An industry identification method, device, storage medium and electronic equipment

Abstract

The specification discloses an industry identification method, device, storage medium and electronic equipment. In the industry identification method provided in the specification, user information of a user is acquired, and corresponding fusion features are extracted; the fusion features are input as input into feature extraction layers of each label level in an identification model, to determine local dimension features of the user under different label levels; meanwhile, according to the upper and lower layer relationship of the label level corresponding to each feature extraction layer, intermediate features extracted by a previous layer feature extraction layer and the fusion features are input as input into a next layer feature extraction layer, to determine global dimension features of the user through each feature extraction layer; the level features of the user are determined according to each local dimension feature and the global dimension feature; and the level features are input into an output subnetwork of the identification model, to output labels of the industry of the user in each label dimension.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a method, apparatus, storage medium, and electronic device for industry identification. Background Technology

[0002] With the development of the internet, business platforms have gradually launched a wide variety of services. However, in reality, many users do not have time to choose from the numerous services available in their daily lives; at the same time, too many services may make it difficult for users with service needs to find suitable ones. Therefore, business platforms need to reasonably recommend services to users to help them choose the right ones.

[0003] Typically, business platforms recommend suitable services to users based on their personal information, with recommendations based on the user's industry being particularly effective. However, currently, due to the protection of user privacy data, the user information available to business platforms is often very limited, and in many cases, they cannot directly determine the user's industry.

[0004] Therefore, determining the industry to which a user belongs is a problem that urgently needs to be solved. Summary of the Invention

[0005] This specification provides an industry identification method and an industry identification device to partially solve the aforementioned problems existing in the prior art.

[0006] The following technical solution is adopted in this specification:

[0007] This specification provides a method for industry identification, including:

[0008] Obtain user information;

[0009] The user information is input into a pre-trained recognition model, and the user's fusion features are determined through the feature processing subnetwork in the model.

[0010] The fused features are used as input and input to the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels.

[0011] According to the hierarchical relationship of the label levels corresponding to each feature extraction layer, the intermediate features extracted by the previous feature extraction layer and the fused features are used as the input of the next feature extraction layer. Through each feature extraction layer, the global dimensional features of the user are determined.

[0012] Based on the local dimensional features and the global dimensional features, the user's hierarchical features are determined, and the hierarchical features are used to characterize the user's industry under each tag level.

[0013] The hierarchical features are input into the output subnet of the recognition model, and the user's industry labels at each label level are output through the output subnet.

[0014] Optionally, user information may be obtained, specifically including:

[0015] Based on the user data obtained from multiple data sources, the user information of the user is determined, including location information, behavioral information, and professional license information.

[0016] Optionally, the feature processing subnetwork includes at least: an extraction layer, a splicing layer, and a fusion layer;

[0017] The user's fusion features are determined through the feature processing subnet in the model, specifically including:

[0018] The location information, behavior information, and professional license information are input into the extraction layer, and the user's location features, behavior features, and professional license features are extracted through the extraction layer.

[0019] The location features and the behavior features are input into the splicing layer, and the splicing features are determined by the splicing layer;

[0020] The industry license feature and the splicing feature are input into the fusion layer, and the fusion feature is determined by the fusion layer.

[0021] Optionally, the extraction layer includes at least: a first extraction layer and a second extraction layer;

[0022] The location information, behavior information, and professional license information are input into the extraction layer, and the extraction layer extracts the user's location features, behavior features, and professional license features, specifically including:

[0023] The professional license information is input into the first extraction layer, and the professional license features of the user are extracted through the first extraction layer.

[0024] The location information and behavior information are input into the second extraction layer, and the user's location features and behavior features are extracted through the second extraction layer.

[0025] Optionally, the feature extraction layer includes a global output layer and a local output layer;

[0026] The fused features are used as input and fed into the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user at different label levels, specifically including:

[0027] For each feature extraction layer, the intermediate features output by the global output layer of the previous feature extraction layer are determined according to the hierarchical relationship of the label levels corresponding to each feature extraction layer.

[0028] The intermediate features are concatenated with the fused features, and the concatenation result is used as input to the feature extraction layer.

[0029] The local dimensional features of the feature extraction layer are determined by sequentially passing through the global output layer and the local output layer of the feature extraction layer, and are used as the local dimensional features of the label level corresponding to the feature extraction layer.

[0030] Within each tag level, the higher the tag level, the finer the granularity of the tags within that level.

[0031] Optionally, according to the hierarchical relationship of the label levels corresponding to each feature extraction layer, the intermediate features extracted by the previous feature extraction layer and the fused features are used as the input of the next feature extraction layer. Through each feature extraction layer, the global dimensional features of the user are determined, specifically including:

[0032] For each feature extraction layer, the intermediate features output by the global output layer of the previous feature extraction layer are determined in order of the label level corresponding to each feature extraction layer from low to high.

