etc., which is not specifically limited in the embodiment of the present invention.

[0122] For audio-type resource objects, the corresponding target segmentation strategies include but are not limited to segmentation by time, segmentation by voice objects, or segmentation by content relevance, etc. Among them, segmentation by voice objects can determine the audio content of the same narrator as a resource segment, and segmentation by content relevance can determine the audio describing the same event or theme as a resource segment.

[0123] For image-type resource objects, the corresponding target segmentation strategies include but are not limited to segmentation according to preset areas, segmentation according to object areas, etc., wherein segmentation according to preset areas can divide the image into a preset number of sub-images, and segmentation according to object areas can treat the image area containing a target object (such as a person, vehicle, building, etc.) as a segment.

[0124] For resource objects of the video type, the corresponding target segmentation strategies may include but are not limited to segmentation by time, segmentation by target object, or segmentation by content relevance, etc. Among them, according to target object segmentation, the video content of the same target can be determined as a resource segment, and according to content relevance segmentation, the video screen showing the same event or theme can be determined as a resource segment.

[0125] It should be noted that the design principle of the segmentation strategy can be based on the differences between the contents, so as to divide the resource objects of the corresponding data types into segments with different contents. The segmentation strategy can also be set in a similar manner for other resource types, and the embodiments of the present invention will not be repeated.

[0126] Sub-step 20322: Generate fragment description information corresponding to each of the resource fragments.

[0127] In the embodiment of the present invention, fragment description information corresponding to each resource fragment can be generated. Resource fragments also have corresponding data types, and the data types of resource fragments can generally include text fragments, audio fragments, image fragments, and video fragments. Different semantic extraction models can be used to generate fragment description information for resource fragments of different data types. Among them, the fragment description information may include but is not limited to fragment features, fragment description text, etc., which are not specifically limited in the embodiment of the present invention.

[0128] Exemplarily, a semantic extraction model for images may be used to extract segment description information corresponding to resource segments of image data types. A semantic extraction model for videos may be used to extract segment description information corresponding to resource segments of video data types.

[0129] Sub-step 20323, generating object description information of the resource object based on the fragment description information.

[0130] In an embodiment of the present invention, the fragment description information corresponding to each resource fragment of the resource object can be spliced ​​to obtain the object description information of the resource object. It is also possible to perform overall feature extraction on the fragment description information corresponding to each resource fragment of the resource object to obtain the object description information of the resource object, which is not specifically limited in the embodiment of the present invention. In the case where the fragment description information is text, the generated object description information can also be text information.

[0131] In addition, a fragment description information sequence can be generated according to the temporal or spatial adjacent relationship of the resource fragments corresponding to the fragment description information, and the fragment description information sequence can be input into the temporal feature extraction model to obtain the object description information output by the temporal feature extraction model. Among them, temporally adjacent resource fragments represent temporal adjacency distributed in the corresponding resource object, such as two adjacent audio fragments in an audio object; spatially adjacent resource fragments represent spatial adjacency distributed in the corresponding resource object, such as two image area fragments in an image object. The above-mentioned temporal feature extraction model may include but is not limited to a recurrent neural network (RNN), a long short-term memory network (LSTM), etc., and the embodiment of the present invention does not make specific limitations.

[0132] Sub-step 20324: Structuring the fragment description information and the object description information according to the target template to obtain the target structured data.

[0133] In an embodiment of the present invention, a target template applicable to resource objects of all data types can be used to perform structured processing on the fragment description information and the object description information, thereby obtaining target structured data with the same structure. This enhances the comparability between target structured data corresponding to resource objects of different data types. In the case where the fragment description information is text, the generated target structured data can also be structured text data.

[0134] By segmenting resource objects based on the target segmentation strategy, multiple resource fragments are obtained; the fragment description information corresponding to each resource fragment is generated; the object description information of the resource object is generated based on the fragment description information; the fragment description information and object description information are structured according to the target template to obtain target structured data. It is possible to generate target structured data with high content consistency and high structural consistency for resource objects of different data types, which helps to improve the accuracy of subsequent classification of multimodal resource objects based on target structured data.

[0135] Sub-step 2032: extracting features from the target structured data based on the target feature generation strategy to obtain target features corresponding to the resource object.

[0136] In an embodiment of the present invention, corresponding feature generation strategies may be set for different data types, so that target features obtained by resource objects of different data types have high comparability. In the case where different data types correspond to structured data with the same data structure, the feature generation strategies corresponding to each data type may be the same, and in the case where different data types correspond to structured data with different data structures, the feature generation strategies corresponding to each data type may be different.

