Creative ability information calculation, training methods, devices, media, equipment and platforms

Abstract

The creative ability information calculation, training method, apparatus, medium, device, and platform provided in this disclosure extract feature data from creator data based on feature dimensions corresponding to preset feature types; the feature data is input into a creative ability evaluation model to obtain and display accurate creative ability evaluation results for the creator. On the one hand, the creative ability evaluation model accurately obtains creative ability evaluation results based on the input feature vectors, effectively improving efficiency compared to manual scoring; on the other hand, it helps the platform more reasonably allocate platform resources to each creator based on the creative ability evaluation results, thereby improving evaluation efficiency; on the other hand, by displaying the creative ability evaluation results, it is convenient for creators to understand their own creative level, which is conducive to promoting the improvement of their creative ability.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of artificial intelligence data processing technology. More specifically, the embodiments of the present disclosure relate to creative ability information calculation, training methods, devices, media, equipment and platforms. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the disclosure that are recited in the claims, and no statement herein is admitted to be prior art by inclusion in this section.

[0003] Work-based online platforms, such as online music platforms and online novel platforms, allow creators to showcase their works, generate revenue from them, and generate revenue for the online platforms, achieving a win-win situation. Therefore, online platforms need to accurately and quantitatively analyze creators' creative abilities to facilitate the rational allocation of platform resources to them, but this technology is currently lacking. Summary of the Invention

[0004] In this context, embodiments of the present disclosure provide creative capability information calculation, training methods, devices, media, equipment, and platforms.

[0005] According to the first aspect of the present disclosure, a method for calculating creative ability information is provided, including: obtaining creator data of a creator to be evaluated; extracting feature data on feature dimensions corresponding to each preset feature type based on the creator data; and inputting the feature data into a creative ability evaluation model to obtain and display a creative ability evaluation result.

[0006] According to a second aspect of the present disclosure, a training method for a creative ability evaluation model is provided, comprising: obtaining historical creator data and creative ability type labels of multiple creators; extracting feature data on feature dimensions corresponding to each preset feature type based on each of the historical creator data, and generating training samples based on the feature vectors and the creative ability type labels of the corresponding creators; inputting the training samples into the creative ability evaluation model for training; and adjusting the parameters of the creative ability evaluation model based on a loss function until the training is completed.

[0007] According to the third aspect of the present disclosure, a creative ability information calculation device is provided, including: an acquisition module for acquiring creator data of a creator to be evaluated; a feature extraction module for extracting feature data on feature dimensions corresponding to each preset feature type based on the creator data; and an evaluation module for inputting the feature data into a creative ability evaluation model to obtain and display a creative ability evaluation result.

[0008] According to a fourth aspect of the present disclosure, a training device for a creative ability evaluation model is provided, comprising: an acquisition module for acquiring historical creator data and creative ability type labels of multiple creators; a training sample generation module for extracting feature data on feature dimensions corresponding to each preset feature type based on each of the historical creator data, and generating training samples based on the feature vectors and the creative ability type labels of the corresponding creators; a training module for inputting the training samples into the creative ability evaluation model for training; and, based on a loss function, adjusting the parameters of the creative ability evaluation model until the training is completed.

[0009] According to the fifth aspect of the present disclosure, a storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, it implements: the creative ability information calculation method as described in any one of the first aspects; or the creative ability evaluation model training method as described in any one of the second aspects.

[0010] According to the sixth aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute, by executing the executable instructions: a method for calculating creative ability information as described in any one of the first aspects; or a method for training a creative ability evaluation model as described in any one of the second aspects.

[0011] According to a seventh aspect of the present disclosure, a work-based network platform is provided, comprising the electronic device as described in the sixth aspect.

[0012] The creative ability information calculation, training method, apparatus, medium, device, and platform provided in this disclosure extract feature data from creator data based on feature dimensions corresponding to preset feature types; the feature data is input into a creative ability evaluation model to obtain and display accurate creative ability evaluation results for the creator. On the one hand, the creative ability evaluation model accurately obtains creative ability evaluation results based on the input feature data, effectively improving efficiency compared to manual scoring; on the other hand, it helps the platform more reasonably allocate platform resources to each creator based on the creative ability evaluation results, thereby improving evaluation efficiency; on the other hand, by displaying the creative ability evaluation results, it is convenient for creators to understand their own creative level, which is conducive to promoting the improvement of their creative ability.

[0013] In addition, in addition to learning the characteristics of the feature dimensions corresponding to each preset feature type, the creative ability evaluation model can also be configured to learn the relationship between the feature dimensions of different preset feature types based on the cross-information between the feature fragments of different preset feature types, so as to learn more in-depth and improve the accuracy of the creative ability evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood by reading the detailed description below with reference to the accompanying drawings, in which several embodiments of the present disclosure are shown by way of example and not limitation, wherein:

[0015] Figure 1 A flow chart of a method for calculating creative capability information in an embodiment of the present disclosure is shown.

[0016] Figure 2 A structural diagram of a creative ability evaluation model in one embodiment of the present disclosure is shown.

[0017] Figure 3 A schematic diagram of the processing flow of the shallow feature representation layer in one embodiment of the present disclosure is shown.

[0018] Figure 4 A schematic diagram showing the principle of Hadamard crossover calculation in an embodiment of the present disclosure is shown.

[0019] Figure 5 A structural diagram of a creative ability evaluation model in a specific embodiment of the present disclosure is shown.

[0020] Figure 6 A flowchart of a method for training a creative ability evaluation model in one embodiment of the present disclosure is shown.

[0021] Figure 7 A schematic diagram of the module architecture of a creative capability information calculation device in one embodiment of the present disclosure is shown.

[0022] Figure 8 A schematic diagram of the module architecture of a training device for a creative ability evaluation model in one embodiment of the present disclosure is shown.

[0023] Figure 9 A schematic diagram of a storage medium in an embodiment of the present disclosure is shown.

[0024] Figure 10 A schematic structural diagram of an electronic device in an embodiment of the present disclosure is shown.

[0025] In the drawings, the same or corresponding reference numerals denote the same or corresponding parts. DETAILED DESCRIPTION

[0026] The principles and spirit of the present disclosure will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present disclosure, and are not intended to limit the scope of the present disclosure in any way. Rather, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0027] Those skilled in the art will appreciate that the embodiments of the present disclosure may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present disclosure may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.

[0028] According to the embodiments of the present disclosure, creative ability information calculation, training methods, devices, media, equipment and platforms are provided.

[0029] In this document, any number of elements in the drawings is for illustration and not for limitation, and any naming is for distinction only and does not have any limiting meaning.

[0030] The principles and spirit of the present disclosure are described in detail below with reference to several representative embodiments of the present disclosure.

