Category recognition method and device

Abstract

This application discloses a category recognition method and apparatus. The method includes: acquiring target data when preset behavioral data of a target object is detected; performing subject recognition on each image in at least one image to obtain one or more subjects, and classifying each subject to obtain a category for each subject; filtering the one or more subjects according to one or more subject filtering strategies to obtain filtered subjects, the filtered subjects including at least one candidate subject; determining the target subject and its corresponding integrated category from the filtered subjects based on the category of each candidate subject, and using the integrated category as the category recognition result of the target data; and outputting the category recognition result of the target data. This application embodiment can obtain target subjects that better match the target data and improve the accuracy of the category recognition result.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a category identification method and apparatus. Background Technology

[0002] Currently, with the continuous development of computer technology, category recognition tasks have been widely applied. For example, category recognition can be performed on target data of interest to a target object to identify the category that the target object is interested in (i.e., the category recognition result of the target data), or the category recognition result can be used to determine the matching category under the matching category system. Existing technologies typically use category recognition models to identify the categories of subjects in the target data to obtain category recognition results. However, target data may include multiple subjects, and the subjects may be vague or have complex categories, making it difficult to determine the target subject among multiple subjects, resulting in low accuracy of the output category recognition results. Therefore, how to determine the target subject and improve the accuracy of category recognition results has become a research hotspot. Summary of the Invention

[0003] This application provides a category identification method and apparatus, which can filter one or more subjects through one or more subject filtering strategies, and determine the target subject that is more compatible with the target data and the integrated classification category corresponding to the target subject (i.e. the category identification result of the target data) from the filtered subjects, thereby improving the accuracy of the category identification result.

[0004] On the one hand, embodiments of this application provide a category identification method, the method comprising:

[0005] When preset behavior data of a target object in response to target data is detected, the target data is acquired; wherein, the target data includes at least one image;

[0006] Subject recognition is performed on each image in the at least one image to obtain one or more subjects, and each of the one or more subjects is classified to obtain the classification category of each subject;

[0007] According to one or more subject filtering strategies, the one or more subjects are filtered to obtain filtered subjects. The filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than the matching degree between the filtered subjects and the target data.

[0008] Based on the classification categories of each candidate subject in the filtered subjects, the target subject and the integrated classification category corresponding to the target subject are determined from the filtered subjects, and the integrated classification category is used as the category identification result of the target data. The matching degree between the target subject and the target data is greater than the matching degree between the subjects other than the target subject in the filtered subjects and the target data.

[0009] Output the category identification results of the target data.

[0010] On the other hand, embodiments of this application provide a category identification device, the device comprising:

[0011] An acquisition unit is configured to acquire the target data when preset behavioral data of a target object in relation to the target data is detected; wherein the target data includes at least one image;

[0012] The processing unit is configured to perform subject recognition on each image in the at least one image to obtain one or more subjects, and to perform classification processing on each of the one or more subjects to obtain the classification category of each subject;

[0013] The processing unit is further configured to perform filtering processing on the one or more subjects according to one or more subject filtering strategies to obtain filtered subjects, wherein the filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than the matching degree between the filtered subjects and the target data.

[0014] The processing unit is further configured to determine the target subject and the integrated classification category corresponding to the target subject from the filtered subjects based on the classification categories of each candidate subject in the filtered subjects, and use the integrated classification category as the category identification result of the target data, wherein the matching degree between the target subject and the target data is greater than the matching degree between the subjects other than the target subject in the filtered subjects and the target data;

[0015] The output unit is used to output the category identification result of the target data.

[0016] In another aspect, embodiments of this application provide a computer device, the computer device including a processor and a memory, wherein the memory is used to store a computer program, and the computer program, when executed by the processor, performs the following steps:

[0017] When preset behavior data of a target object in response to target data is detected, the target data is acquired; wherein, the target data includes at least one image;

[0018] Subject recognition is performed on each image in the at least one image to obtain one or more subjects, and each of the one or more subjects is classified to obtain the classification category of each subject;

[0019] According to one or more subject filtering strategies, the one or more subjects are filtered to obtain filtered subjects. The filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than the matching degree between the filtered subjects and the target data.

[0020] Based on the classification categories of each candidate subject in the filtered subjects, the target subject and the integrated classification category corresponding to the target subject are determined from the filtered subjects, and the integrated classification category is used as the category identification result of the target data. The matching degree between the target subject and the target data is greater than the matching degree between the subjects other than the target subject in the filtered subjects and the target data.

[0021] Output the category identification results of the target data.

[0022] In another aspect, embodiments of this application provide a computer storage medium storing a computer program adapted for loading by a processor and executing the following steps:

[0023] When preset behavior data of a target object in response to target data is detected, the target data is acquired; wherein, the target data includes at least one image;

[0024] Subject recognition is performed on each image in the at least one image to obtain one or more subjects, and each of the one or more subjects is classified to obtain the classification category of each subject;

[0025] According to one or more subject filtering strategies, the one or more subjects are filtered to obtain filtered subjects. The filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than the matching degree between the filtered subjects and the target data.

[0026] Based on the classification categories of each candidate subject in the filtered subjects, the target subject and the integrated classification category corresponding to the target subject are determined from the filtered subjects, and the integrated classification category is used as the category identification result of the target data. The matching degree between the target subject and the target data is greater than the matching degree between the subjects other than the target subject in the filtered subjects and the target data.

[0027] Output the category identification results of the target data.

[0028] In another aspect, embodiments of this application provide a computer program product, which includes a computer program that, when executed by a processor, implements the category identification method mentioned above.

[0029] This application embodiment can acquire target data, which includes at least one image, when it detects preset behavior data of a target object targeting target data. Based on this, subject recognition can be performed on each image in the at least one image to obtain one or more subjects, and each subject in the one or more subjects can be classified to obtain the classification category of each subject. Then, one or more subjects can be filtered according to one or more subject filtering strategies to obtain filtered subjects, which include at least one candidate subject. Based on the classification category of each candidate subject in the filtered subjects, a target subject that matches the target data better and the integrated classification category corresponding to the target subject can be determined from the filtered subjects. The integrated classification category is used as the category recognition result of the target data, and the category recognition result of the target data is output, which can effectively improve the accuracy of the category recognition result. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1a This is a flowchart illustrating a category identification scheme provided in an embodiment of this application;

[0032] Figure 1b This is a schematic diagram illustrating the interaction between a terminal and a server, provided in an embodiment of this application.

[0033] Figure 2 This is a flowchart illustrating a category identification method provided in an embodiment of this application;

[0034] Figure 3a This is a schematic diagram of a main body provided in an embodiment of this application;

[0035] Figure 3b This is a schematic diagram of another type of subject provided in an embodiment of this application;

[0036] Figure 3c This is a schematic diagram of another type of subject provided in an embodiment of this application;

[0037] Figure 3d This is a schematic diagram of another type of subject provided in an embodiment of this application;

[0038] Figure 3e This is a schematic diagram of another type of subject provided in an embodiment of this application;

[0039] Figure 3f This is a schematic diagram of another type of subject provided in an embodiment of this application;

[0040] Figure 3g This is a schematic diagram of a classification category provided in an embodiment of this application;

[0041] Figure 4 This is a flowchart illustrating another category identification method provided in an embodiment of this application;

[0042] Figure 5 This is a flowchart illustrating another category identification method provided in the embodiments of this application;

[0043] Figure 6 This is a schematic diagram of the structure of a category identification device provided in an embodiment of this application;

[0044] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0045] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0046] To obtain target subjects that better match the target data and improve the accuracy of category identification results for the target data, this application proposes a category identification scheme; see [link to relevant documentation]. Figure 1a As shown, the general principle of the category recognition scheme proposed in this application embodiment is as follows: First, when the preset behavior data of the target object against the target data is detected, the target data can be obtained. The target data may include at least one image. Then, one or more subjects can be identified from the images included in the target data, and the classification category of each subject can be obtained. Thus, one or more subjects can be filtered according to one or more subject filtering strategies, and the target subject and the integrated classification category corresponding to the target subject can be determined from the filtered subjects. The integrated classification category is used as the category recognition result of the target data, and then the category recognition result is output.