[0033] The intermediate features are concatenated with the fused features, and the concatenation result is used as input to the feature extraction layer.

[0034] The global dimensional features of the user are determined based on the intermediate features output by the global output layer of the feature extraction layer at the highest label level.

[0035] Optionally, the recognition model further includes a matching subnet, in which the label features of each label are pre-stored;

[0036] Before inputting the hierarchical features into the output subnet of the recognition model, the method further includes:

[0037] The hierarchical features are input into the matching subnet, and among each label feature, the label features that match the dimensional features contained in the hierarchical features are determined.

[0038] Based on the matching results and the matched label features, at least some of the dimensional features in the hierarchical features are adjusted, and the hierarchical features are redefined.

[0039] Optionally, the tag features of each tag are extracted based on the tag features of each tag under the highest tag level;

[0040] Based on the matching results and the matched label features, at least some dimensions of the hierarchical features are adjusted to redetermine the hierarchical features, specifically including:

[0041] By using label features that match the dimensional features contained in the hierarchical features, the dimensional features of the highest label level in the hierarchical features are replaced to obtain the redefined hierarchical features.

[0042] Optionally, the user's industry tags at each tag level are output through the output subnet, specifically including:

[0043] Based on the dimensional features of each tag level in the hierarchical features, the tags of the user's industry under each tag level are output in ascending order of tag level.

[0044] Optionally, a recognition model is pre-trained, specifically including:

[0045] Obtain sample user information and labeling of sample users;

[0046] The sample user information is input into a pre-trained recognition model, and the fusion features of the sample user are determined through the feature processing subnetwork in the model.

[0047] The fused features are used as input and input to the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels.

[0048] According to the hierarchical relationship of the label levels corresponding to each feature extraction layer, the intermediate features extracted by the previous feature extraction layer and the fused features are used as the input of the next feature extraction layer. Through each feature extraction layer, the global dimensional features of the user are determined.

[0049] Based on the local dimensional features and the global dimensional features, the user's hierarchical features are determined, and the hierarchical features are used to characterize the user's industry under each tag level.

[0050] The hierarchical features are input into the output subnet of the recognition model, and the industry of the sample user is output through the output subnet at each label level.

[0051] The recognition model is trained with the goal of minimizing the difference between the label output by the recognition model and the labeled label.

[0052] This specification provides an industry identification device, including:

[0053] The acquisition module is used to acquire user information.

[0054] The processing module is used to input the user information into a pre-trained recognition model and determine the user's fusion features through the feature processing subnet in the model;

[0055] The local module is used to take the fused features as input and input them into the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels.

[0056] The global module is used to determine the global dimensional features of the user by taking the intermediate features extracted by the previous feature extraction layer and the fused features as inputs to the next feature extraction layer according to the hierarchical relationship of the label levels corresponding to each feature extraction layer.

[0057] The hierarchical module is used to determine the user's hierarchical features based on the local dimensional features and the global dimensional features. The hierarchical features are used to characterize the user's industry under each tag level.

[0058] The output module is used to input the hierarchical features into the output subnet of the recognition model, and output the user's industry labels at each label level through the output subnet.

[0059] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned industry identification method.

[0060] This specification provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned industry identification method.

[0061] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects:

[0062] The industry identification method provided in this specification acquires user information and extracts corresponding fusion features. These fusion features are then input into the feature extraction layer of each label level in the identification model to determine the local dimensional features of the user at different label levels. Simultaneously, according to the hierarchical relationship of each feature extraction layer's label level, the intermediate features extracted by the previous feature extraction layer and the fusion features are used as input to the next feature extraction layer. Through each feature extraction layer, the user's global dimensional features are determined. Based on the local and global dimensional features, the user's hierarchical features are determined. These hierarchical features are then input into the output subnet of the identification model to output the user's industry label at each label dimension. Attached Figure Description

[0063] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings:

[0064] Figure 1 This is a flowchart illustrating one industry identification method provided in this specification;

[0065] Figure 2 This is a schematic diagram of the structure of a recognition model provided in this specification;

[0066] Figure 3 This is a schematic diagram of the structure of a multidimensional subnet in an identification model provided in this specification;

[0067] Figure 4 This specification provides a schematic diagram of an industry identification device.

[0068] Figure 5 This specification provides a corresponding Figure 1 A schematic diagram of an electronic device. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0070] The technical solutions provided in the various embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0071] Figure 1 This is a flowchart illustrating an industry identification method provided in this specification, including the following steps:

[0072] S100: Obtain user information.

[0073] In this specification, the executing entity used to implement the industry identification method can refer to a designated device such as a server set up on the business platform. For ease of description, this specification will only use the server as the executing entity as an example to illustrate one industry identification method provided in this specification.