[0031] The data involved in the present disclosure may be data authorized by the user or fully authorized by all parties, and the implementation methods / embodiments of the present disclosure may be combined with each other. SUMMARY OF THE INVENTION

[0033] Currently, there is a need to evaluate the creative ability of creators on online platforms based on works, such as online music platforms and online novel platforms. However, there is currently a lack of relevant technical solutions that can accurately evaluate creative ability. Specifically, in one method, the creative ability of creators can be evaluated through manual scoring. However, the use of manual scoring increases the manual workload and is relatively inefficient. It may also require the scorers to have a certain ability to appreciate works, which increases the difficulty and cost of implementation and is not suitable for scenarios such as online platforms with a large number of creators. In addition, although there are some evaluation methods based on mathematical statistics in the relevant technology, they cannot be directly applied to the scenario of evaluating the creative ability of creators, because both online music platforms and online novel platforms need to go deep into the specific scenario to explore and select feature dimensions that are more relevant to creative ability for analysis in order to obtain accurate evaluation results.

[0034] In view of this, an embodiment of the present disclosure may provide a method for calculating creative ability information to solve the above problems.

[0035] Exemplary Method Embodiments

[0036] refer to Figure 1 As shown, a flow chart showing a method for calculating creative capability information in one embodiment of the present disclosure is shown. Figure 1 The creative capability information calculation method includes:

[0037] Step S101: Obtain the creator data of the creator to be evaluated.

[0038] In some embodiments, the creator data may be obtained from a work-based web platform.

[0039] Step S102: extracting feature data on feature dimensions corresponding to each preset feature type based on the creator data.

[0040] In some embodiments, the preset feature types and feature dimensions may be determined based on the type of work and the operating mode of the network platform, for example, the number and / or frequency of publications of the work; the number and / or frequency of positive and negative reviews received by the work, etc.

[0041] In some embodiments, the selection of characteristic dimensions can be based on the comprehensiveness of the factors that affect the creative ability evaluation results; or, the characteristic dimensions can be selected based on the factors that mainly affect the creative ability evaluation results. For example, if there are a total of 10 characteristic dimensions, and 6 of them have an impact of more than 90% on the creative ability evaluation results, then these 6 characteristic dimensions are selected.

[0042] Step S103: inputting the characteristic data into a creative ability evaluation model to obtain and display a creative ability evaluation result.

[0043] In some embodiments, the creative ability evaluation model can be implemented based on a machine learning model, such as a deep neural network model. The creative ability evaluation model can be pre-trained to learn the relationship between the feature data of each feature dimension corresponding to each of the preset feature types and the required output results, so that after training, the feature data of the creator to be evaluated can be input into the creative ability evaluation model to obtain the creative ability evaluation result. In a possible example, the creative ability evaluation model can be a creative ability evaluation model of the present disclosure. Figure 6 The training method described in the embodiment is used for training.

[0044] It should be noted that the creative ability evaluation model is a scoring model. In addition to the deep neural network model, it can also be implemented by algorithms such as linear or logistic regression, and its specific implementation method is not limited.

[0045] In some embodiments, the creative ability evaluation model may directly output a creative ability evaluation result; or, it may also output a preliminary evaluation result, which is then processed in a predetermined manner to form the creative ability evaluation result.

[0046] Thus, the creative ability evaluation model automatically calculates the creator's creative ability evaluation result based on the feature vectors of each characteristic dimension determined from the specific scenario. On the one hand, the creative ability evaluation model accurately obtains the creative ability evaluation result based on the input feature vector, effectively improving efficiency compared to manual scoring. On the other hand, it helps the platform more reasonably allocate platform resources to each creator based on the creative ability evaluation results, thereby improving evaluation efficiency. On the other hand, by displaying the creative ability evaluation results, it is convenient for creators to understand their own creative level, which is conducive to promoting the improvement of their creative ability.

[0047] The following uses specific embodiments to exemplify possible implementations of each step in the above creative ability information calculation method.

[0048] Regarding the implementation of step S101, in some embodiments, the creator data may include relevant data of the creator to be evaluated currently on the network platform, such as work data, behavioral interaction data, etc. Taking the online music platform as an example, music creators will publish music works through the online music platform, and users of each platform can visit the creator's column to play music, as well as comment, collect, like, etc. Therefore, the work may be a song directly created by the creator to be evaluated, or it may be a playlist formed by the creator to be evaluated arranging songs, or it may be a dynamic post by the creator to be evaluated. For another example, the behavioral interaction data may include: data on the play, comment, like, and collection of works of other platform users by the creator to be evaluated, or data on the play, comment, like, and collection of works by users of other platforms by the creator to be evaluated.

[0049] Regarding the implementation of step S102, in some embodiments, taking an online music platform as an example, the preset feature types and feature dimensions can be exemplarily shown in Table 1 below. Table 1 provides 25 feature dimensions, which are classified into three preset feature types. Among them, the three preset feature types may include a first feature type related to the original creative behavior of the creator to be evaluated, exemplarily shown in the table as the "original" type; a second feature type related to the dissemination effect of the creator's work, exemplarily shown in the table as the "dissemination effect" type; and a third feature type related to the ecological contribution of the network platform where the creator's work is located, exemplarily shown in the table as the "platform ecological contribution" type.

[0050] Table 1

[0051]

[0052]

[0053]

[0054]

[0055] Specifically, originality corresponds to the creation of works by the creator to be evaluated. In Table 1 above, 8 feature dimensions related to the first feature type are exemplarily shown, including: the total number of songs released, the number of songs released in the last A / B / C days, the number of playlists created, and the number of playlists created in the last A days / B days / C days, etc., and the corresponding identifiers are x0 to x7. These feature dimensions are mainly related to the number of works and the frequency of release. Combined with the description in the table, it can be seen that the feature dimensions related to the original feature type reflect the short-term or long-term creative enthusiasm of the creator to be evaluated. It can be understood that the greater the number of works generated and the higher the frequency, the higher the creative enthusiasm.

[0056] The dissemination effect corresponds to the reaction behavior of the recipients generated by the dissemination of the works of the creator to be evaluated. In the example shown in Table 1 above, 7 feature dimensions related to the second feature type are displayed, including: the total number of complete play times of the song, the average number of play times of the song, the average red heart rate of the released songs, the average complete play rate of the released songs, the percentage of songs with a red heart rate exceeding a%, the percentage of songs with a complete play rate exceeding b%, and the average collection rate of the playlist, and the corresponding labels are x8 to x14. These feature dimensions mainly reflect the quality of the works of the creator to be evaluated. Among them, red heart refers to the inclusion of the song in the red heart playlist.

[0057] The platform ecological contribution corresponds to the contribution made by the creator to be evaluated to the platform ecology through interactions between the creator and other users. In the example shown in Table 1 above, 10 feature dimensions related to the third feature type are displayed, including: number of comments posted, number of comments posted in the last A days / B days / C days, number of dynamic posts, number of dynamic posts in the last A days / B days / C days, average like rate of comments, average like rate of dynamic posts, and corresponding labels are x15 to x24. These feature dimensions mainly reflect the interaction between the creator to be evaluated and other platform users, such as comments, likes, etc.

[0058] It is understandable that the creative ability evaluation results calculated based on the feature data of the feature dimensions related to the above feature types do not only consider direct factors such as the creator's creative volume and frequency, but also consider the quality of the work reflected by the dissemination effect of the creative works, as well as the commercial value of the creator's contribution to the platform ecology, so as to better comprehensively and accurately evaluate the creator's creative ability.

[0059] It should be noted that the feature types and feature dimensions listed above are only exemplary and can be added or deleted according to needs, but are not limited to this.