[0047] Practice has shown that the category identification scheme proposed in this application can have at least the following beneficial effects: ① It can obtain a target subject that is more closely matched with the target data, so that the integrated classification category corresponding to the target subject can more accurately represent the category of the target data, thereby obtaining the category identification result of the target data; ② It can effectively improve the accuracy of the category identification result of the target data.

[0048] In its implementation, the category identification scheme mentioned above can be executed by a computer device, which can be a terminal or a server. The terminal mentioned here can include, but is not limited to, smartphones, tablets, laptops, desktop computers, smartwatches, smart voice interaction devices, smart home appliances, vehicle terminals, and aircraft. Various clients (apps) can run on the terminal, such as video playback clients, social media clients, browser clients, news feed clients, educational clients, and so on. The server mentioned here can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms, etc. Furthermore, the computer device mentioned in the embodiments of this application can be located outside or inside the blockchain network, and there is no limitation on this. The so-called blockchain network is a network composed of a peer-to-peer network (P2P network) and a blockchain. The blockchain refers to a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm. In essence, it is a decentralized database, which is a series of data blocks (or blocks) linked together using cryptographic methods.

[0049] Alternatively, in other embodiments, the category identification scheme mentioned above can also be jointly executed by a server and a terminal; the terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited herein. For example, the terminal can be responsible for acquiring target data and sending it to the server, so that the server can acquire one or more subjects and their respective classification categories based on the target data, thereby filtering one or more subjects to determine the target subject and its corresponding integrated classification category from the filtered subjects, and sending the integrated classification category as the category identification result of the target data to the terminal, which then outputs the category identification result, such as... Figure 1bAs shown. For example, the terminal can acquire target data, obtain one or more entities and their corresponding category categories based on the target data, and then filter these entities to obtain filtered entities. The filtered entities are then sent to the server, which can then determine the target entity and its corresponding integrated category from the filtered entities. The server uses this integrated category category as the category identification result for the target data and sends it back to the terminal, which then outputs the category identification result, and so on. It should be understood that this is merely an illustrative example of two scenarios where the terminal and server jointly execute the above category identification scheme; it is not an exhaustive list.

[0050] Based on the above description of the category identification scheme, this application proposes a category identification method, which can be executed by the aforementioned computer device (terminal or server); or, the category identification method can be executed jointly by the terminal and the server. For ease of explanation, the following description will use the execution of the category identification method by a computer device as an example; please refer to [link to relevant documentation]. Figure 2 The category identification method may include the following steps S201-S205:

[0051] S201, when the preset behavior data of the target object in relation to the target data is detected, the target data is acquired; wherein the target data includes at least one image.

[0052] The aforementioned preset behavioral data can be data generated by screenshotting, data generated by saving, data generated by liking, etc.; this application does not limit this. For example, taking data generated by screenshotting as an example, suppose that when a target object is browsing a target application (such as a community application for sharing), the target object can take a screenshot of the interface containing the target data (i.e., the notes composed of the target data). Then, the computer device can detect the preset behavioral data of the target object regarding the target data (i.e., the data generated by screenshotting), thereby obtaining the target data. For ease of explanation, the following description will use data generated by screenshotting as an example.

[0053] In the embodiments of this application, the target data mentioned above includes, but is not limited to: at least one image, target category (a category used to reflect the subject information of the target data, also known as a note category), and category words extracted from text structure (i.e., text category words), etc.; this application does not limit this.

[0054] S202, perform subject recognition on each image in at least one image to obtain one or more subjects, and classify each subject in the one or more subjects to obtain the classification category of each subject.

[0055] Specifically, the computer device can invoke a visual basic model module to perform recognition processing on various images (i.e., pictures) to obtain one or more subjects and their classification categories. This visual basic model can also be called a visual model, and the visual basic model module includes, but is not limited to: detection models, subject models, subject classification models (i.e., classification models), and subject ranking models, etc. Based on this, the computer device can invoke the detection model and the subject model to perform subject recognition on various images. That is, the computer device can first invoke the detection model to perform subject detection on various images, obtaining subject detection boxes in each image, and then invoke the subject model to perform subject recognition on the subject detection boxes in each image, to obtain one or more subjects. The models in the visual basic model module can be convolutional neural networks, recurrent neural networks, etc.; this application does not limit this. It should be noted that the computer device can also obtain the detection categories of each subject detection box through the detection model, thereby obtaining the detection categories of each subject, etc.

[0056] For example, such as Figure 3a As shown, assuming the target data includes an image containing three main subjects (i.e., toothpaste), then a computer device can identify the three main subjects from the image included in the target data; for example, ... Figure 3b As shown, assuming the target data includes an image comprising four parts, and these four parts contain a total of 12 main objects (i.e., shampoo), then the computer device can identify the 12 main objects from the image included in the target data; similarly, as... Figure 3c and Figure 3d As shown, the computer device can identify two subjects separately, and so on.

[0057] Furthermore, the computer device can invoke the subject classification model to classify each subject among one or more subjects, obtaining the classification categories for each subject, and thus outputting the classification categories for each subject (i.e., the algorithm category system). The number of category levels can be four or five, etc., and this application does not limit this; when the number of category levels is four, this application may also refer to the subject's classification categories as subject fourth-level classification categories, and so on. The number of category levels refers to the number of hierarchical levels of categories included in a single classification category.

[0058] It should be noted that when the visual basic model module includes a subject sorting model, the computer device can further call the subject sorting model to sort the subjects according to their position information and / or display area in the image, and output the subjects according to the sorting results.

[0059] For example, such as Figure 3e and Figure 3fAs shown, assuming the computer device sorts the subjects according to their display area in the image, and the subject detection box numbered 0 (i.e., the subject detection box where the clothes are located) has the largest display area, then when sorting according to the display area from largest to smallest, the computer device can put the subject corresponding to the subject detection box numbered 0 at the front.

[0060] S203, according to one or more subject filtering strategies, filter one or more subjects to obtain filtered subjects, the filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than the matching degree between the filtered subjects and the target data.

[0061] Specifically, computer equipment can filter one or more subjects and their respective category groups according to one or more subject filtering strategies, resulting in filtered subjects and filtered category groups. The filtered subjects include at least one candidate subject, and the filtered category groups include the category groups of each candidate subject. In other words, after filtering one or more subjects and obtaining the filtered subjects, only the category groups of each candidate subject in the filtered subjects can be retained.