[0074] In the industry identification method provided in this specification, user information may include at least the user's location information, behavioral information, and industry license information. The user's location information may refer to the latitude and longitude of the user's permanent residence. When a user has multiple permanent residences, the latitude and longitude of one of these residences may be selected as the user's location information. The user's behavioral information may refer to any behavior the user has performed on the business platform, such as performing various business transactions or making purchases. The user's industry license information may refer to information about the business or behaviors the user is capable of performing.

[0075] In this application, user information can be determined based on user data obtained from multiple data sources. This user information includes location information, behavioral information, and professional license information. For example, user location information can be obtained directly through maps and positioning systems; user behavioral information can be obtained based on various behaviors the user has performed on the business platform in historical records; and user professional license information can be obtained through business licenses, skill certificates, etc., submitted by the user on the business platform. It is worth noting that in practical applications, all user behavioral information and all professional license information can be obtained, or, depending on the application scenario, only a portion of key behavioral information and professional license information can be obtained to reduce the amount of data to be processed and shorten the model's recognition time. When obtaining user professional license information, the user's business license and / or skill certificate can be scanned to obtain the text, which is then presented as the user's professional license information in text form.

[0076] S102: Input the user information into a pre-trained recognition model, and determine the user's fusion features through the feature processing subnet in the model.

[0077] The industry identification method provided in this specification uses an identification model to identify the user's industry. The model structure of the identification model can be as follows: Figure 2 As shown, the recognition model may include a feature processing subnet, a multidimensional subnet, and an output subnet.

[0078] After obtaining the user's information in step S100, the user information can be input into the recognition model, and the user's fused features can be extracted through the feature processing subnet in the recognition model.

[0079] Furthermore, in this application, user information may include at least the user's location information, behavior information, and professional license information. Different features can be extracted from each of these three types of information, and then these three different features can be fused into a fused feature. Specifically, the feature processing subnet includes at least an extraction layer, a stitching layer, and a fusion layer. The location information, behavior information, and professional license information can be input into the extraction layer, which extracts the user's location features, behavior features, and professional license features. The location features and behavior features are input into the stitching layer, which determines the stitched feature. The professional license feature and the stitched feature are input into the fusion layer, which determines the fused feature.

[0080] The extraction layer can be implemented by setting up a model in the network layer to extract parameters. Typically, location and behavioral information are various types of data. Separate models can be trained specifically for extracting location and behavioral features. Furthermore, the models for extracting location and behavioral features can have the same structure but different parameters. Professional license information is usually text information scanned from business licenses or skill certificates; therefore, existing text feature extraction models, such as BERT, can be directly used to extract professional license features. Specifically, the extraction layer includes at least: a first extraction layer and a second extraction layer; professional license information can be input into the first extraction layer to extract the user's professional license features; the location and behavioral information can be input into the second extraction layer to extract the user's location and behavioral features.

[0081] Typically, models with the same structure but different parameters can be used to extract location and behavioral features. However, extracting professional license features requires a different model structure. Therefore, in most cases, the extracted location and behavioral features reside in the same feature space, while the professional license feature often resides in a different feature space. To better integrate location, behavioral, and professional license features into a single fused feature, after concatenating the location and behavioral features, an additional network layer can be added to change the feature space of the feature—that is, to change the feature's dimension. This alters the dimensionality of the concatenated feature, either increasing or decreasing it to the same dimension as the professional license feature, allowing for better integration. This network layer for changing the feature's dimension can be based on a self-attention mechanism.

[0082] Furthermore, the fusion layer can also employ a network layer based on attention mechanisms, including but not limited to soft-attention, hard-attention, and cross-attention mechanisms. The attention mechanism fuses the concatenated features and the industry-licensed features into a fused feature.

[0083] S104: Using the fused features as input, input them into the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels.

[0084] In this step, the fused features obtained in step S104 can be input into the feature extraction layers of each label level in the recognition model to determine the local dimensional features of the user at different label levels. For example, Figure 2 As shown, the feature extraction layers of each label level in the recognition model are located in the multidimensional subnet of the recognition model.

[0085] In the industry identification method provided in this specification, the identification model outputs the user's industry in the form of labels, and multiple labels are output simultaneously. Each label has a different label level, and these multiple labels describe the user's industry from multiple different label levels. It should be noted that the label level here can be seen as the granularity of the label, representing the specificity of a label in describing an industry, or the scope of the industry it covers. Specifically, for each label level, the higher the label level, the finer the granularity of the labels within that level. Higher-level labels are more specific and granular in describing an industry. For example, suppose a user's specific industry is corn cultivation. However, in reality, describing this user's industry could be more vaguely described as staple food cultivation, or even more broadly as the broader field of agriculture. In the example above, the user's industry could be labeled as agriculture at the first label level, staple food cultivation at the second label level, and corn cultivation at the third label level. The higher the label level, the smaller the scope of the label's description, and the higher the degree of granularity. It is worth mentioning that the number of feature extraction layers at the label level in the recognition model is not fixed and can be set according to the specific needs of the label level.