[0060] The principle of step S103 is described exemplarily.

[0061] In some embodiments, the creative ability evaluation model is implemented based on a deep neural network model.

[0062] like Figure 2 As shown, a structural diagram of the creative ability evaluation model in one embodiment of the present application is shown. In order to improve the accuracy of the model evaluation, in this embodiment, the creative ability evaluation model is implemented as an innovative deep network model. In the process of processing the feature data, the model can learn the relationship between the feature dimensions under different preset feature types by realizing the intersection between the feature dimensions under different preset feature types, such as the relationship between the feature dimensions of native creation type, dissemination effect type, and platform ecological contribution type, so as to obtain more accurate creative ability evaluation results.

[0063] refer to Figure 2 The creative ability evaluation model in this embodiment includes an input layer 201, a hidden layer 202, and an output layer 203. The hidden layer 202 includes a shallow feature representation layer 221, an explicit feature cross layer 222, and an implicit feature cross layer 223. Possibly, each of the shallow feature representation layer 221, the explicit feature cross layer 222, and the implicit feature cross layer 223 may include one or more neural network layers.

[0064] The input layer 201 is used to receive the feature data. For example, referring to the 25 dimensions in the previous table, each of the 25 dimensions corresponds to a creator to be evaluated. The feature values ​​of the creator to be evaluated on these 25 dimensions are obtained to form a 25-dimensional feature data, which is then input into the creative ability evaluation model. Accordingly, the input layer 201 of the creative ability evaluation model may correspond to 25 neurons to respectively transmit each feature value in the feature data.

[0065] The shallow feature representation layer 221 is used to obtain each feature representation result based on the feature data segment mapping corresponding to each preset feature type in the feature data, and obtain a shallow representation result based on each feature representation result and the intersection results between feature representation results.

[0066] You can refer to it together Figure 3 As shown, a schematic diagram of the processing flow of the shallow feature representation layer in one embodiment is shown.

[0067] S301: Obtain a feature representation matrix according to mapping of each feature data segment.

[0068] Specifically, each eigenvalue in a multi-dimensional feature data segment can be mapped to a feature vector, and the feature representation result of the feature data segment can be in the form of a feature matrix in which multiple feature vectors are stacked.

[0069] For example, referring to the 25-dimensional feature dimensions in Table 1 above, the original type corresponds to 8 feature dimensions, that is, it corresponds to an 8-dimensional feature data segment in the 25-dimensional feature data; the dissemination effect type corresponds to 7 feature dimensions, that is, it corresponds to a 7-dimensional feature data segment in the 25-dimensional feature data; the platform ecological contribution type corresponds to 10 feature dimensions, that is, it corresponds to a 10-dimensional feature data segment in the 25-dimensional feature data. Furthermore, the 8-dimensional, 7-dimensional, and 10-dimensional feature data segments can be mapped into feature representation results, respectively, to obtain 3 feature representation results. Furthermore, the 7-dimensional feature data segment, in which each eigenvalue is mapped into a 16-dimensional feature vector, is stacked to form a 7*16 feature matrix, for example, represented as Embedding(7*16).

[0070] In a possible example, the implementation method is based on mapping eigenvalues ​​to eigenvectors. For example, it can be: randomly create initial eigenvectors for each feature dimension in advance, such as 7 16-dimensional eigenvectors; then calculate the weight value according to each eigenvalue, and assign one of the initial eigenvectors to multiply the weight value by the assigned initial eigenvector so that the eigenvalue information is added to this eigenvector, then the 7 initial eigenvectors are each multiplied by the weight to obtain Embedding (7*16). Similarly, the feature data fragments corresponding to the propagation effect type can obtain the feature matrix Embedding (8*16), and the feature data fragments corresponding to the platform ecological contribution type can obtain the feature matrix Embedding (10*16).

[0071] S302: Process each feature representation matrix into a vector form, and map it into a first feature vector of a preset dimension.

[0072] In some embodiments, the feature vectors in the feature representation matrix can be connected (e.g., concat) and / or flattened into a one-dimensional vector form through a fully connected layer (FC), and the preset dimension can be controlled by parameters of the fully connected layer.

[0073] For example, for Embedding (7*16), seven 16-dimensional feature vectors are connected to form a 1*112-dimensional vector, which can be passed through a fully connected layer and multiplied with a 112*A parameter matrix to obtain an A-dimensional first feature vector output, where A is the preset dimension.

[0074] S303: Obtain a feature intersection result based on a first intersection calculation between feature representation matrices.

[0075] In some embodiments, the first cross calculation may be a Hadamard Cross calculation, that is, performing a pairwise cross calculation between each eigenvector in two feature representation matrices, and calculating a Hadamard product between the two crossed eigenvectors.

[0076] You can refer to Figure 4 As shown, the principle of Hadamard cross calculation is demonstrated through an example. The first feature representation matrix is ​​exemplarily shown as [x0, x1, x2, x3], where each element is a feature vector, represented as x0 = [x 0_1 ,x 0_2 ,....,x 0_d ], x 0_d The second feature representation matrix is ​​exemplarily shown as [y0, y1, y2], where each element is a feature vector, such as y0 = [y 0_1, y0 _2, ....,y 0_d ], y 0_d is the eigenvalue. Figure 4 As shown, the Hadamard cross calculation of [x0,x1,x2,x3] and [y0,y1,y2] is [x0*y0,x0*y1,x0*y2,x1*y0,.....x3*y2], and x0*y0 is the Hadamard product [x 0_1 ·y 0_1 ,x 0_2 ·y 0_2 ,....,x 0_d ·y 0_d ], [x0*y0, x0*y1, x0*y2, x1*y0, .....x3*y2] will result in a vector of length 12*d. This means that through the Hadamard crossover, an eigenvector can be calculated from the two matrices as the feature crossover result. For example, the crossover between Embedding(7*16) and Embedding(8*16) yields a 56*16=896-dimensional vector as their feature crossover result.

[0077] S304: Map each feature cross result into a second feature vector of a preset dimension.

[0078] For example, an 896-dimensional vector is obtained by crossing Embedding (7*16) and Embedding (8*16). This vector may be too large and may need to be reduced to a preset dimension, such as 64. Then, a fully connected layer (64) can be passed through, and the 1*896-dimensional feature cross result is multiplied by an 896*B parameter matrix to obtain a B-dimensional second eigenvector. Optionally, the dimensions of the first eigenvector and the second eigenvector can be equal, that is, A=B.

[0079] S305: Output each first eigenvector and second eigenvector as the shallow representation result.

[0080] For example, for the three feature representation matrices of Embedding (7*16), Embedding (8*16), and Embedding (10*16), three first eigenvectors and three second eigenvectors will be generated by cross-generation. The corresponding shallow representation result will be 6 feature vector outputs of preset dimensions.

[0081] Back to Figure 2 In an embodiment, the explicit feature intersection layer is used to obtain an explicit intersection result based on the intersection between partial results corresponding to different preset feature types in the shallow representation results.

[0082] In some embodiments, the explicit feature intersection layer is based on the intersection between partial results corresponding to different preset feature types in the shallow representation results to obtain an explicit intersection result, including any one of the following:

[0083] 1) Concatenate the first eigenvectors and the second eigenvectors to obtain the explicit crossover result.