[0062] S204. Based on the classification categories of each candidate subject in the filtered subjects, determine the target subject and the integrated classification category corresponding to the target subject from the filtered subjects, and use the integrated classification category as the category identification result of the target data. The matching degree between the target subject and the target data is greater than the matching degree between the subjects other than the target subject in the filtered subjects and the target data.

[0063] The integrated category corresponding to the target entity can refer to: the category resulting from the integration of at least one level category from the target entity's existing categories, or it can refer to: the target entity's existing categories, etc.; this application does not limit this. For example, such as Figure 3g As shown, the integrated category corresponding to the target entity can refer to the category after at least one level of the target entity's category has been integrated. For example, assuming the target entity includes entity A and entity B, entity A's category is Other_Drying Agents_Oral Care_Toothpaste, and entity B's category is Other_Drying Agents_Clothing Care_Laundry Detergent. When the first two levels of the category are integrated (i.e., deduplicated), the first three levels of the integrated category corresponding to the target entity can be: Other_Drying Agents_Oral Care / Clothing Care. Furthermore, the fourth level under the third level category (i.e., the third-level category) "Oral Care" is toothpaste, and the fourth level under the third level category "Clothing Care" is laundry detergent.

[0064] S205, outputs the category identification results of the target data.

[0065] It should be noted that the embodiments of this application can be applied to e-commerce scenarios and recommendation scenarios, etc.; this application does not limit them. For example, in a recommendation scenario, a computer device can identify the category identification result of the target data and use the category identification result as the category that the target object is interested in, thereby recommending data with the same category as the category identification result to the target object.

[0066] For example, in e-commerce scenarios, the system can automatically and in real-time convert the target user's screenshot behavior on the interface containing the target data into their purchase intent. This helps merchants establish daily sales and promotional areas, linking target users and merchants to solve the problem of not being able to purchase goods in community applications used for sharing. Specifically, computer devices can identify the category recognition results of the target data and map these results to a matching category system to obtain the matching category (i.e., e-commerce category information in the e-commerce scenario). This enables intelligent conversion of the target user's screenshot behavior into e-commerce category information, allowing merchants to send corresponding product data (such as promotional products, discount information, etc.) to the target user based on the matching category. In e-commerce scenarios, this application can also refer to the method of obtaining the matching category as a mapping technology from user-generated content (UGC) to the category system of goods with transactional value, i.e., e-commerce category mapping technology.

[0067] For example, such as Figure 3d As shown, the category recognition result (i.e., prototype, model) is Accessories_Watches & Eyeglasses_Functional Eyeglasses_Sunglasses. Then, the computer device can map the category recognition result to the matching category system, and obtain the matching category (i.e., good) under the matching category system as Clothing, Shoes & Bags_Eyeglasses / Accessories_Functional Eyeglasses_Sunglasses.

[0068] This application embodiment can acquire target data, which includes at least one image, when it detects preset behavior data of a target object targeting target data. Based on this, subject recognition can be performed on each image in the at least one image to obtain one or more subjects, and each subject in the one or more subjects can be classified to obtain the classification category of each subject. Then, one or more subjects can be filtered according to one or more subject filtering strategies to obtain filtered subjects, which include at least one candidate subject. Based on the classification category of each candidate subject in the filtered subjects, a target subject that matches the target data better and the integrated classification category corresponding to the target subject can be determined from the filtered subjects. The integrated classification category is used as the category recognition result of the target data, and the category recognition result of the target data is output, which can effectively improve the accuracy of the category recognition result.

[0069] Please see Figure 4 This is a flowchart illustrating another category identification method provided in this application. This category identification method can be executed by the computer device (terminal or server) mentioned above; or, it can be executed jointly by the terminal and the server. For ease of explanation, the following description will use the execution of this category identification method by a computer device as an example; please refer to [link to relevant documentation]. Figure 4 The category identification method may include the following steps S401-S407:

[0070] S401, when the preset behavior data of the target object in relation to the target data is detected, the target data is acquired; wherein the target data includes at least one image.

[0071] S402, Obtain data filtering instruction information, which includes at least one of the following: value assessment data and filtering categories, wherein the value assessment data includes entities with value.

[0072] It should be noted that target data and data filtering instructions can be collectively referred to as data source, that is, data source can include target data and data filtering instructions; among them, when the value judgment data is presented in tabular form, it can also be called a note value table, which is used to reflect the transaction value information of the corresponding subject, such as the transaction value information of each product in e-commerce scenarios.

[0073] S403, based on the data filtering instruction information, perform data filtering processing on the target data in order to identify the category of the target data that has passed the data filtering processing. The target data that has passed the data filtering processing refers to: target data that includes valuable subjects and / or target data that includes target categories other than the filtered categories.

[0074] In this embodiment, since the target data may contain a large amount of non-transactional data (i.e., data without valuable components), meaning the target data may not include any valuable components, it may lack category identification significance in certain scenarios. Therefore, the computer device can filter out the target data, thus avoiding category identification and saving bandwidth and preventing resource waste. For example, in an e-commerce scenario, if the target data contains a beach, and the beach has no value (i.e., it is not a commodity), the computer device can filter out the target data. It should be noted that the filtering categories can be set based on experience or actual needs; this application does not limit this.

[0075] Specifically, target data may include target categories, which are categories used to reflect the thematic information of the target data. Therefore, when the data filtering instruction information includes a filtering category, and the target data includes target categories other than the filtering category (i.e., target categories different from the filtering category), the target data can be processed through data filtering. When the data filtering instruction information includes value assessment data, and the images in the target data include valuable subjects (i.e., the target data includes valuable subjects), the target data can be processed through data filtering. When the data filtering instruction information includes both value assessment data and a filtering category, and the target data includes target categories other than the filtering category, and the images in the target data include valuable subjects, the target data can be processed through data filtering, and so on. It should be noted that when the data filtering instruction information includes value assessment data, the process of filtering the target data according to the value assessment data in the data filtering instruction information can also be called data value filtering (i.e., note value filtering). When the data filtering instruction information includes a filtering category, the process of filtering the target data according to the filtering category in the data filtering instruction information can also be called target category (i.e., note category) filtering.

[0076] Optionally, the computer device may include a pre-filtering module. In this case, the computer device can perform data pre-filtering processing (i.e., data filtering processing) on ​​the target data through the pre-filtering module. This can be done by using value judgment data and filtering categories to filter out target data with transaction value, which can cover about 90% of the category data of e-commerce and notes.

[0077] S404, perform subject recognition on each image in at least one image to obtain one or more subjects, and classify each subject in the one or more subjects to obtain the classification category of each subject.

[0078] S405, according to one or more subject filtering strategies, filter one or more subjects to obtain filtered subjects, the filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than the matching degree between the filtered subjects and the target data.

[0079] In the embodiments of this application, the above-mentioned one or more subject filtering strategies may include at least one of the following: category inconsistency filtering strategy (i.e., detection of category & classification category inconsistency filtering strategy), mutually exclusive category filtering strategy, specified category filtering strategy (i.e., special category filtering strategy), and category consistency filtering strategy (i.e., image subject and text category consistency filtering strategy), etc.; this application does not limit this. Each subject filtering strategy can be executed by a corresponding module; for example, the category inconsistency filtering strategy can be executed by the category inconsistency filtering module (i.e., detection of category & classification category inconsistency filtering module), the mutually exclusive category filtering strategy can be executed by the mutually exclusive category module, the specified category filtering strategy can be executed by the specified category filtering module, etc.