[0086] In the industry identification method provided in this manual, the final output of the identification model is the user's industry label at each label level. For example... Figure 3As shown, in the recognition model used in this application, each label level's feature extraction layer has two output layers: a global output layer and a local output layer. In this step, the fused features are separately input into each label level's feature extraction layer. The local output layer of each label level's feature extraction layer outputs the local dimensional features of that label level. Specifically, the local dimensional features output by the feature extraction layer of a label level are used to represent the label at that label level.

[0087] Specifically, the fused features can be used as input and fed into the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user corresponding to different label levels. This includes: for each feature extraction layer, determining the intermediate features output by the global output layer of the previous feature extraction layer according to the hierarchical relationship of the label levels; concatenating the intermediate features with the fused features, and inputting the concatenation result into the feature extraction layer; sequentially passing through the global output layer and local output layer of the feature extraction layer to determine the local dimensional features output by the feature extraction layer, which are then used as the local dimensional features of the label level corresponding to the feature extraction layer; wherein, for each label level, the higher the label level, the finer the granularity of the labels at that label level.

[0088] S106: According to the hierarchical relationship of the label levels corresponding to each feature extraction layer, the intermediate features extracted by the previous feature extraction layer and the fused features are used as the input of the next feature extraction layer. Through each feature extraction layer, the global dimensional features of the user are determined.

[0089] It should be noted that there is no strict order between this step and step S104. In the actual execution of this method, steps S104 and S106 can be performed simultaneously.

[0090] While determining the local dimensional features at each label level in step S104, step S106 determines the user's global dimensional features. The intermediate features output by the global output layer of the previous feature extraction layer are used as the input to the next feature extraction layer, until the global output layer of the feature extraction layer at the highest label level outputs the user's global dimensional features.

[0091] Furthermore, it is possible to... Figure 3As shown, the fused features and the intermediate features extracted by the previous feature extraction layer are used as input to the next feature extraction layer. The global dimensional features of the user are determined through each feature extraction layer. Specifically, for each feature extraction layer, the intermediate features output by the global output layer of the previous feature extraction layer are determined according to the label level of each feature extraction layer from low to high. These intermediate features are then concatenated with the fused features, and the concatenation result is used as input to the feature extraction layer. Based on the intermediate features output by the global output layer of the feature extraction layer with the highest label level, the global dimensional features of the user are determined.

[0092] When predicting a user's industry, the broader the prediction scope and the coarser the required granularity, the easier it is to obtain accurate results. For example, based on the same information, identifying a user's industry as agriculture is certainly simpler than identifying them specifically as a corn grower. In this application, based on the same fusion features, identifying lower-level labels is often more accurate than identifying higher-level labels. Therefore, one can start with lower-level labels and gradually determine higher-level features to obtain the most accurate results.

[0093] Suppose there are three feature extraction layers at three label levels in the multidimensional subnet: the first feature extraction layer at the first label level, the second feature extraction layer at the second label level, and the third feature extraction layer at the third label level. When the fused features enter the multidimensional subnet, they first enter the first feature extraction layer, which has a lower dimension, and the global output layer outputs the intermediate features of the first label level. Subsequently, the fused features, along with the intermediate features of the first label level, enter the second feature extraction layer, which outputs the intermediate features of the second label level. Then, the fused features, along with the intermediate features of the second label level, enter the third feature extraction layer. At this point, the global output layer of the third feature extraction layer outputs the user's global dimensional features.

[0094] The global dimensional features represent the user's tags at each tag level. In fact, the global dimensional features can be viewed as a concatenation of dimensional features from multiple different tag levels, with each dimensional feature representing a tag at a specific tag level.

[0095] S108: Determine the user's hierarchical features based on the local dimensional features and the global dimensional features. The hierarchical features are used to characterize the user's industry under each tag level.

[0096] Each local dimensional feature represents the characteristics of a label at a single label level, while the global dimensional features contain multiple dimensional features, representing labels at multiple label levels. Typically, the number of local dimensional features is the same as the number of feature extraction layers at each label level; similarly, the number of global dimensional features is also the same as the number of feature extraction layers at each label level. The local dimensional features obtained from each layer can be fused with the global dimensional features to obtain hierarchical features.

[0097] During fusion, it can be as follows Figure 3 As shown, the local dimensional features of each label level are first concatenated together, and then the concatenated features are weighted and summed with the global dimensional features, where β represents the weight of the global dimensional features.

[0098] S110: Input the hierarchical features into the output subnet of the recognition model, and output the user's industry labels at each label level through the output subnet.