[0084] For example, according to Embedding(7*16), Embedding(8*16), and Embedding(10*16), six first eigenvectors and second eigenvectors of preset dimensions are obtained. By connecting the six eigenvectors and setting the preset dimension to 64, a 384-dimensional eigenvector will be formed as the explicit crossover result.

[0085] 2) performing a pairwise second crossover calculation based on at least a portion of each of the first eigenvectors and the second eigenvectors to obtain a third eigenvector; and concatenating each of the first eigenvectors, the second eigenvectors, and the third eigenvector to obtain the explicit crossover result.

[0086] For example, the second crossover calculation can be performed by subtracting or calculating the Hadamard product between each pair of eigenvectors. The resulting third eigenvector can represent the distance or correlation between each pair of eigenvectors involved in the calculation. The third eigenvector is then added to the explicit crossover result, so that the explicit crossover result carries the information represented by the third eigenvector.

[0087] 3) calculating the attention weights of the first eigenvectors and the second eigenvectors, and adjusting the eigenvalues ​​of the first eigenvectors and the second eigenvectors according to the attention weights; and connecting the adjusted first eigenvectors and the second eigenvectors to obtain the explicit cross-concatenation result.

[0088] Specifically, considering that the contributions of each first eigenvector and second eigenvector to the final creative ability evaluation may be different, in order to add this part of information, the self-attention mechanism can be used to calculate the respective attention weights of the first eigenvector and the second eigenvector to highlight the information that contributes more to the evaluation results.

[0089] There are more than one way to calculate the attention weight and the corresponding explicit crossover result. In one example, by calculating the dot product of each of the six eigenvectors with each of the other eigenvectors, for example, by calculating the dot product between eigenvector A and B, C, D, E, and F, we can obtain the probability values ​​z1, z2, z3, z4, and z5 respectively. The average value z of z1, z2, z3, z4, and z5 is used as the attention weight of A, and the attention weights of B, C, D, E, and F can be obtained by analogy. It can be found that in this process, crossover actually occurs between A and F through the dot product calculation. By calculating z*A=A1, the adjusted eigenvector A1 can be obtained, and similarly, B1, C1, D1, and E1 can be obtained. Optionally, the above attention weight calculation process can be repeated for A1 to E1 to obtain the new eigenvector after considering the attention weight. The above iterative process can be continued multiple times so that the relationship between the eigenvectors can more accurately describe the relationship between the actual preset feature types, thereby obtaining a more accurate creative ability evaluation result.

[0090] In another example, the calculation method of the attention weight can also be relatively simplified. For example, the feature vectors A, B, C, D, E, and F are respectively multiplied by a vector of corresponding dimensions to obtain a probability value between [0, 1] as the attention weight. For example, if A is a 1*64-dimensional vector, multiplying it by a 64*1 vector can obtain the corresponding probability value, and multiplying it by A can obtain the adjusted feature vector A2. In this example, although there is no direct intersection between A and F, the back propagation of the loss calculated by the loss function during training of the deep neural network model to adjust the parameters will actually be affected by the adjusted A-F relationship, that is, there is a certain degree of intersection.

[0091] The implicit feature cross layer calculates the implicit cross result based on the explicit cross result as the hidden layer output result. In some embodiments, the implicit feature cross layer includes at least one fully connected layer, and the fully connected layer outputs a feature vector in the form of a second preset dimension. The fully connected layer can be configured with a first activation function, and the explicit cross result is mapped to an intermediate vector of a preset dimension through the fully connected layer, and the intermediate vector is processed by the first activation function to obtain the implicit cross result as the hidden layer output result. In some implementation examples, there can be multiple fully connected layers, and their dimensions can be gradually reduced, such as reducing the dimension to 64 dimensions through a FC (64), and then reducing the dimension to 32 dimensions through a FC (32). The number of FC layers needs to be set appropriately to take into account both output accuracy and efficiency. Increasing the number of FC layers will be more helpful in refining deep information, but there will be a certain amount of information loss after each FC layer. The increase in FC layers will also cause an increase in information loss, and the increase in FC layers also means more model parameters and a more complex structure. In some embodiments, the first activation function used by at least one FC layer in the implicit feature cross layer may be a hyperbolic tangent function (tanh), which has a value in [-1, 1] and can carry relatively more information.

[0092] The output layer calculates the model output result based on the parameters of the current layer and the output result of the hidden layer, and obtains and displays the creative ability evaluation result based on the model output result. In some embodiments, the output layer can be a fully connected layer FC(1) to calculate an output value. The output layer can be configured with a second activation function, such as sigmoid, so that the output is a probability value of [0,1].

[0093] The probability value can be used to form a creative ability evaluation result. A larger probability value may correspond to a stronger creative ability of the creator to be evaluated. In one example, the probability value can be used to form a creative ability evaluation result. Get the creative ability score of the creator to be evaluated and display the creative ability score. Specifically, the formula The probability value is converted into z between 1 and 100 and output as the creative ability score. ceil() means returning the smallest integer greater than or equal to the current given number, for example: ceil(20.5) = 21, that is The value is 0.205, which is processed into a creative ability score of 21 and displayed to the user.

[0094] In one example, creative ability type information can be obtained based on the model output and displayed. The creative ability type can be "high" or "low," specifically by comparing the probability value output by the model with a preset threshold, such as 0.6. When the probability value output by the model reaches 0.6, it indicates that the creative ability of the creator to be evaluated is high, and the creative ability type is displayed to the user as "high." Otherwise, the creative ability type is displayed as "low."

[0095] In one example, the creative ability scores and creative ability type information in the above two examples can be displayed together. For example, the model output probability value in the above example is 0.205, and the corresponding creative ability type is "low", so the output is "low" and the corresponding creative ability score is "21 points", so as to more quantitatively display the user's actual creative ability. Users can compare the creative ability scores to more clearly determine the specific gap in each other's creative ability. For example, two creators both have a low creative ability type, but by comparing 59 points and 20 points, it can be found that there is actually a large gap in their creative ability, which is more conducive to creators to intuitively understand their own creative ability level.

[0096] like Figure 5 FIG. 1 is a schematic diagram showing the structure of a creative ability evaluation model in a specific embodiment. The creative ability evaluation model is presented as a layered multi-tower parallel deep cross network.

[0097] As shown in the figure, the feature data of 25 feature dimensions of the creator to be evaluated are input into the input layer of the creative ability evaluation model, and further passed to the shallow feature representation layer, where the feature data segments of the first feature type corresponding to originality (8 feature dimensions), the feature data segments of the second feature type corresponding to the dissemination effect (7 feature dimensions), and the feature data segments of the third feature type corresponding to the platform ecological contribution (10 feature dimensions) are mapped to Embedding (8*16), Embedding (7*16) and Embedding (10*16) respectively.

[0098] Embedding(8*16), Embedding(7*16) and Embedding(10*16) enter a tower respectively, and after Concat processing, they are converted into 128-dimensional, 112-dimensional and 160-dimensional vector forms, and then input into FC(64) to form three 64-dimensional first feature vectors. Any combination of two of Embedding(8*16), Embedding(7*16) and Embedding(10*16) enters other towers respectively, and after Hadamard cross calculation, three feature cross results are obtained, which are 56*16=896-dimensional, 70*16=1120-dimensional and 80*16=1280-dimensional vectors respectively, and then respectively input into FC(64) to form three 64-dimensional vectors. Thus, the shallow feature representation layer obtains six 64-dimensional vectors as the shallow representation result output to be input into the explicit cross layer.