[0080] In one implementation, the one or more subject filtering strategies include at least a category inconsistency filtering strategy. The computer device can then obtain the detection categories of each subject and, based on the category inconsistency filtering strategy, identify inconsistent subjects from the one or more subjects whose detection categories and classification categories are inconsistent. Then, the inconsistent subjects are filtered out from the one or more subjects to obtain the filtered subjects. The detection categories of each subject can be obtained through the aforementioned detection model, thereby obtaining the detection categories of each subject based on the output of the detection model.

[0081] It should be noted that when the detection category (i.e., the detection class) and the classification category (i.e., the classification category) differ significantly, the subject may be unclear, affecting the accuracy of predicting the target object's intent. Therefore, the category inconsistency filtering strategy can filter out subjects whose detection category and classification category are inconsistent, thereby improving the quality of the filtered subjects and thus improving the accuracy of the category identification results. In the embodiments of this application, inconsistency between the detection category and the classification category can refer to the detection category and the classification category being different, or it can refer to the category type involved in the detection category being different from the category type involved in the classification category, etc.; this application does not limit this.

[0082] In a specific implementation, if the detection category and the classification category are inconsistent, it means that the detection category and the classification category are different. Then, when the category levels of the detection category and the classification category are different, the inconsistency between the detection category and the classification category can mean that the corresponding category in the detection category is different from that in the classification category. For example, if the category level of the detection category is one level and the category level of the classification category is four levels, and the detection category corresponds to the first-level category (i.e., the first-level category) of the classification category, if the detection category and the first-level category of the classification category are the same, then the detection category and the classification category are consistent. If the detection category and the first-level category of the classification category are different, then the detection category and the classification category are inconsistent.

[0083] In another specific implementation, if the inconsistency between the detection category and the classification category means that the category type involved in the detection category is different from that involved in the classification category, then when the category levels of the detection category and the classification category are different, the inconsistency between the detection category and the classification category can mean that the category type involved in the detection category is different from that involved in the corresponding category in the classification category. For example, when the detection category has a level 1 classification and the categorization category has a level 4 classification, and the detection category corresponds to the level 1 classification category, suppose an entity's detection category is "clothing," and the entity's categorization category has a level 1 classification category such as "tops, pants, or skirts," and suppose the category type corresponding to "clothing" is "apparel," and the category type corresponding to "tops, pants, and skirts" is "apparel," then in this case, the category type involved in the entity's detection category is the same as the category type involved in the entity's categorization category's level 1 classification. That is, the category type involved in the entity's detection category is the same as the category type involved in the entity's categorization category, thus confirming that the entity's detection category and categorization category are consistent. For example, suppose a subject's detection category is "clothing," and its primary category for classification is "mobile phones, etc." Furthermore, suppose the category type for "clothing" is "clothing," and the category type for "mobile phones" is "communication devices." In this case, the category types involved in the subject's detection category are different from the category types involved in the primary category of its classification. Therefore, it can be determined that the subject's detection category and classification category are inconsistent.

[0084] In another implementation, one or more strategies include at least a mutually exclusive category filtering strategy, and the target data includes target categories, which are categories used to reflect the subject information of the target data. Based on this, the computer device can obtain category mapping information based on the mutually exclusive category filtering strategy. The category mapping information is used to indicate the relevance and mutual exclusion between various categories in multiple categories. Further, based on the category mapping information, relevant subjects can be found from one or more subjects. The relevance between the detection category of the relevant subject and the target category is greater than the relevance between the detection category of the subjects other than the relevant subjects and the target category. Then, the computer device can find mutually exclusive subjects whose detection categories are mutually exclusive with the detection categories of the relevant subjects from one or more subjects, and filter out the mutually exclusive subjects from one or more subjects to obtain the filtered subjects.

[0085] In this scenario, the computer device can retain only the entities corresponding to the detection categories that are strongly relevant to the target category. For example, suppose the category mapping information indicates that beauty products are mutually exclusive with pets, clothing, shoes, furniture, accessories, and others. Since clothing, shoes, furniture, and accessories are not included in the beauty product category, if one or more entities have a related entity whose detection category is beauty products, and one or more entities include an entity whose detection category is clothing (i.e., mutually exclusive entities), then the computer device can filter out entities whose detection category is clothing from one or more entities, ensuring that the filtered entities do not include entities whose detection category is clothing. When the category mapping information exists in tabular form, it can be called a category mapping table.

[0086] In another implementation, one or more subject filtering strategies include at least a specified category filtering strategy, and a category includes multiple levels of categories. Based on this, the computer device can obtain preset specified category setting information based on the specified category filtering strategy. This specified category setting information includes at least one specified category and the restricted categories under each specified category. Further, a specified filtering subject can be found among one or more subjects. The category of the specified filtering subject includes the target specified category in the specified category setting information and any restricted category under the target specified category. The specified filtering subject is then filtered from one or more subjects to obtain the filtered subject. The specified category setting information can be set based on experience or based on actual needs; this application does not limit this.

[0087] It should be noted that computer equipment can perform special processing on categories with high error rates in classification models, such as maternal and infant products, fashion activities, nail salons, digital products, masks, and accessories. These categories can be set as designated categories. For example, the designated category of "fashion activities" can only include ["clothing," "bags," "hats," "outdoor / mountain / travel camping supplies"], excluding masks. In other words, masks can be a restricted category under "fashion activities." For instance, assuming the designated category is "fashion activities" and the restricted category under "fashion activities" is masks, when one or more subjects include the designated filtering subject, the designated filtering subject can be filtered out from one or more subjects, so that the filtered subjects do not include the designated filtering subject. In this case, the classification category of the designated filtering subject includes the target designated category (i.e., fashion activities) and any restricted category under the target designated category (i.e., masks).

[0088] In another implementation, one or more subject filtering strategies include at least a category consistency filtering strategy, and the target data also includes text category terms. Based on this, the computer device can determine, based on the category consistency filtering strategy, a category matching subject whose classification category matches the text category term among one or more subjects, wherein any first-level category in the classification category of the category matching subject is the same as any category term in the text category term; and the category matching subject is used as the filtered subject.

[0089] It should be noted that when category terms exist in the text (coverage around 20%), the target data can include textual category terms (i.e., category words) extracted from the text. This allows for filtering based on a category consistency strategy, ensuring that the filtered subjects only include those whose category matches the textual category terms. Here, "category matching" means that any first-level category in the category matches any category term in the text.

[0090] Furthermore, when one or more subject filtering strategies include a category inconsistency filtering strategy and a mutually exclusive category filtering strategy, the computer device can determine inconsistent subjects from one or more subjects based on the category inconsistency filtering strategy, find mutually exclusive subjects from one or more subjects whose detection categories are mutually exclusive with the detection categories of related subjects based on the mutually exclusive category filtering strategy, and filter out inconsistent subjects and mutually exclusive subjects from one or more subjects to obtain the filtered subjects; when one or more subject filtering strategies include a category inconsistency filtering strategy and a specified category filtering strategy, the computer device can determine inconsistent subjects from one or more subjects based on the category inconsistency filtering strategy, and find mutually exclusive subjects from one or more subjects based on the specified category filtering strategy. The category filtering strategy identifies a specified filtering subject from one or more subjects and filters out inconsistent subjects and the specified filtering subject from one or more subjects. When the filtering strategy for one or more subjects includes a mutually exclusive category filtering strategy and a specified category filtering strategy, the computer device can filter out mutually exclusive subjects and the specified filtering subject from one or more subjects to obtain the filtered subjects. When the filtering strategy for one or more subjects includes a category inconsistency filtering strategy, a mutually exclusive category filtering strategy, and a specified category filtering strategy, the computer device can filter out inconsistent subjects, mutually exclusive subjects, and the specified filtering subject from one or more subjects to obtain the filtered subjects, and so on. This application does not limit this aspect.