[0137] Specifically, the target feature generation strategy can be a rule model, through which the data items in the target structured data can be converted, spliced, and other operations can be performed to obtain the corresponding target features. Exemplarily, the target feature generation strategy can extract the various data items in the target structured data and sequentially splice them to obtain the target features. The target feature generation strategy can also be a neural network model, through which the target structured data is subjected to feature extraction to obtain the corresponding target features. The embodiments of the present invention are not specifically limited.

[0138] By structuring resource objects based on the target structuring strategy, the target structured data corresponding to the resource objects are obtained; by extracting features from the target structured data based on the target feature generation strategy, the target features corresponding to the resource objects are obtained. This can improve the feature consistency and comparability of the target features corresponding to resource objects of different data types, and help improve the accuracy of subsequent resource object classification based on target features.

[0139] Optionally, step 2032 may include:

[0140] Sub-step 20321: extracting features from the target structured data based on the first feature model to obtain a first feature corresponding to the resource object.

[0141] In one embodiment, since the target structured data generated by resource objects of different data types have differences in content features, in order to improve the accuracy of generating target features, the corresponding first feature model can be trained for different data types, and the target structured data is feature extracted by the first feature model. The network structure of the first feature model can be set according to the data content of the corresponding target structured data. For example, the first feature model for processing structured data of text content can include but is not limited to a bag of words model (Bag of Words, BoW), a word embedding model (WordEmbeddings), a document embedding model (Document Embeddings), etc. The first feature model for processing structured data of image content can include but is not limited to SIFT (Scale-Invariant Feature Transform), ResNet (Residual Networks), DenseNet, etc. The first feature model for processing structured data of audio content can include but is not limited to an MFCC model (Mel-Frequency Cepstral Coefficients), Audio-CNN, LSTM model, etc. Other types of structured data can use an adapted model as their first feature model, which is not specifically limited in the embodiment of the present invention.

[0142] In another embodiment, since the target structured data of resource objects of different data types can all be structured text data, the first feature models corresponding to different data types can also be set to perform text feature extraction. Although the structured text data generated by resource objects of different data types have the same data structure, there are still differences in text volume, text content, etc. In order to accurately extract features from the structured text data of resource objects of different data types, the structured text data of sample objects of a specific data type can be used to train the first feature model corresponding to the data type, that is, resource objects of different data types can use different first feature models to extract features from structured text data.

[0143] Sub-step 20322: input the first feature into a second feature model to obtain a target feature corresponding to the resource object output by the second feature model.

[0144] In an embodiment of the present invention, a second feature model can also be set. The second feature model can be applied to target structured data corresponding to resource objects of all data types. By uniformly setting the second feature model, the comparability of target features output by the second feature model can be further improved, which helps to improve the accuracy of subsequent multimodal data classification based on target features.

[0145] Among them, the second feature model can be obtained by training the first features of samples of sample objects of various data types. During the training process, unlabeled training data can be used to complete the training in a self-supervised and unsupervised manner. The second feature model can be flexibly selected according to actual business needs, such as Transformer-based models, RNN and its variants (LSTM / GRU), graph neural networks (GNN), convolutional neural networks, etc., which are not specifically limited in the embodiments of the present invention.

[0146] By extracting features from the target structured data based on the first feature model, the first feature corresponding to the resource object is obtained; the first feature is input into the second feature model to obtain the target feature corresponding to the resource object output by the second feature model, which can further improve the comparability of the target feature and improve the accuracy of subsequent batch classification of resource objects of different data types based on the target feature.

[0147] Step 204: Map the target features to a target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space.

[0148] This step can refer to step 103, and will not be described in detail in this embodiment of the present invention.

[0149] Step 205: input the feature vector set into the first classification sub-model to obtain a first classification result output by the first classification sub-model.

[0150] The target classification model can be a fusion classification model based on the first classification sub-model and the second classification sub-model. The first classification sub-model and the second classification sub-model can include one of a random forest sub-model, a K-nearest neighbor algorithm sub-model, a multi-layer perceptron sub-model (MLP), etc. The first classification sub-model and the second classification sub-model are different. For example, the first classification sub-model can be a random forest sub-model, and the second classification sub-model can be a K-nearest neighbor sub-model, which is not specifically limited in the embodiment of the present invention. Among them, Random Forest is an ensemble learning method based on decision trees, which completes the classification task by constructing multiple decision trees and voting or averaging their output results. K-Nearest Neighbors (KNN) is a classification algorithm that determines the classification of a data point by examining the "neighbor" category closest to the data point.