[0099] The explicit cross layer is configured to perform a concatenation operation on six 64-dimensional vectors to form a 384-dimensional vector as the explicit cross result output to be input to the implicit cross layer.

[0100] The implicit cross layer contains two FC layers, FC(64) and FC(32). The 384-dimensional vector is first processed into 64 dimensions by FC(64) and calculated by the first activation function tanh. Then it is processed into 32 dimensions by FC(32) and calculated by the first activation function tanh. The obtained 32-dimensional vector is output as the hidden layer output result and input into the output layer.

[0101] The output layer contains an FC layer, FC(1), which converts the 32-dimensional vector into a 1-dimensional vector and converts it into a probability value of [0,1] through the second activation function sigmod, which is output as the model output result.

[0102] It should be noted that, in combination with the previous embodiments, Figure 5 The model network structure is only exemplary and can be changed as needed, but is not limited to this.

[0103] The creative ability evaluation model is implemented based on a machine learning model and needs to be trained to perform accurate evaluation. In some embodiments, the present disclosure may also provide a training method for the creative ability evaluation model.

[0104] like Figure 6 As shown, a flow chart showing the training method of the creative ability evaluation model in one embodiment of the present application is shown. It should be noted that the creative ability evaluation model used in the previous embodiment can be trained by the training method in this embodiment or by other training methods, and is not necessarily subject to Figure 6 Limitations of the presented training method.

[0105] exist Figure 6 In the training method, the training method includes:

[0106] Step S601: Acquire historical creator data and creative ability type labels of multiple creators.

[0107] In some embodiments, historical creator data of multiple creators, such as hundreds of creators, can be obtained to construct a training sample set. For example, the historical creator data can be obtained from a work-based online platform. The historical creator data can include relevant data about the creator on the online platform, such as historical work data and historical behavioral interaction data.

[0108] In some embodiments, the method for obtaining the creative ability type label may be to first obtain the creator's creative ability reference score; then, based on the comparison result of the creative ability reference score and a preset threshold, obtain the creator's creative ability type label.

[0109] The reference score can, for example, come from an expert score. If this score is used directly as a label so that the trained creative ability evaluation model directly outputs a score of a similar principle, the score quantification is too detailed, and the accuracy of the actual description of the creator's creative ability is not ideal. For example, the difference between 75 points and 80 points in creative ability cannot be accurately learned by the creative ability evaluation model. Accordingly, the score made by the creative ability evaluation model is actually not accurate. Therefore, in the embodiment of the present disclosure, the reference score is abandoned as a classification label, and the creative ability type information "high" and "low" (which can be represented by, for example, "1" and "0") obtained according to the reference score is used as a label, so that the trained creative ability evaluation model is a binary classification model based on the creative ability type information, and a probability value is output to indicate that the creative ability tends to be "high" or "low". This not only makes the evaluation result of the creative ability evaluation model more accurate, but also simplifies the structure and parameters of the creative ability evaluation model compared to a large number of classifications using reference scores as classification labels, thereby improving the model calculation efficiency.

[0110] To illustrate this through an example, the reference score z obtained from the expert, which can be an integer between 1 and 100, can be converted into a category value of 0 or 1. The specific conversion can be obtained by comparing the reference score with a preset threshold to obtain a corresponding binary creative ability type label. The preset threshold can be exemplarily set to 60, and z is converted into a creative ability type label y through the following formula.

[0111]

[0112] Among them, users whose reference score of creative ability is not less than 60 are called high creativity, represented by 1, that is, y=1; users whose music creativity score is less than 60 are called low creativity, represented by 0, that is, y=0.

[0113] Step S602: extracting feature data on feature dimensions corresponding to each preset feature type based on the historical creator data, and generating training samples based on the feature vectors and the creative ability type labels of the corresponding creators.

[0114] In some embodiments, each of the preset feature types includes at least one of the following: a first feature type related to the creator's original creative behavior; a second feature type related to the dissemination effect of the creator's work; and a third feature type related to the ecological contribution of the network platform where the creator's work is located. For example, the three preset feature types and 25 feature dimensions in Table 1 can be referenced.

[0115] According to the value ranges of each feature dimension in Table 1 above, the values ​​of the feature data of different feature dimensions may be between [0, 1] or between [0, +∞]. Therefore, in some embodiments, the feature values ​​of multiple creators in each feature dimension can be processed by a normalization method to obtain corresponding processed feature values, and the feature data of each creator can be constructed based on each processed feature value. In this way, the value of [0, +∞] can also be converted to [0, 1], which can improve the convergence speed of the model training stage and improve the training effect of the model.

[0116] Exemplarily, the normalization method may adopt, for example, a Min-Max Normalization method to process the feature data into feature values ​​ranging from [0, 1] to form a feature vector.

[0117] The Min-Max Normalization calculation formula is as follows:

[0118]

[0119] Here, min and max are the maximum and minimum values ​​of x, respectively. For example, let's take the feature data of total number of released songs. Suppose there are five users with total released songs of 5, 10, 15, 20, and 50, respectively. Then, min = 5 and max = 50. After Min-MaxNormalization, these become: 0, 1 / 9, 2 / 9, 1 / 3, and 1.

[0120] The normalized feature values ​​are combined into feature data for each user according to the preset feature classifications in Table 1 above, which is represented by x in the following formula:

[0121] x=[x0,x1,x2,...,x 24 ]

[0122] Each element is the feature value of the user in each feature dimension.

[0123] Furthermore, let's take an example to illustrate the construction of training samples based on feature data and labels. Suppose we collect creator data from 500 platform users and calculate each feature data in 25 feature dimensions. Combine each feature data calculated with the classification label y one by one to form the final 500 training samples. Use t0 to t499 to identify these 500 training samples, and we can get

[0124]

[0125] in, represents the i-th feature of the j-th sample, j∈{0, 1, ..., 499}, i∈{0, 1, 2, ..., 24}, y i Represents the creative ability type label of the i-th sample.

[0126] Step S603: input the training samples into the creative ability evaluation model for training.

[0127] In some embodiments, the creative ability evaluation model is implemented based on a deep neural network model. For example, the deep neural network model can be, for example Figure 2 ,or Figure 5 The model structure shown here, the deep neural network model, may include an input layer, a hidden layer, and an output layer. The hidden layer includes a shallow feature representation layer, an explicit feature cross layer, and an implicit feature cross layer. Since the specific structure and principles of this model have been described in detail in the previous embodiment, reference can be made here and no further details will be given.

[0128] Step S604: Based on the loss function, adjust the parameters of the creative ability evaluation model until the training is completed.

[0129] Specifically, model training is an iterative process of calculating loss, adjusting parameters, and then calculating loss again. Therefore, based on the loss function of this round, the parameters of the creative ability evaluation model can be derived and updated; based on the updated parameters, the next round of loss function is determined until the loss function meets the conditions and the training is ended.