[0099] Finally, the hierarchical features obtained in step S108 can be input into the output subnet of the recognition model. The output subnet then outputs the user's industry labels at each label level based on the hierarchical features, providing a detailed reflection of the user's industry. It is worth noting that while the recognition model outputs the user's industry labels at each label level, it also outputs the hierarchical relationships between the labels. That is, the output layer can output the user's industry labels at each label level according to the dimensional features of each label level in the hierarchical features, in ascending order of label level.

[0100] For example, in the output layer, if the first label level for a user's industry is "planting," the second label level is "staple food," and the third label level is "corn," then the labels will be output in the format {first label level - second label level - third label level}, which is {planting - staple food - corn}. Furthermore, to simplify the output, a numerical identifier can be set for each label. Only the corresponding numerical identifier is output. The numerical identifiers for labels at different label levels can be repeated or not. Since the labels at each label level are output in a fixed order from low to high label levels, misidentification will not occur as long as the numerical identifiers for labels within the same label level are different. For example, in the first label level, the numerical identifier for "planting" can be set to 1; in the second label level, the numerical identifier for "staple food" can be set to 1; and in the third label level, the numerical identifier for "corn" can be set to 1. In this case, the final output of the recognition model will be {1-1-1}.

[0101] When using the industry identification method provided in this manual, multi-source information can be utilized to extract multi-modal features. By combining these multi-modal features, industry labels at multiple label levels can be predicted for the user's industry, thereby providing a detailed and comprehensive output for the user's industry.

[0102] The industry identification method provided in this specification also includes a matching subnet in its identification model. In the matching subnet, label features for each label are pre-stored. Therefore, when hierarchical features enter the matching subnet, these hierarchical features can be input into the matching subnet. Among the label features, label features that match the dimensional features included in the hierarchical features are determined. Based on the matching results and the matched label features, at least some dimensional features in the hierarchical features are adjusted, and the hierarchical features are redefined. Specifically, the matching degree can be calculated using a network layer based on a soft attention mechanism.

[0103] As one might expect, the higher the tag level, the more detailed the industry segmentation, naturally leading to fewer data points available for training. Therefore, in practical applications, when the model outputs tags at each tag level, higher-level tags tend to have lower accuracy, especially the highest-level tags. For example, among three users, one grows corn, one grows rice, and one grows wheat. When using these three users' data as training samples, the first tag level (planting) has three training samples, the second tag level (staple grains) also has three training samples, while the third tag level (corn, wheat, and rice) each has only one sample. This results in lower training efficiency compared to the lower tag levels.

[0104] Therefore, to address this potential issue, the label features stored in the matching subnet can be extracted based on the labels at the highest label level. After determining the label features that match the hierarchical features, the label features that match the dimensional features contained in the hierarchical features can be used to replace the dimensional features of the highest label level in the hierarchical features, thus obtaining the redefined hierarchical features.

[0105] Additionally, the industry identification method provided in this specification uses an identification model that can be pre-trained. Specifically, sample user information and labeled tags can be obtained; the sample user information is input into a pre-trained identification model, and the fusion features of the sample user are determined through the feature processing subnet of the model; the fusion features are used as input to the feature extraction layer of each label level in the identification model to determine the local dimensional features of the user under different label levels; according to the hierarchical relationship of the label levels corresponding to each feature extraction layer, the intermediate features extracted by the previous feature extraction layer and the fusion features are used as input to the next feature extraction layer, and the global dimensional features of the user are determined through each feature extraction layer; based on the local dimensional features and the global dimensional features, the hierarchical features of the user are determined, and the hierarchical features are used to characterize the features of the user's industry under each label level; the hierarchical features are input into the output subnet of the identification model, and the labels of the sample user's industry under each label level are output through the output subnet; the identification model is trained with the optimization objective of minimizing the difference between the labels output by the identification model and the labeled tags.

[0106] The labels can be manually determined based on the sample data. During training, the parameters of each network layer in each subnet can be adjusted simultaneously.

[0107] It is worth mentioning that the industry identification method provided in this manual can be used for industry identification in different needs and scenarios, depending on the samples used when training the identification model. For example, in the scenario of identifying the rural user industry, rural users can be selected as sample users when training the model, and the model can be trained using the sample information of each rural user to obtain an identification model for identifying the rural user industry. Of course, when training the identification model, all users in various scenarios can also be used as sample users to train the model, resulting in an industry identification model that can identify various industries. It is conceivable that an identification model that only targets one scenario should have higher accuracy than an identification model that can identify multiple scenarios.

[0108] The above describes one or more methods for implementing industry identification in this specification. Based on the same approach, this specification also provides corresponding industry identification devices, such as... Figure 4 As shown.

[0109] Figure 4 A schematic diagram of an industry identification device provided in this specification includes:

[0110] Module 200 is used to obtain user information;

[0111] Processing module 202 is used to input the user information into a pre-trained recognition model and determine the user's fusion features through the feature processing subnet in the model;

[0112] Local module 204 is used to take the fused features as input and input them into the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels.