[0151] In an embodiment of the present invention, the feature vector set may be input into the first classification sub-model to obtain the first classification result corresponding to each resource object output by the first classification sub-model, wherein the first classification result may include the first probability distribution of the resource object for each preset category.

[0152] Step 206: Obtain a second classification result output by the second classification sub-model based on the feature vector set and the second classification sub-model.

[0153] The feature vector set may be input into the second classification sub-model to obtain a second classification result output by the second classification sub-model, wherein the second classification result may include a second probability distribution of the resource object for each preset category.

[0154] Optionally, step 206 may include:

[0155] Sub-step 2061, fusing the feature vector in the feature vector set with the first classification result corresponding to the feature vector to obtain a fused feature vector set.

[0156] In an embodiment of the present invention, the feature vector corresponding to the resource object in the feature vector set can be fused with the first classification result corresponding to the feature vector to obtain a fused feature vector corresponding to the resource object, so that the fused feature vectors of all resource objects can constitute a fused feature vector set. Specifically, the above fusion method can include weighted fusion, splicing fusion, etc., which is not specifically limited in the embodiment of the present invention.

[0157] Sub-step 2062, inputting the fused feature vector set into the second classification sub-model to obtain a second classification result output by the second classification sub-model.

[0158] The feature vector set may be input into the second classification sub-model to obtain a second classification result output by the second classification sub-model, wherein the second classification result may include a second probability distribution of the resource object for each preset category.

[0159] A fused feature vector set is obtained by fusing the feature vectors in the feature vector set with the first classification results corresponding to the feature vectors; the fused feature vector set is input into the second classification sub-model to obtain the second classification results output by the second classification sub-model. Since the fused feature vector set contains information about the original features and the first classification results, it increases the prior knowledge of the category, which can help the second classification model learn the feature distribution of the category more accurately, and help improve the accuracy of the second classification results.

[0160] Step 207: Determine the target category corresponding to each of the resource objects based on the first classification result and the second classification result.

[0161] In an embodiment of the present invention, the first classification result and the second classification result of the resource object can be fused to obtain the target category of the resource object. The fusion method can include an average method, that is, calculating the average probability of the resource object for each preset category, and then selecting the preset category with the highest average probability as the target category of the resource object. The fusion method can also include a weighted average method, which is not specifically limited in the embodiment of the present invention.

[0162] It should be noted that if the second classification result is obtained based on the fused feature vector set, a higher weight can be set for the second classification result, and a weighted average can be performed with the first classification result to obtain the target category.

[0163] By inputting the feature vector set into the first classification sub-model, the first classification result output by the first classification sub-model is obtained; based on the feature vector set and the second classification sub-model, the second classification result output by the second classification sub-model is obtained; based on the first classification result and the second classification result, the target category corresponding to each resource object is determined, and the target category of the resource object can be determined by fusing multiple classification models, which helps to improve the accuracy of classifying resource objects.

[0164] In summary, another resource classification method provided by an embodiment of the present invention includes: obtaining a resource object set consisting of resource objects of multiple data types; performing feature extraction on each resource object in the resource object set to obtain target features corresponding to each resource object; mapping the target features to the target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space; inputting the feature vector set into a target classification model to obtain target categories corresponding to each resource object output by the target classification model. The resource objects in a multimodal resource object set can be classified, and in the classification process, the features of resource objects of different data types can be mapped to the same feature space, so that the features of the resource objects have high comparability, and batch classification operations are completed on the resource objects in the same feature space, thereby improving the classification accuracy while ensuring the efficiency of multimodal resource classification.

[0165] Reference Fig. 9 , Fig. 9 is an architecture diagram of a resource classification system provided by an embodiment of the present invention, such as Fig. 9As shown, the resource classification system may include a multimodal data resource layer, a preprocessing layer, a feature fusion layer, and a resource classification layer. The multimodal data resource layer is used to carry resource objects, which may include but are not limited to structured data, NoSQL data, text, images, audio, video, etc. The preprocessing layer is used to carry various data processing algorithms, models, etc. These algorithms and models can be used to process and detect resource objects of different data types to obtain data processing results of resource objects. The above algorithms and models may include but are not limited to convolutional neural network model (CNN), recurrent neural network model (RNN), SDD target detection algorithm, SORT target tracking algorithm, DeepSORT target tracking algorithm, optical character recognition algorithm (OCR), digital signal processing algorithm (DSP), hidden Markov model (HMM), YOLO target detection model, Faster R-CNN target detection model, word segmentation algorithm, etc. The feature fusion layer can extract and map features of the data processing results of resource objects to obtain feature vectors of resource objects; the resource classification layer contains preset classification information and classification models, and resource objects can be classified by preset classification information, classification models, and feature vectors.