[0130] The following describes this in detail through specific examples.

[0131] The BP (Back Propagation) algorithm can be used to train the parameters of the creative ability evaluation model. Before using the BP algorithm, it is necessary to define the loss function for this task. For a binary classification model, the loss function can be defined as a logarithmic loss function, as shown below:

[0132]

[0133] Where N=500, which means there are 500 samples. The BP algorithm uses gradient descent to update the parameters. The specific formula is as follows:

[0134]

[0135] in, Denotes the derivative of L with respect to w, and η is the learning rate. Repeated gradient descent updates w, and when w eventually converges, training can be stopped.

[0136] For example, a training algorithm process can be as follows:

[0137] Randomly initialize the model parameters; set the total number of model iterations, recorded as epoch; set the minimum convergence value of the loss function, recorded as σ; set the learning rate, recorded as η. For the tth iteration, execute the following strategy:

[0138] The 25-dimensional features calculated for N users are input into the deep neural network model. The model architecture can be, for example, Figure 5 The example, or other DNN models, calculate the scores of the model output of N users, which are recorded as

[0139] According to the Loss Function formula Calculate the loss function L for this round and determine whether L is less than or equal to σ. If so, the model has converged and training stops immediately; otherwise, training continues. The loss is backpropagated and the model parameters are updated using gradient descent.

[0140] Exemplary Device Embodiments

[0141] After introducing the exemplary method embodiments of the present disclosure, next, reference is made to Figure 7 A creative ability information calculation device according to an exemplary embodiment of the present disclosure will be described.

[0142] Since the various functional modules or sub-modules of the creative ability information calculation device in the embodiment of the present disclosure are based on the same principles as the corresponding steps or sub-steps of the creative ability information calculation method in the above-mentioned exemplary method embodiment, the specific implementation in this embodiment can refer to the previous content, and the same technical content will not be repeated.

[0143] refer to Figure 7 As shown, an exemplary embodiment of the present disclosure provides a creative ability information calculation device 700, which is characterized by including: an acquisition module 701 for acquiring creator data of a creator to be evaluated; a feature extraction module 702 for extracting feature data on feature dimensions corresponding to each preset feature type based on the creator data; and an evaluation module 703 for inputting the feature data into a creative ability evaluation model to obtain and display a creative ability evaluation result.

[0144] In some embodiments, the evaluation module 703 includes: a creative ability score acquisition module, used to input the feature data into the creative ability evaluation model to obtain the creative ability score of the creator to be evaluated; and a display module, used to display the creative ability score.

[0145] In some embodiments, the evaluation module 703 includes: a creative ability type acquisition module, used to input the feature data into a creative ability evaluation model to obtain creative ability type information; and a display module, used to display the creative ability type information.

[0146] In some embodiments, the creative ability evaluation model is implemented based on a deep neural network model; the deep neural network model includes: an input layer, a hidden layer and an output layer, and the hidden layer includes: a shallow feature representation layer, an explicit feature cross layer and an implicit feature cross layer; the input layer is used to receive the feature data; the shallow feature representation layer is used to obtain each feature representation result based on the feature data segment mapping corresponding to each preset feature type in the feature data; based on each feature representation result and the cross results between the feature representation results, a shallow representation result is obtained; the explicit feature cross layer is used to obtain an explicit cross result based on the cross between the partial results corresponding to different preset feature types in the shallow representation result; the implicit feature cross layer is used to calculate an implicit cross result based on the explicit cross result as the hidden layer output result; the output layer is used to calculate the model output result based on the parameters of this layer and the hidden layer output result, and to obtain and display the creative ability evaluation result based on the model output result.

[0147] In some embodiments, the shallow feature representation layer includes: a first mapping module, which is used to map each feature data segment to obtain a feature representation matrix; a second mapping module, which processes each feature representation matrix into a vector form and maps it into a first eigenvector of a preset dimension; a cross calculation module, which is used to obtain a feature cross result based on the first cross calculation between feature representation matrices; a third mapping module, which is used to map each feature cross result into a second eigenvector of a preset dimension; and an output module, which is used to output each first eigenvector and second eigenvector as the shallow representation result.

[0148] In some embodiments, the explicit feature cross-layer includes any one of the following: 1) a first explicit cross-processing module, used to connect each first eigenvector and the second eigenvector to obtain the explicit cross-result; 2) a second explicit cross-processing module, used to perform a pairwise second cross-calculation based on at least part of each first eigenvector and the second eigenvector to obtain a third eigenvector; connect each first eigenvector, second eigenvector and third eigenvector to obtain the explicit cross-result; 3) a third explicit cross-processing module, used to calculate the attention weight of each first eigenvector and the second eigenvector, and adjust the eigenvalues ​​of each first eigenvector and the second eigenvector according to the attention weight; connect based on the adjusted first eigenvector and the second eigenvector to obtain the explicit cross-result.

[0149] In some embodiments, the implicit feature crossover layer includes at least one fully connected layer, which outputs a feature vector in a second predetermined dimension. The fully connected layer may be configured with a first activation function, and the fully connected layer maps the explicit crossover result into an intermediate vector of a predetermined dimension, and processes the intermediate vector through the first activation function to obtain the implicit crossover result as the hidden layer output result.

[0150] In some embodiments, each of the preset feature types includes at least one of the following: a first feature type related to the original creative behavior of the creator; a second feature type related to the dissemination effect of the creator's work; and a third feature type related to the ecological contribution of the network platform where the creator's work is located.

[0151] refer to Figure 7 A creative ability information calculation device according to an exemplary embodiment of the present disclosure will be described.

[0152] Since the various functional modules or sub-modules of the training device for the creative ability evaluation model in the embodiment of the present disclosure have the same principles as the corresponding steps or sub-steps of the training method for the creative ability evaluation model in the above-mentioned exemplary method embodiment, the specific implementation in this embodiment can refer to the previous content, and the same technical content will not be repeated.

[0153] refer to Figure 8 As shown, an exemplary embodiment of the present disclosure provides a training device 800 for a creative ability evaluation model, including: an acquisition module 801, used to acquire historical creator data and creative ability type labels of multiple creators; a training sample generation module 802, used to extract feature data on feature dimensions corresponding to each preset feature type based on each of the historical creator data, and generate training samples based on the feature vectors and the creative ability type labels of the corresponding creators; a training module 803, used to input the training samples into the creative ability evaluation model for training; and, based on the loss function, adjust the parameters of the creative ability evaluation model until the training is completed.

[0154] In some embodiments, the creative ability evaluation model is implemented based on a deep neural network model; the deep neural network model includes: an input layer, a hidden layer and an output layer; the hidden layer includes: a shallow feature representation layer, an explicit feature cross layer and an implicit feature cross layer.

[0155] In some embodiments, the acquisition module 801 includes: a reference score acquisition module for acquiring the creator's creative ability reference score; a comparison module for obtaining the creator's creative ability type label based on the comparison result of the creative ability reference score and a preset threshold.

[0156] In some embodiments, each of the preset feature types includes at least one of the following: a first feature type related to the original creative behavior of the creator; a second feature type related to the dissemination effect of the creator's work; and a third feature type related to the ecological contribution of the network platform where the creator's work is located.