[0091] In another implementation, one or more subject filtering strategies may include at least a first subject filtering strategy and a second subject filtering strategy. The first subject filtering strategy includes at least one of the following: a category inconsistency filtering strategy, a mutually exclusive category filtering strategy, and a specified category filtering strategy. The second subject filtering strategy includes at least a category consistency filtering strategy. Based on this, the computer device can filter one or more subjects based on the first subject filtering strategy to obtain a first filtered subject; and filter one or more subjects based on the second subject filtering strategy to obtain a second filtered subject; thereby using the first filtered subject and the second filtered subject as the filtered subjects.

[0092] The first and second subject filtering strategies can each be a branch. Specifically, addressing the issue of slightly low coverage of text category words in notes (around 20%), this application designs a visual subject (i.e., image subject) filtering branch (i.e., visual multimodal subject filtering fusion branch) incorporating the first subject filtering strategy. This branch can obtain the model category output and model subject output of a single visual branch without text category words through at least one of the following modules: mutually exclusive category filtering module, category detection & classification category inconsistency filtering module, and specified category filtering module. Correspondingly, this application also designs an image subject (i.e., visual subject) & text category consistency branch incorporating the second subject filtering strategy. When text category words exist in the text, the computer device can directly use the image subject & text category consistency branch. Thus, when the classification category of the image subject matches the text category words, it outputs all consistent categories and corresponding category-matching subjects, such as... Figure 5 As shown.

[0093] S406, based on the classification categories of each candidate subject in the filtered subjects, determine the target subject and the integrated classification category corresponding to the target subject from the filtered subjects, and use the integrated classification category as the category identification result of the target data. The matching degree between the target subject and the target data is greater than the matching degree between the subjects other than the target subject in the filtered subjects and the target data.

[0094] In a specific implementation, the classification categories of any subject include M-level categories, where M is an integer greater than 1. When the computer device determines the target subject and the integrated classification category corresponding to the target subject from the filtered subjects based on the classification categories of each candidate subject, it can perform deduplication processing on the m-th level category of each candidate subject's classification category to obtain the deduplicated m-th level category, where m∈[1,H], and H is an integer greater than or equal to 1 and less than M. Then, the display parameters of each candidate subject are calculated respectively, and based on the display parameters of each candidate subject, the candidate subjects whose display parameters are greater than the parameter threshold are determined from the filtered subjects to obtain the target subject. Further, the deduplicated first H-level categories corresponding to the target subject can be used as the first H-level categories of the integrated classification category, and the i-th level category of the target subject can be used as the i-th level category of the integrated classification category, where i∈[H+1,M].

[0095] The display parameters can be the number of similar subjects among one or more subjects, the proportion of similar subjects among one or more subjects, the number of similar subjects among filtered subjects, or the display area of ​​a subject in the image, etc.; this application does not limit this. Similar subjects can refer to subjects in the same M-level category of a classification category, or subjects in all levels of the classification category (i.e., subjects in the same classification category); this application does not limit this. In this embodiment, if the display parameter is the proportion of similar subjects among one or more subjects, or the proportion of the display area in the image, the display parameter can also be called the display probability, to select subjects with high probability from the filtered subjects. Furthermore, the above parameter thresholds can be set based on experience or based on actual needs; this application does not limit this.

[0096] For example, let's take the display parameter being the number of similar subjects among one or more subjects, where similar subjects refer to subjects with the same M-level category in the classification category. Suppose that one or more subjects include subject A, subject B, subject C, and subject D, and the M-level categories in the classification categories of subject A, subject B, subject C, and subject D are: toothpaste, water cup, and laundry detergent, respectively. Then, the display parameter for subject A is 2, the display parameter for subject B is 2, the display parameter for subject C is 1, and the display parameter for subject D is 1. In this case, when the filtered subjects include subject A, subject B, and subject C, and the parameter threshold is 1, the computer device can determine subject A and subject B from the filtered subjects to obtain the target subject.

[0097] Furthermore, when determining candidate subjects whose display parameters are greater than a parameter threshold from the filtered subjects based on the display parameters of each candidate subject, in order to obtain the target subject, the computer device can determine candidate subjects whose display parameters are greater than a parameter threshold from the filtered subjects based on the display parameters of each candidate subject, in order to obtain a candidate subject set; and perform deduplication processing on the M-level category of the classification category of the candidate subjects in the candidate subject set, so as to select leaf category deduplicated subjects from the candidate subject set, and select the target subject from the leaf category deduplicated subjects; wherein, the number of leaf category deduplicated subjects is equal to the number of deduplicated M-level categories, and a leaf category deduplicated subject is determined according to the sorting position of the candidate subjects with the same M-level category in the target sorting result.

[0098] The target ranking result can refer to the ranking result obtained through the above-mentioned subject ranking model, or it can be obtained by ranking the number of subjects with the same classification category in historical data, or it can be obtained by ranking the number of subjects with the same classification category in the candidate subject set. This application does not limit this. The historical data can include subjects that have undergone classification processing before the current system time, or it can include subjects that have undergone classification processing within a preset time period. This application does not limit this. It should be noted that the target ranking result can be arranged in descending order or in ascending order; this application does not limit this. When the target ranking result is arranged in descending order, the deduplicated leaf category subjects among the candidate subjects with the same M-level category in the candidate subject set will be ranked before other candidate subjects with the same M-level category in the target ranking result. When the target ranking result is arranged in ascending order, the deduplicated leaf category subjects among the candidate subjects with the same M-level category in the candidate subject set will be ranked before other candidate subjects with the same M-level category in the target ranking result. Here, "other candidate subjects" refers to candidate subjects other than the deduplicated leaf category subjects among the candidate subjects with the same M-level category.

[0099] For example, let's take the case where the target ranking result is based on the number of entities with the same category in historical data. Assume that there are 100 entities with the category "Other_Hydrating Agents_Oral Care_Toothpaste" and 10 entities with the category "Other_Baby & Maternity_Oral Care_Toothpaste" in historical data. Candidate entity A has the category "Other_Hydrating Agents_Oral Care_Toothpaste" and candidate entity B has the category "Other_Baby & Maternity_Oral Care_Toothpaste". When the target ranking result is arranged in descending order, the computer device can determine that candidate entity A is ranked before candidate entity B in the target ranking result. Therefore, candidate entity A is selected as a leaf category deduplicated entity among the candidates with the Mth level category "Toothpaste".