[0113] The global module 206 is used to determine the global dimensional features of the user by taking the intermediate features extracted by the previous feature extraction layer and the fused features as inputs to the next feature extraction layer according to the hierarchical relationship of the label levels corresponding to each feature extraction layer.

[0114] The hierarchical module 208 is used to determine the hierarchical features of the user based on the local dimensional features and the global dimensional features. The hierarchical features are used to characterize the features of the user's industry under each tag level.

[0115] The output module 210 is used to input the hierarchical features into the output subnet of the recognition model, and output the user's industry labels at each label level through the output subnet.

[0116] Optionally, the acquisition module 200 is specifically used to determine the user information of the user based on the user data of the user obtained from multiple data sources, wherein the user information includes location information, behavior information, and professional license information.

[0117] Optionally, the feature processing subnetwork includes at least: an extraction layer, a splicing layer, and a fusion layer;

[0118] The processing module 202 is specifically used to input the location information, behavior information, and professional license information into the extraction layer, and extract the user's location features, behavior features, and professional license features through the extraction layer; input the location features and behavior features into the stitching layer, and determine the stitching features through the stitching layer; input the professional license features and the stitching features into the fusion layer, and determine the fusion features through the fusion layer.

[0119] Optionally, the extraction layer includes at least: a first extraction layer and a second extraction layer;

[0120] The processing module 202 is specifically used to input the business license information into the first extraction layer and extract the business license features of the user through the first extraction layer; input the location information and behavior information into the second extraction layer and extract the location features and behavior features of the user through the second extraction layer.

[0121] Optionally, the feature extraction layer includes a global output layer and a local output layer;

[0122] The local module 204 is specifically used to take the fused features as input and input them to the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels. Specifically, it includes: for each feature extraction layer, determining the intermediate features output by the global output layer of the previous feature extraction layer according to the upper and lower layer relationships of the label levels corresponding to each feature extraction layer; concatenating the intermediate features with the fused features and using the concatenation result as input to the feature extraction layer; and sequentially passing the global output layer and local output layer of the feature extraction layer to determine the local dimensional features output by the feature extraction layer as the local dimensional features under the label level corresponding to the feature extraction layer; wherein, for each label level, the higher the label level, the finer the granularity of the labels under that label level.

[0123] Optionally, the global module 206 is specifically used to determine, for each feature extraction layer, the intermediate features output by the global output layer of the previous feature extraction layer in ascending order of the label levels corresponding to each feature extraction layer; concatenate the intermediate features with the fused features, and input the concatenation result into the feature extraction layer; and determine the user's global dimensional features based on the intermediate features output by the global output layer of the feature extraction layer at the highest label level.

[0124] Optionally, the recognition model further includes a matching subnet, in which the label features of each label are pre-stored;

[0125] The device further includes a matching module 212, specifically used to input the hierarchical features into the matching subnet, determine the label features that match the dimensional features contained in the hierarchical features among the label features, and adjust at least some of the dimensional features in the hierarchical features according to the matching results and the matched label features, and redetermine the hierarchical features.

[0126] Optionally, the tag features of each tag are extracted based on the tag features of each tag under the highest tag level;

[0127] The matching module 212 is specifically used to replace the label dimension feature with the highest label dimension in the hierarchical features by using the label feature that matches the hierarchical features, so as to obtain the redefined hierarchical features.

[0128] Optionally, the output module 210 is specifically used to output the user's industry tags at each tag level according to the dimensional features of each tag level in the hierarchical features, in order from low to high tag levels.

[0129] Optionally, the device further includes a training module 214, specifically used to acquire sample user information and labeled tags of sample users; input the sample user information into a pre-trained recognition model, and determine the fusion features of the sample user through the feature processing subnet in the model; use the fusion features as input to the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels; according to the hierarchical relationship of the label levels corresponding to each feature extraction layer, use the intermediate features extracted by the previous feature extraction layer and the fusion features as input to the next feature extraction layer, and determine the global dimensional features of the user through each feature extraction layer; determine the hierarchical features of the user based on the local dimensional features and the global dimensional features, the hierarchical features being used to characterize the features of the user's industry under each label level; input the hierarchical features into the output subnet of the recognition model, and output the labels of the sample user's industry under each label level through the output subnet; train the recognition model with the optimization objective of minimizing the difference between the labels output by the recognition model and the labeled tags.

[0130] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 This provides a method for industry identification.

[0131] This instruction manual also provides Figure 5 One of the corresponding Figure 1 A schematic diagram of the structure of an electronic device. (e.g.) Figure 5 At the hardware level, the electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for the business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 The method for industry identification described above. Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. In other words, the execution subject of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.