[0157] In some embodiments, the training sample generation module 802 includes: a normalization processing module, which is used to process the characteristic values ​​of multiple creators in each characteristic dimension through a normalization method to obtain corresponding processed characteristic values; and a characteristic data generation module, which is used to construct the characteristic data of each creator based on each processed characteristic value.

[0158] In some embodiments, the creative ability evaluation model is implemented based on a deep neural network model; based on the loss function, the parameters of the creative ability evaluation model are adjusted to complete the training, including: based on the current round of loss function, the parameters of the creative ability evaluation model are derived and updated; based on the updated parameters, the next round of loss function is determined until the loss function meets the conditions and the training is ended.

[0159] Exemplary Storage Media

[0160] After introducing the method and apparatus of the exemplary embodiment of the present disclosure, next, reference is made to Figure 9A storage medium according to an exemplary embodiment of the present disclosure is described.

[0161] refer to Figure 9 As shown, a storage medium 900 according to an embodiment of the present disclosure is described, which may contain program code and can be run on a device, such as a server, to implement the execution of each step and sub-step in the above method embodiment of the present disclosure. In this document, a readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0162] The program code can be in any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0163] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0164] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0165] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0166] Exemplary electronic devices

[0167] After introducing the storage medium of the exemplary embodiment of the present disclosure, next, reference is made to Figure 9 An electronic device according to an exemplary embodiment of the present disclosure will be described.

[0168] Figure 10 The electronic device 1000 shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure. The electronic device 900 can be implemented in a server, etc.

[0169] like Figure 10 As shown, electronic device 1000 is implemented as a general-purpose computing device. Components of electronic device 1000 may include, but are not limited to, the aforementioned at least one processing unit 1010, the aforementioned at least one storage unit 1020, and a bus 1030 connecting various system components (including storage unit 1020 and processing unit 1010).

[0170] The storage unit stores program codes, which can be executed by the processing unit 1010, so that the processing unit 1010 performs the steps and sub-steps of the method described in the above embodiment of the present disclosure. For example, the processing unit 1010 can perform the following steps: Figure 1 、 Figure 3 or Figure 6 etc. Steps in the embodiments.

[0171] In some embodiments, the storage unit 1020 may include a volatile storage unit, such as a random access memory unit (RAM) 10201 and / or a cache memory unit 10202 , and may further include a read-only memory unit (ROM) 10203 .

[0172] In some embodiments, the storage unit 1020 may also include a program / utility 10204 having a set (at least one) of program modules 10205, such program modules 10205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0173] In some embodiments, bus 1030 may include a data bus, an address bus, and a control bus.

[0174] In some embodiments, the electronic device 1000 may also communicate with one or more external devices 1100 (e.g., a keyboard, a pointing device, a Bluetooth device, etc.), and such communication may be performed via an input / output (I / O) interface 1050. Optionally, the electronic device 1000 also includes a display unit 1040, which is connected to the input / output (I / O) interface 1050 for display. Furthermore, the electronic device 1000 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 1060. As shown, the network adapter 1060 communicates with other modules of the electronic device 1000 via a bus 1030. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0175] It should be noted that although several modules or sub-modules such as the creative ability information calculation device and the creative ability model training device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiment of the present disclosure, the features and functions of two or more units / modules described above can be concretized in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided to be concretized by multiple units / modules.

[0176] Furthermore, although the operations of the disclosed method are described in a particular order in the accompanying drawings, this does not require or imply that the operations must be performed in this particular order, or that all illustrated operations must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0177] Although the principles of the present disclosure have been described with reference to several specific embodiments, it should be understood that the present disclosure is not limited to the specific embodiments disclosed, and the division into various aspects does not mean that the features of these aspects cannot be combined to benefit. Such division is only for convenience of expression. The present disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A creative ability information calculation method, characterized in that: include: Get the creator data of the creator to be evaluated; Extracting feature data on feature dimensions corresponding to each preset feature type based on the creator data; Inputting the characteristic data into a creative ability evaluation model to obtain and display a creative ability evaluation result; The creative ability evaluation model is implemented based on a deep neural network model; the deep neural network model includes: An input layer and a hidden layer for receiving the feature data, and an output layer for calculating a model output result based on the parameters of the current layer and the output result of the hidden layer, and obtaining and displaying a creative ability evaluation result based on the model output result; The hidden layer includes: a shallow feature representation layer, an explicit feature cross layer and an implicit feature cross layer; The shallow feature representation layer obtains each feature representation result based on the feature data segment mapping corresponding to each preset feature type in the feature data; obtains a shallow representation result based on each feature representation result and the intersection result between the feature representation results; The explicit feature intersection layer obtains an explicit intersection result based on the intersection between partial results corresponding to different preset feature types in the shallow representation results; and The implicit feature cross layer calculates an implicit cross result based on the explicit cross result to serve as the hidden layer output result.

2. The creative ability information calculation method according to claim 1, characterized in that: The step of inputting the characteristic data into a creative ability evaluation model to obtain and display a creative ability evaluation result includes: Inputting the characteristic data into a creative ability evaluation model to obtain a creative ability score of the creator to be evaluated; The creative ability score is displayed.

3. The creative ability information calculation method according to claim 1, characterized in that: The step of inputting the characteristic data into a creative ability evaluation model to obtain and display a creative ability evaluation result includes: The characteristic data is input into a creative ability evaluation model to obtain creative ability type information; and the creative ability type information is displayed.

4. The creative ability information calculation method according to claim 1, characterized in that: Obtaining each feature representation result based on mapping feature data segments corresponding to each preset feature type in the feature data; Based on each of the feature representation results and the cross-results between the feature representation results, a shallow representation result is obtained, including: Obtain a feature representation matrix based on mapping each feature data segment; Processing each feature representation matrix into a vector form and mapping it into a first eigenvector of a preset dimension; Based on the first cross calculation between the feature representation matrices, a feature cross result is obtained; Mapping each feature intersection result into a second feature vector of a preset dimension; The first eigenvectors and the second eigenvectors are output as the shallow representation results.

5. The creative ability information calculation method according to claim 1, characterized in that: The explicit feature intersection layer is based on the intersection between the partial results corresponding to different preset feature types in the shallow representation results to obtain an explicit intersection result, including any one of the following; 1) connecting each first eigenvector and the second eigenvector to obtain the explicit crossover result; 2) performing a pairwise second crossover calculation based on at least a portion of each of the first eigenvectors and the second eigenvectors to obtain a third eigenvector; and concatenating each of the first eigenvectors, the second eigenvectors, and the third eigenvector to obtain the explicit crossover result; 3) calculating the attention weights of the first eigenvectors and the second eigenvectors, and adjusting the eigenvalues ​​of the first eigenvectors and the second eigenvectors according to the attention weights; and connecting the adjusted first eigenvectors and the second eigenvectors to obtain the explicit cross-concatenation result.

6. The creative ability information calculation method according to claim 1, characterized in that: The implicit feature cross layer includes at least one fully connected layer, and the fully connected layer outputs a feature vector in a second preset dimensional form; the implicit feature cross layer calculates an implicit cross result based on the explicit cross result to serve as a hidden layer output result, including: Mapping the explicit cross-convolution result into an intermediate vector of a second preset feature dimension through a fully connected layer; The intermediate vector is processed by a first activation function to obtain the implicit cross-pollination result as the hidden layer output result.