[0100] It should be noted that the deduplication process of the computer device for the first H levels of categories, and the process of determining candidate subjects based on display parameters, can be referred to as the same-category deduplication logic. For example, when H is 2, the computer device can prioritize deduplication of the first and second levels of the categorized categories to obtain the first two levels of categories after deduplication, and output the third and fourth level categories of candidate subjects whose display parameters are greater than the parameter threshold. Correspondingly, the process of the computer device deduplicating the Mth level of the categorized categories can be referred to as leaf category deduplication, thereby retaining a limited number of the top (first) subjects (i.e., leaf category deduplicated subjects) at the first M-1 levels. For example, when H is 2, the first, second, and third level categories of the top subjects can be retained preferentially. Optionally, the computer device may include a deduplication merging logic module, and use this module to deduplicate the filtered subjects to obtain leaf category deduplicated subjects.

[0101] Furthermore, when selecting the target subject from the deduplicated leaf category subjects, the computer device can use the deduplicated leaf category subjects as the target subjects; alternatively, it can truncate the deduplicated leaf category subjects to select the top N subjects (i.e., top N subjects) and use these selected subjects as the target subjects, where N is a positive integer less than the number of deduplicated leaf category subjects, and so on. Specifically, the computer device can select the top N subjects from the deduplicated leaf category subjects based on the truncation and sorting results. This truncation and sorting result can be the same as or different from the target sorting result mentioned above; this application does not limit this. For example, when N is 3, the computer device can output the integrated classification categories corresponding to the top three subjects (i.e., target subjects) in the deduplicated leaf category subjects, at which point the basic subject results can all be output.

[0102] S407 outputs the category identification results of the target data.

[0103] It should be noted that in e-commerce and other scenarios requiring category mapping, after outputting the category identification results of the target data, the computer device can use fuzzy text matching (a logic of matching progressively upwards from leaf nodes) to associate the category identification results with the matching categories, and use this as the final output, thereby establishing the association between the UGC graph and the matching categories. It should also be noted that the matching categories can include Q-level categories, where Q is a positive integer; for example, when Q is 5 and the category includes four levels, the model can predict the association between the level 4 categories and the matching category (level 5).

[0104] This application embodiment can perform data filtering processing on target data based on the data filtering instruction information after obtaining the data filtering instruction information, thereby identifying the category of the target data that has passed the data filtering processing. This can effectively save traffic and avoid resource waste. Furthermore, this application embodiment can solve the technical difficulties of how to retain the subject in at least one image included in the target data, how many subjects to retain, and the inconsistency between the detected category and the classification category by complementarity between weakly supervised information (i.e., various subject filtering strategies, etc.), so as to improve the accuracy of category identification through subject filtering strategies. In addition, this application embodiment can involve many technical issues such as detection, subject identification, category consistency, and image-text consistency, so as to design image subject and text category consistency, as well as visual subject filtering branches. By complementary fusion of the two branches, the association recall can be maximized while ensuring accuracy. That is, the recall rate of subject classification categories can be improved by combining various filtering strategies and multimodal information such as various branches, thereby achieving accurate prediction of the target subject (i.e., the intention subject of the target object) and the target subject classification category. Furthermore, embodiments of this application can be designed with reasonable deduplication and merging logic, truncation logic, etc., to ensure high-accuracy output of the target subject, and then achieve comprehensive association between the two category systems through text fuzzy matching.

[0105] Based on the description of the relevant embodiments of the above category identification method, this application also proposes a category identification device, which can be a computer program (including program code) running on a computer device. The category identification device can execute... Figure 2 or Figure 4 The category identification method shown; please refer to [link / reference]. Figure 6 The category identification device can operate the following units:

[0106] The acquisition unit 601 is used to acquire the target data when a preset behavior data of a target object is detected in relation to the target data; wherein the target data includes at least one image;

[0107] The processing unit 602 is used to perform subject recognition on each image in the at least one image to obtain one or more subjects, and to perform classification processing on each of the one or more subjects to obtain the classification category of each subject;

[0108] The processing unit 602 is further configured to perform filtering processing on the one or more subjects according to one or more subject filtering strategies to obtain filtered subjects, wherein the filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than the matching degree between the filtered subjects and the target data.

[0109] The processing unit 602 is further configured to determine, based on the classification categories of each candidate subject in the filtered subjects, a target subject and the integrated classification category corresponding to the target subject from the filtered subjects, and use the integrated classification category as the category identification result of the target data, wherein the matching degree between the target subject and the target data is greater than the matching degree between the subjects other than the target subject in the filtered subjects and the target data.

[0110] The output unit 603 is used to output the category identification result of the target data.

[0111] In one implementation, the one or more subject filtering strategies include at least a category inconsistency filtering strategy; when the processing unit 602 performs filtering processing on the one or more subjects according to the one or more subject filtering strategies to obtain filtered subjects, it may specifically be used to:

[0112] Obtain the detection categories for each of the aforementioned entities;

[0113] Based on the category inconsistency filtering strategy, inconsistent subjects are identified from the one or more subjects, and the detection category and classification category of the inconsistent subjects are inconsistent.

[0114] The inconsistent subjects are filtered out from the one or more subjects to obtain the filtered subjects.

[0115] In another implementation, the one or more strategies include at least a mutually exclusive category filtering strategy, and the target data includes a target category, which refers to a category used to reflect the topic information of the target data; when the processing unit 602 performs filtering processing on the one or more subjects according to one or more subject filtering strategies to obtain filtered subjects, it may specifically be used to:

[0116] Based on the mutually exclusive category filtering strategy, category mapping information is obtained, which is used to indicate the relevance and mutual exclusion between various categories in multiple categories;

[0117] Based on the category mapping information, relevant entities are found from the one or more entities. The correlation between the detection category of the relevant entity and the target category is greater than the correlation between the detection category of the entities other than the relevant entity and the target category.

[0118] Find mutually exclusive entities from the one or more entities whose detection categories are mutually exclusive with the detection categories of the related entities; and filter out the mutually exclusive entities from the one or more entities to obtain the filtered entities.

[0119] In another embodiment, the one or more subject filtering strategies include at least a specified category filtering strategy, and a category includes multi-level categories; when the processing unit 602 filters the one or more subjects according to the one or more subject filtering strategies to obtain filtered subjects, it can be specifically used for:

[0120] Based on the specified category filtering strategy, the preset specified category setting information is obtained. The specified category setting information includes: at least one specified category, and the restricted categories under each specified category.

[0121] The specified filtering subject is found among the one or more subjects, and the classification category of the specified filtering subject includes: the target specified category in the specified category setting information and any restricted category under the target specified category;

[0122] The specified filter subject is filtered out from the one or more subjects to obtain the filtered subject.

[0123] In another embodiment, the one or more subject filtering strategies include at least a category consistency filtering strategy, and the target data further includes text category terms; when the processing unit 602 filters the one or more subjects according to the one or more subject filtering strategies to obtain the filtered subjects, it can be specifically used for:

[0124] Based on the category consistency filtering strategy, category matching entities that match the category category in the text are identified from the one or more entities, and any first-level category in the category matching entity is the same as any category word in the text.

[0125] The category matching subject is used as the filtered subject.

[0126] In another embodiment, the one or more subject filtering strategies include at least a first subject filtering strategy and a second subject filtering strategy. The first subject filtering strategy includes at least one of the following: a category inconsistency filtering strategy, a mutually exclusive category filtering strategy, and a specified category filtering strategy. The second subject filtering strategy includes at least a category consistency filtering strategy. When the processing unit 602 performs filtering processing on the one or more subjects according to the one or more subject filtering strategies to obtain filtered subjects, it can be specifically used for:

[0127] Based on the first subject filtering strategy, the one or more subjects are filtered to obtain the first filtered subject;

[0128] Based on the second subject filtering strategy, the one or more subjects are filtered to obtain the second filtered subject;

[0129] The first filter body and the second filter body are used as the filtered body.