[0166] It is able to classify resource objects in a multimodal resource object set, and in the classification process it is able to map the features of resource objects of different data types to the same feature space, making the features of resource objects highly comparable, and completing batch classification operations on resource objects in the same feature space, thereby improving classification accuracy while ensuring the efficiency of multimodal resource classification.

[0167] Reference Fig.10 , Fig.10 : is a structural diagram of a resource classification device provided by an embodiment of the present invention. The resource classification device 30 may include:

[0168] An acquisition module 31 is used to acquire a resource object set consisting of resource objects of multiple data types;

[0169] A feature module 32 is used to extract features from each resource object in the resource object set to obtain target features corresponding to each resource object;

[0170] A mapping module 33, used for mapping the target feature to a target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space;

[0171] The classification module 34 is used to input the feature vector set into the target classification model to obtain the target category corresponding to each resource object output by the target classification model.

[0172] Optionally, the feature module includes:

[0173] An extraction strategy submodule, used to determine the target data type corresponding to the resource object, and obtain a target feature extraction strategy corresponding to the target data type;

[0174] The feature extraction submodule is used to extract features from the resource object based on the target feature extraction strategy to obtain target features corresponding to the resource object.

[0175] Optionally, the target feature extraction strategy includes a target structuring strategy and a target feature generation strategy, and the feature extraction submodule includes:

[0176] A structured data unit, used to perform structured processing on the resource object based on the target structuring strategy to obtain target structured data corresponding to the resource object;

[0177] The structured data extraction unit is used to extract features from the target structured data based on the target feature generation strategy to obtain target features corresponding to the resource object.

[0178] Optionally, the target structured strategy includes a target segmentation strategy and a target template, and the structured data unit includes:

[0179] A resource fragment subunit, configured to perform segmentation processing on the resource object based on the target segmentation strategy to obtain a plurality of resource fragments;

[0180] A fragment description information subunit, used to generate fragment description information corresponding to each of the resource fragments;

[0181] An object description information subunit, configured to generate object description information of the resource object based on the fragment description information;

[0182] The structured data generating subunit is used to perform structured processing on the fragment description information and the object description information according to the target template to obtain the target structured data.

[0183] Optionally, the target feature generation strategy includes a first feature model corresponding to a data type of the resource object and a second feature model corresponding to all data types, and the structured data extraction unit includes:

[0184] A first feature subunit, configured to extract features from the target structured data based on the first feature model to obtain a first feature corresponding to the resource object;

[0185] The target feature generation subunit is used to input the first feature into the second feature model to obtain the target feature corresponding to the resource object output by the second feature model.

[0186] Optionally, the target classification model includes a first classification sub-model and a second classification sub-model, and the classification module includes:

[0187] A first classification result submodule, used for inputting the feature vector set into the first classification submodel to obtain a first classification result output by the first classification submodel;

[0188] A second classification result submodule, used for obtaining a second classification result output by the second classification submodel based on the feature vector set and the second classification submodel;

[0189] A target category submodule is used to determine the target category corresponding to each of the resource objects based on the first classification result and the second classification result.

[0190] Optionally, the second classification result submodule comprises:

[0191] A fusion feature vector set unit, used to fuse the feature vector in the feature vector set with the first classification result corresponding to the feature vector to obtain a fused feature vector set;

[0192] A fusion unit is used to input the fused feature vector set into the second classification sub-model to obtain a second classification result output by the second classification sub-model.

[0193] In summary, the resource classification device provided by the embodiment of the present invention can classify resource objects in a multimodal resource object set, and in the classification process can map the features of resource objects of different data types to the same feature space, so that the features of the resource objects have high comparability, and complete batch classification operations on resource objects in the same feature space, thereby improving classification accuracy while ensuring the efficiency of multimodal resource classification.