[0132] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0133] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, ASICs, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0134] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0135] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0136] Those skilled in the art will understand that embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0137] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0138] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0139] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0140] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0141] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0142] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0143] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0144] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0145] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0146] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0147] The above description is merely an embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for industry identification, comprising: Obtain user information, including location information, behavior information, and business license information; The user information is input into a pre-trained recognition model. The feature processing subnetwork in the model is used to determine the user's fused features. The feature processing subnetwork includes at least an extraction layer, a splicing layer, and a fusion layer. The location information, behavior information, and professional license information are input into the extraction layer to extract the user's location features, behavior features, and professional license features. The location features and behavior features are input into the splicing layer to determine the splicing features. The professional license features and the splicing features are input into the fusion layer to determine the fused features. The fused features are used as input and input to the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels. According to the hierarchical relationship of the label levels corresponding to each feature extraction layer, the intermediate features extracted by the previous feature extraction layer and the fused features are used as the input of the next feature extraction layer. Through each feature extraction layer, the global dimensional features of the user are determined. Based on the local dimensional features and the global dimensional features, the user's hierarchical features are determined. The hierarchical features are used to characterize the user's industry under each tag level. The hierarchical features are input into the output subnet of the recognition model, and the user's industry labels at each label level are output through the output subnet.

2. The method as described in claim 1, wherein obtaining user information specifically includes: The user information of the user is determined based on the user data obtained from multiple data sources.

3. The method of claim 1, wherein the extraction layer comprises at least: First extraction layer and second extraction layer; The location information, behavior information, and professional license information are input into the extraction layer, and the extraction layer extracts the user's location features, behavior features, and professional license features, specifically including: The professional license information is input into the first extraction layer, and the professional license features of the user are extracted through the first extraction layer. The location information and behavior information are input into the second extraction layer, and the user's location features and behavior features are extracted through the second extraction layer.

4. The method of claim 1, wherein the feature extraction layer comprises: Global output layer and local output layer; The fused features are used as input and fed into the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user at different label levels, specifically including: For each feature extraction layer, the intermediate features output by the global output layer of the previous feature extraction layer are determined according to the hierarchical relationship of the label levels corresponding to each feature extraction layer. The intermediate features are concatenated with the fused features, and the concatenation result is used as input to the feature extraction layer. The local dimensional features of the feature extraction layer are determined by sequentially passing through the global output layer and the local output layer of the feature extraction layer, and are used as the local dimensional features of the label level corresponding to the feature extraction layer. Within each tag level, the higher the tag level, the finer the granularity of the tags within that level.

5. The method as described in claim 4, wherein, according to the hierarchical relationship of the label levels corresponding to each feature extraction layer, the intermediate features extracted by the previous feature extraction layer and the fused features are used as the input to the next feature extraction layer, and the global dimensional features of the user are determined through each feature extraction layer, specifically including: For each feature extraction layer, the intermediate features output by the global output layer of the previous feature extraction layer are determined in order of the label level corresponding to each feature extraction layer from low to high. The intermediate features are concatenated with the fused features, and the concatenation result is used as input to the feature extraction layer. The global dimensional features of the user are determined based on the intermediate features output by the global output layer of the feature extraction layer at the highest label level.

6. The method of claim 1, wherein the recognition model further comprises: A matching subnet, wherein the tag features of each tag are pre-stored in the matching subnet; Before inputting the hierarchical features into the output subnet of the recognition model, the method further includes: The hierarchical features are input into the matching subnet, and among each label feature, the label features that match the dimensional features contained in the hierarchical features are determined. Based on the matching results and the matched label features, at least some of the dimensional features in the hierarchical features are adjusted, and the hierarchical features are redefined.

7. The method as described in claim 6, wherein the tag features of each tag are extracted based on the tag features of each tag under the highest tag level; Based on the matching results and the matched label features, at least some dimensions of the hierarchical features are adjusted to redetermine the hierarchical features, specifically including: By using label features that match the dimensional features contained in the hierarchical features, the dimensional features of the highest label level in the hierarchical features are replaced to obtain the redefined hierarchical features.

8. The method as described in claim 1, wherein the user's industry tags at each tag level are output through the output subnet, specifically including: Based on the dimensional features of each tag level in the hierarchical features, the tags of the user's industry under each tag level are output in ascending order of tag level.

9. The method as described in claim 1, wherein the pre-trained recognition model specifically includes: Obtain sample user information and labeling of sample users; The sample user information is input into a pre-trained recognition model, and the fusion features of the sample user are determined through the feature processing subnetwork in the model. The fused features are used as input and input to the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels. According to the hierarchical relationship of the label levels corresponding to each feature extraction layer, the intermediate features extracted by the previous feature extraction layer and the fused features are used as the input of the next feature extraction layer. Through each feature extraction layer, the global dimensional features of the user are determined. Based on the local dimensional features and the global dimensional features, the user's hierarchical features are determined. The hierarchical features are used to characterize the user's industry under each tag level. The hierarchical features are input into the output subnet of the recognition model, and the industry of the sample user is output through the output subnet at each label level. The recognition model is trained with the goal of minimizing the difference between the label output by the recognition model and the labeled label.