[0130] In another implementation, the classification categories of any subject include M levels of categories, where M is an integer greater than 1; when the processing unit 602 determines the target subject and the integrated classification category corresponding to the target subject from the filtered subjects based on the classification categories of each candidate subject, it can be specifically used for:

[0131] For the m-th level category in the classification category, the m-th level category in the classification category of each candidate subject is deduplicated to obtain the deduplicated m-th level category, where m∈[1,H], and H is an integer greater than or equal to 1 and less than M;

[0132] The display parameters of each candidate subject are calculated respectively, and based on the display parameters of each candidate subject, candidate subjects whose display parameters are greater than the parameter threshold are determined from the filtered subjects to obtain the target subject;

[0133] The first H-level categories after deduplication of the target subject are taken as the first H-level categories of the integrated classification category, and the i-th level category of the target subject is taken as the i-th level category of the integrated classification category, i∈[H+1,M].

[0134] In another embodiment, when the processing unit 602 determines candidate subjects whose display parameters are greater than a parameter threshold from the filtered subjects based on the display parameters of each candidate subject, in order to obtain the target subject, it may be specifically used for:

[0135] Based on the display parameters of each candidate subject, candidate subjects whose display parameters are greater than the parameter threshold are determined from the filtered subjects to obtain a candidate subject set;

[0136] The M-th level category of the classification category of the candidate subjects in the candidate subject set is deduplicated to select leaf category deduplicated subjects from the candidate subject set, and the target subject is selected from the leaf category deduplicated subjects;

[0137] The number of deduplicated leaf category subjects is equal to the number of deduplicated M-level categories, and a deduplicated leaf category subject is determined by the ranking position of the candidate subjects with the same M-level category in the target ranking result.

[0138] In another embodiment, the target data includes target categories, which refer to categories used to reflect the subject information of the target data; the acquisition unit 601 can also be used for:

[0139] Obtain data filtering instruction information, which includes at least one of the following: value assessment data and filtering categories, wherein the value assessment data includes valuable entities;

[0140] Processing unit 602 can also be used for:

[0141] Based on the data filtering instruction information, the target data is subjected to data filtering processing in order to identify the category of the target data that has passed the data filtering processing. The target data that has passed the data filtering processing refers to: target data that includes valuable subjects and / or target data that includes target categories other than the filtered categories.

[0142] According to one embodiment of this application, Figure 2 or Figure 4 Each step involved in the method shown can be derived from... Figure 6 This is performed by each unit in the category identification device shown. For example, Figure 2 Step S201 shown can be performed by Figure 6 The acquisition unit 601 shown is executed, and steps S202-S204 can all be performed by... Figure 6 The processing unit 602 shown executes the step S205, which can be performed by... Figure 6 The output unit 603 shown is executed. For example, Figure 4 Steps S401 and S402 shown can both be performed by Figure 6 The acquisition unit 601 shown is executed, and steps S403-S406 can all be performed by... Figure 6 The processing unit 602 shown executes the step S407, which can be performed by... Figure 6 The output unit 603 shown executes, etc.

[0143] According to another embodiment of this application, Figure 6The categories identification device shown can be composed of individual or combined units into one or more other units, or some of the units can be further divided into multiple functionally smaller units. This achieves the same operation without affecting the technical effects of the embodiments of this application. The above units are based on logical function division. In practical applications, the function of one unit can be implemented by multiple units, or the function of multiple units can be implemented by one unit. In other embodiments of this application, the categories identification device may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0144] According to another embodiment of this application, the following can be achieved by running on a general-purpose computing device, such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM), a device capable of performing operations such as... Figure 2 or Figure 4 The computer program (including program code) for each step involved in the corresponding method shown, to construct such... Figure 6 The category identification device shown herein, and the category identification method for implementing the embodiments of this application, are described. The computer program may be recorded on, for example, a computer storage medium, loaded onto the aforementioned computing device via the computer storage medium, and run therein.

[0145] This application embodiment can acquire target data, which includes at least one image, when it detects preset behavior data of a target object targeting target data. Based on this, subject recognition can be performed on each image in the at least one image to obtain one or more subjects, and each subject in the one or more subjects can be classified to obtain the classification category of each subject. Then, one or more subjects can be filtered according to one or more subject filtering strategies to obtain filtered subjects, which include at least one candidate subject. Based on the classification category of each candidate subject in the filtered subjects, a target subject that matches the target data better and the integrated classification category corresponding to the target subject can be determined from the filtered subjects. The integrated classification category is used as the category recognition result of the target data, and the category recognition result of the target data is output, which can effectively improve the accuracy of the category recognition result.

[0146] Based on the description of the above method and apparatus embodiments, this application also provides a computer device. Please refer to... Figure 7 The computer device includes at least a processor 701, an input interface 702, an output interface 703, and a computer storage medium 704. The processor 701, input interface 702, output interface 703, and computer storage medium 704 within the computer device can be connected via a bus or other means.

[0147] Computer storage medium 704 can be stored in the memory of a computer device. The computer storage medium 704 is used to store computer programs, which include program instructions. The processor 701 is used to execute the program instructions stored in the computer storage medium 704. The processor 701 (or CPU (Central Processing Unit)) is the computing and control core of the computer device, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. In one embodiment, the processor 701 described in this application embodiment can be used to perform a series of category recognitions, specifically including: when preset behavior data of a target object against target data is detected, acquiring the target data; wherein, the target data includes at least one image; performing subject recognition on each image in the at least one image to obtain one or more subjects, and classifying each subject in the one or more subjects to obtain the classification category of each subject; according to one or more... A subject filtering strategy is employed to filter one or more subjects, resulting in filtered subjects. The filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than that between the filtered subjects and the target data. Based on the classification categories of each candidate subject in the filtered subjects, a target subject and its corresponding integrated classification category are determined from the filtered subjects. This integrated classification category is used as the category identification result for the target data. The matching degree between the target subject and the target data is greater than that between the other subjects in the filtered subjects (excluding the target subject) and the target data. The category identification result for the target data is then output, and so on.

[0148] This application embodiment also provides a computer storage medium (memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer storage medium provides storage space that stores the operating system of the computer device. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device; optionally, it can also be at least one computer storage medium located remotely from the aforementioned processor. In one embodiment, the processor can load and execute one or more instructions stored in the computer storage medium to implement the above-mentioned... Figure 2 or Figure 4 The various method steps in the embodiments of the category identification method shown.

[0149] This application embodiment can acquire target data, which includes at least one image, when it detects preset behavior data of a target object targeting target data. Based on this, subject recognition can be performed on each image in the at least one image to obtain one or more subjects, and each subject in the one or more subjects can be classified to obtain the classification category of each subject. Then, one or more subjects can be filtered according to one or more subject filtering strategies to obtain filtered subjects, which include at least one candidate subject. Based on the classification category of each candidate subject in the filtered subjects, a target subject that matches the target data better and the integrated classification category corresponding to the target subject can be determined from the filtered subjects. The integrated classification category is used as the category recognition result of the target data, and the category recognition result of the target data is output, which can effectively improve the accuracy of the category recognition result.