[0194] Reference Fig.11 , the present invention also provides a structural diagram of an electronic device, such as Fig.11 As shown, it includes: a processor 1101, a memory 1102, and a computer program 11021 stored in the memory and executable on the processor, and the processor implements the resource classification method of the aforementioned embodiment when executing the program.

[0195] The present invention also provides a readable storage medium. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute the resource classification method of the aforementioned embodiment.

[0196] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0197] It should be noted that the various information and data obtained in the embodiments of the present invention are all obtained with the authorization of the information / data holder.

[0198] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0199] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0200] Similarly, it should be understood that in order to streamline the present invention and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, the various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the following intention: that the claimed invention requires more features than the features explicitly recited in each claim. More specifically, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

[0201] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0202] The various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components in the sorting device according to the present invention. The present invention may also be implemented as a device or apparatus program for executing part or all of the methods described herein. Such a program for implementing the present invention may be stored on a computer-readable medium, or may be in the form of one or more signals. Such a signal may be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0203] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim enumerating a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc., does not indicate any order. These words may be interpreted as names.

[0204] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0205] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0206] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A resource classification method, characterized in that: The method comprises: Get a resource object set consisting of resource objects of multiple data types; Extracting features from each resource object in the resource object set to obtain target features corresponding to each resource object; Mapping the target feature to a target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space; The feature vector set is input into a target classification model to obtain a target category corresponding to each resource object output by the target classification model.

2. The method according to claim 1, characterized in that The extracting features of each resource object in the resource object set to obtain target features corresponding to each resource object includes: Determine the target data type corresponding to the resource object, and obtain the target feature extraction strategy corresponding to the target data type; Based on the target feature extraction strategy, feature extraction is performed on the resource object to obtain target features corresponding to the resource object.

3. The method according to claim 2, characterized in that The target feature extraction strategy includes a target structuring strategy and a target feature generation strategy. The feature extraction of the resource object based on the target feature extraction strategy to obtain the target feature corresponding to the resource object includes: Performing structured processing on the resource object based on the target structured strategy to obtain target structured data corresponding to the resource object; Based on the target feature generation strategy, feature extraction is performed on the target structured data to obtain target features corresponding to the resource object.

4. The method according to claim 3, characterized in that The target structuring strategy includes a target segmentation strategy and a target template. The resource object is structured based on the target structuring strategy to obtain target structured data corresponding to the resource object, including: Segmenting the resource object based on the target segmentation strategy to obtain multiple resource fragments; Generate fragment description information corresponding to each of the resource fragments; generating object description information of the resource object based on the fragment description information; The segment description information and the object description information are structured according to the target template to obtain the target structured data.

5. The method according to claim 3, characterized in that: The target feature generation strategy includes a first feature model corresponding to the data type of the resource object and a second feature model corresponding to all data types, and extracting features from the target structured data based on the target feature generation strategy to obtain the target features corresponding to the resource object includes: Extracting features from the target structured data based on the first feature model to obtain a first feature corresponding to the resource object; The first feature is input into a second feature model to obtain a target feature corresponding to the resource object output by the second feature model.

6. The method according to claim 1, characterized in that The target classification model includes a first classification sub-model and a second classification sub-model, and the inputting of the feature vector set into the target classification model to obtain the target category corresponding to each resource object output by the target classification model includes: Inputting the feature vector set into the first classification sub-model to obtain a first classification result output by the first classification sub-model; Based on the feature vector set and the second classification sub-model, obtaining a second classification result output by the second classification sub-model; The target category corresponding to each of the resource objects is determined based on the first classification result and the second classification result.

7. The method according to claim 6, characterized in that The obtaining, based on the feature vector set and the second classification sub-model, a second classification result output by the second classification sub-model comprises: Fusion of the feature vector in the feature vector set with the first classification result corresponding to the feature vector to obtain a fused feature vector set; The fused feature vector set is input into the second classification sub-model to obtain a second classification result output by the second classification sub-model.

8. A resource classification device, characterized in that: The device comprises: An acquisition module, used to acquire a resource object set consisting of resource objects of multiple data types; A feature module, used to extract features from each resource object in the resource object set to obtain target features corresponding to each resource object; A mapping module, used to map the target feature to a target feature space to obtain a feature vector set corresponding to the resource object set in the target feature space; The classification module is used to input the feature vector set into the target classification model to obtain the target category corresponding to each resource object output by the target classification model.

9. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the program.

10. A readable storage medium, characterized in that: When the instructions in the readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute any one of the methods described in claims 1-7.

Images