10. An industry identification device, comprising: The acquisition module is used to acquire user information, including location information, behavior information, and business license information. The processing module is used to input the user information into a pre-trained recognition model, and determine the user's fusion features through the feature processing subnetwork in the model. The feature processing subnetwork includes at least an extraction layer, a splicing layer, and a fusion layer. The location information, behavior information, and professional license information are input into the extraction layer to extract the user's location features, behavior features, and professional license features. The location features and behavior features are input into the splicing layer to determine the splicing features. The professional license features and the splicing features are input into the fusion layer to determine the fusion features. The local module is used to take the fused features as input and input them into the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels. The global module is used to determine the global dimensional features of the user by taking the intermediate features extracted by the previous feature extraction layer and the fused features as inputs to the next feature extraction layer according to the hierarchical relationship of the label levels corresponding to each feature extraction layer. The hierarchical module is used to determine the user's hierarchical features based on the local dimensional features and the global dimensional features. The hierarchical features are used to characterize the user's industry under each tag level. The output module is used to input the hierarchical features into the output subnet of the recognition model, and output the user's industry labels at each label level through the output subnet.

11. The apparatus of claim 10, wherein the acquisition module is specifically configured to determine the user information of the user based on the user data of the user acquired from multiple data sources.

12. The apparatus of claim 10, wherein the extraction layer comprises at least: First extraction layer, second extraction layer; The processing module is specifically used to input the business license information into the first extraction layer and extract the user's business license features through the first extraction layer. The location information and behavior information are input into the second extraction layer, and the user's location features and behavior features are extracted through the second extraction layer.

13. The apparatus of claim 10, wherein the feature extraction layer comprises: Global output layer and local output layer; The local module is specifically used to take the fused features as input and input them to the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user under different label levels. Specifically, it includes: for each feature extraction layer, determining the intermediate features output by the global output layer of the previous feature extraction layer according to the upper and lower layer relationships of the label levels corresponding to each feature extraction layer; concatenating the intermediate features with the fused features and using the concatenation result as input to the feature extraction layer; and sequentially passing the global output layer and local output layer of the feature extraction layer to determine the local dimensional features output by the feature extraction layer as the local dimensional features under the label level corresponding to the feature extraction layer; wherein, for each label level, the higher the label level, the finer the granularity of the labels under that label level.

14. The apparatus of claim 13, wherein the global module is specifically configured to, for each feature extraction layer, determine the intermediate features output by the global output layer of the previous feature extraction layer in ascending order of the label levels corresponding to each feature extraction layer; concatenate the intermediate features with the fused features, and input the concatenation result as input to the feature extraction layer; and determine the global dimensional features of the user based on the intermediate features output by the global output layer of the feature extraction layer at the highest label level.

15. The apparatus of claim 10, wherein the recognition model further comprises: A matching subnet, wherein the tag features of each tag are pre-stored in the matching subnet; The device further includes a matching module, specifically used to input the hierarchical features into the matching subnet, determine the label features that match the dimensional features contained in the hierarchical features among the label features, and adjust at least some of the dimensional features in the hierarchical features according to the matching results and the matched label features, and redetermine the hierarchical features.

16. The apparatus of claim 15, wherein the tag features of each tag are extracted based on the tag features of each tag at the highest tag level; The matching module is specifically used to replace the label dimension feature with the highest label dimension in the hierarchical features by using the label features that match the hierarchical features, so as to obtain the redefined hierarchical features.

17. The apparatus of claim 10, wherein the output module is specifically configured to output the user's industry tags at each tag level according to the dimensional features of each tag level in the hierarchical features, in order from low to high tag levels.

18. The apparatus of claim 10, further comprising a training module, specifically configured to acquire sample user information and labeled tags of sample users; input the sample user information into a pre-trained recognition model, and determine the fusion features of the sample users through a feature processing subnet in the model; and input the fusion features as input to the feature extraction layer of each label level in the recognition model to determine the local dimensional features of the user corresponding to different label levels. According to the hierarchical relationship of the label levels corresponding to each feature extraction layer, the intermediate features extracted by the previous feature extraction layer and the fused features are used as the input of the next feature extraction layer. Through each feature extraction layer, the global dimensional features of the user are determined. Based on the local dimensional features and the global dimensional features, the hierarchical features of the user are determined. The hierarchical features are used to characterize the features of the user's industry under each label level. The hierarchical features are input into the output subnet of the recognition model, and the industry of the sample user is output through the output subnet at each label level. The recognition model is trained with the goal of minimizing the difference between the label output by the recognition model and the labeled label.

19. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in any one of claims 1 to 9.

20. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 9.

Images