[0150] It should be noted that, according to one aspect of this application, a computer program product or computer program is also provided, which includes computer instructions stored in a computer storage medium. The processor of a computer device reads the computer instructions from the computer storage medium, executes the computer instructions, and causes the computer device to perform the aforementioned actions. Figure 2 or Figure 4 The methods provided in various alternative ways in the examples of the category identification method shown.

[0151] Furthermore, it should be understood that the above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application are still within the scope of this application.

Claims

1. A category identification method, characterized in that, include: When preset behavior data of a target object in response to target data is detected, the target data is acquired; wherein, the target data includes at least one image; Subject recognition is performed on each image in the at least one image to obtain one or more subjects, and each of the one or more subjects is classified to obtain the classification category of each subject; According to one or more subject filtering strategies, the one or more subjects are filtered to obtain filtered subjects. The filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than the matching degree between the filtered subjects and the target data. Based on the classification categories of each candidate subject in the filtered subjects, the target subject and the integrated classification category corresponding to the target subject are determined from the filtered subjects, and the integrated classification category is used as the category identification result of the target data. The matching degree between the target subject and the target data is greater than the matching degree between the subjects other than the target subject in the filtered subjects and the target data. Output the category identification results of the target data.

2. The method according to claim 1, characterized in that, The one or more subject filtering strategies include at least a category inconsistency filtering strategy; The step of filtering the one or more subjects according to one or more subject filtering strategies to obtain filtered subjects includes: Obtain the detection categories for each of the aforementioned entities; Based on the category inconsistency filtering strategy, inconsistent subjects are identified from the one or more subjects, and the detection category and classification category of the inconsistent subjects are inconsistent. The inconsistent subjects are filtered out from the one or more subjects to obtain the filtered subjects.

3. The method according to claim 1, characterized in that, The one or more strategies include at least a mutually exclusive category filtering strategy, and the target data includes a target category, which refers to a category used to reflect the topic information of the target data; The step of filtering the one or more subjects according to one or more subject filtering strategies to obtain filtered subjects includes: Based on the mutually exclusive category filtering strategy, category mapping information is obtained, which is used to indicate the relevance and mutual exclusion between various categories in multiple categories; Based on the category mapping information, relevant entities are found from the one or more entities. The correlation between the detection category of the relevant entity and the target category is greater than the correlation between the detection category of the entities other than the relevant entity and the target category. Find mutually exclusive entities from the one or more entities whose detection categories are mutually exclusive with the detection categories of the related entities; and filter out the mutually exclusive entities from the one or more entities to obtain the filtered entities.

4. The method according to claim 1, characterized in that, The one or more subject filtering strategies include at least a specified category filtering strategy, and a category includes multiple levels of categories; The step of filtering the one or more subjects according to one or more subject filtering strategies to obtain filtered subjects includes: Based on the specified category filtering strategy, the preset specified category setting information is obtained. The specified category setting information includes: at least one specified category, and the restricted categories under each specified category. The specified filtering subject is found among the one or more subjects, and the classification category of the specified filtering subject includes: the target specified category in the specified category setting information and any restricted category under the target specified category; The specified filter subject is filtered out from the one or more subjects to obtain the filtered subject.

5. The method according to claim 1, characterized in that, The one or more subject filtering strategies include at least a category consistency filtering strategy, and the target data further includes text category terms; the step of filtering the one or more subjects according to the one or more subject filtering strategies to obtain filtered subjects includes: Based on the category consistency filtering strategy, category matching entities that match the category category in the text are identified from the one or more entities, and any first-level category in the category matching entity is the same as any category word in the text. The category matching subject is used as the filtered subject.

6. The method according to claim 1, characterized in that, The one or more subject filtering strategies include at least a first subject filtering strategy and a second subject filtering strategy. The first subject filtering strategy includes at least one of the following: category inconsistency filtering strategy, mutually exclusive category filtering strategy and specified category filtering strategy. The second subject filtering strategy includes at least a category consistency filtering strategy. The step of filtering the one or more subjects according to one or more subject filtering strategies to obtain filtered subjects includes: Based on the first subject filtering strategy, the one or more subjects are filtered to obtain the first filtered subject; Based on the second subject filtering strategy, the one or more subjects are filtered to obtain the second filtered subject; The first filter body and the second filter body are used as the filtered body.

7. The method according to any one of claims 1-6, characterized in that, The classification categories of any subject include M levels of categories, where M is an integer greater than 1; the step of determining the target subject and its corresponding integrated classification category from the filtered subjects based on the classification categories of each candidate subject includes: For the m-th level category in the classification category, the m-th level category in the classification category of each candidate subject is deduplicated to obtain the deduplicated m-th level category, where m∈[1,H], and H is an integer greater than or equal to 1 and less than M; The display parameters of each candidate subject are calculated respectively, and based on the display parameters of each candidate subject, candidate subjects whose display parameters are greater than the parameter threshold are determined from the filtered subjects to obtain the target subject; The first H-level categories after deduplication of the target subject are taken as the first H-level categories of the integrated classification category, and the i-th level category of the target subject is taken as the i-th level category of the integrated classification category, i∈[H+1,M].

8. The method according to claim 7, characterized in that, The step of determining candidate subjects whose display parameters are greater than a parameter threshold from the filtered subjects based on the display parameters of each candidate subject, in order to obtain the target subject, includes: Based on the display parameters of each candidate subject, candidate subjects whose display parameters are greater than the parameter threshold are determined from the filtered subjects to obtain a candidate subject set; The M-th level category of the classification category of the candidate subjects in the candidate subject set is deduplicated to select leaf category deduplicated subjects from the candidate subject set, and the target subject is selected from the leaf category deduplicated subjects; The number of deduplicated subjects in the leaf category is equal to the number of deduplicated M-level categories, and the deduplicated subject of a leaf category is determined by the ranking position of the candidate subjects with the same M-level category in the target ranking result.

9. The method according to any one of claims 1-6, characterized in that, The target data includes target categories, whereby a target category refers to a category used to reflect the thematic information of the target data; the method further includes: Obtain data filtering instruction information, which includes at least one of the following: value assessment data and filtering categories, wherein the value assessment data includes valuable entities; Based on the data filtering instruction information, the target data is subjected to data filtering processing in order to identify the category of the target data that has passed the data filtering processing. The target data that has passed the data filtering processing refers to: target data that includes valuable subjects and / or target data that includes target categories other than the filtered categories.

10. A category identification device, characterized in that, include: An acquisition unit is configured to acquire the target data when preset behavioral data of a target object in relation to the target data is detected; wherein the target data includes at least one image; The processing unit is configured to perform subject recognition on each image in the at least one image to obtain one or more subjects, and to perform classification processing on each of the one or more subjects to obtain the classification category of each subject; The processing unit is further configured to perform filtering processing on the one or more subjects according to one or more subject filtering strategies to obtain filtered subjects, wherein the filtered subjects include at least one candidate subject, and the matching degree between the filtered subjects and the target data is greater than the matching degree between the filtered subjects and the target data. The processing unit is further configured to determine the target subject and the integrated classification category corresponding to the target subject from the filtered subjects based on the classification categories of each candidate subject in the filtered subjects, and use the integrated classification category as the category identification result of the target data, wherein the matching degree between the target subject and the target data is greater than the matching degree between the subjects other than the target subject in the filtered subjects and the target data; The output unit is used to output the category identification result of the target data.

Images