7. The creative ability information calculation method according to claim 1 or 3, characterized in that: Each of the preset feature types includes at least one of the following: a first feature type related to the original creative behavior of the creator; The second type of characteristics is related to the dissemination effect of the creator's work; The third characteristic type is related to the ecological contribution of the network platform where the creator's work is located.

8. A training method for a creative ability evaluation model, characterized in that: include: Obtain historical creator data and creative ability type labels for multiple creators; Extracting feature data on feature dimensions corresponding to each preset feature type based on each of the historical creator data, and generating training samples based on the feature data and the creative ability type label of the corresponding creator; Inputting the training samples into the creative ability evaluation model for training; Based on the loss function, adjusting the parameters of the creative ability evaluation model until training is completed; The creative ability evaluation model is implemented based on a deep neural network model; the deep neural network model includes: An input layer and a hidden layer for receiving the feature data, and an output layer for calculating a model output result based on the parameters of the current layer and the output result of the hidden layer, and obtaining and displaying a creative ability evaluation result based on the model output result; The hidden layer includes: a shallow feature representation layer, an explicit feature cross layer and an implicit feature cross layer; The shallow feature representation layer obtains each feature representation result based on the feature data segment mapping corresponding to each preset feature type in the feature data; obtains a shallow representation result based on each feature representation result and the intersection result between the feature representation results; The explicit feature intersection layer obtains an explicit intersection result based on the intersection between partial results corresponding to different preset feature types in the shallow representation results; and The implicit feature cross layer calculates an implicit cross result based on the explicit cross result to serve as the hidden layer output result.

9. The training method according to claim 8, characterized in that The step of obtaining the creator's creative capability type label includes: Obtain the creator's creative ability reference score; Based on the comparison result of the creative ability reference score and the preset threshold, the creative ability type label of the creator is obtained.

10. The training method according to claim 8 or 9, characterized in that: Each of the preset feature types includes at least one of the following: a first feature type related to the original creative behavior of the creator; The second type of characteristics is related to the dissemination effect of the creator's work; The third characteristic type is related to the ecological contribution of the network platform where the creator's work is located.

11. The training method according to claim 8, characterized in that: The feature data extracted based on the historical creator data on the feature dimensions corresponding to each preset feature type includes: The eigenvalues ​​of multiple creators in each eigendimension are processed by a normalization method to obtain corresponding processed eigenvalues; The feature data of each creator is constructed based on the processed feature values.

12. The training method according to claim 8, characterized in that: The creative ability evaluation model is implemented based on a deep neural network model; and adjusting the parameters of the creative ability evaluation model based on the loss function to complete the training includes: Based on the loss function of this round, the parameters of the creative ability evaluation model are derived and updated; Based on the updated parameters, the next round of loss function is determined until the loss function meets the conditions and the training ends.

13. A creative ability information calculation device, characterized in that: include: An acquisition module, used to obtain the creator data of the creator to be evaluated; A feature extraction module, configured to extract feature data on feature dimensions corresponding to each preset feature type based on the creator data; An evaluation module, configured to input the characteristic data into a creative ability evaluation model to obtain and display a creative ability evaluation result; The creative ability evaluation model is implemented based on a deep neural network model; the deep neural network model includes: An input layer and a hidden layer for receiving the feature data, and an output layer for calculating a model output result based on the parameters of the current layer and the output result of the hidden layer, and obtaining and displaying a creative ability evaluation result based on the model output result; The hidden layer includes: a shallow feature representation layer, an explicit feature cross layer and an implicit feature cross layer; The shallow feature representation layer obtains each feature representation result based on the feature data segment mapping corresponding to each preset feature type in the feature data; obtains a shallow representation result based on each feature representation result and the intersection result between the feature representation results; The explicit feature intersection layer obtains an explicit intersection result based on the intersection between partial results corresponding to different preset feature types in the shallow representation results; and The implicit feature cross layer calculates an implicit cross result based on the explicit cross result to serve as the hidden layer output result.

14. The creative ability information calculation device according to claim 13, characterized in that: The modules evaluated include: A creative ability score acquisition module, configured to input the feature data into a creative ability evaluation model to obtain the creative ability score of the creator to be evaluated; A display module is used to display the creative ability score.

15. The creative ability information calculation device according to claim 13, characterized in that: The modules evaluated include: a creative ability type acquisition module, configured to input the characteristic data into a creative ability evaluation model to acquire creative ability type information; A display module is used to display the creative capability type information.

16. The creative ability information calculation device according to claim 13, characterized in that: The shallow feature representation layer includes: A first mapping module, configured to map each feature data segment to obtain a feature representation matrix; A second mapping module processes each feature representation matrix into a vector form and maps it into a first feature vector of a preset dimension; A cross calculation module, configured to obtain a feature cross result based on a first cross calculation between feature representation matrices; A third mapping module, configured to map each feature cross result into a second feature vector of a preset dimension; An output module is used to output each first eigenvector and second eigenvector as the shallow representation result.

17. The creative ability information calculation device according to claim 13, characterized in that: The explicit feature cross layer includes any of the following: 1) a first explicit crossover processing module, configured to connect each first eigenvector and second eigenvector to form a first intermediate vector as the explicit crossover result; 2) a second explicit crossover processing module, configured to perform a pairwise second crossover calculation based on at least a portion of each of the first eigenvectors and the second eigenvectors to obtain a third eigenvector; and concatenate each of the first eigenvectors, the second eigenvectors, and the third eigenvector to obtain the explicit crossover result; 3) A third explicit cross-processing module, configured to calculate the attention weights of the first and second eigenvectors, and adjust the eigenvalues ​​of the first and second eigenvectors according to the attention weights; and to form the explicit cross-processing result based on the connection of the adjusted first and second eigenvectors.

18. The creative ability information calculation device according to claim 13, characterized in that: The implicit feature crossover layer includes at least one fully connected layer, which outputs a feature vector in the form of a second preset dimension; the fully connected layer is used to map the explicit crossover result into an intermediate vector of the second preset feature dimension; and, the intermediate vector is processed by a first activation function to obtain the implicit crossover result as the hidden layer output result.

19. The creative ability information calculation device according to claim 13 or 15, characterized in that: Each of the preset feature types includes at least one of the following: a first feature type related to the original creative behavior of the creator; The second type of characteristics is related to the dissemination effect of the creator's work; The third characteristic type is related to the ecological contribution of the network platform where the creator's work is located.

20. A training device for a creative ability evaluation model, characterized in that: include: An acquisition module, used to obtain historical creator data and creative ability type labels of multiple creators; A training sample generation module is used to extract feature data on feature dimensions corresponding to each preset feature type based on each of the historical creator data, and generate training samples based on the feature data and the creative ability type label of the corresponding creator; A training module, configured to input the training samples into the creative ability evaluation model for training; and, based on a loss function, adjust the parameters of the creative ability evaluation model until the training is completed; The creative ability evaluation model is implemented based on a deep neural network model; the deep neural network model includes: An input layer and a hidden layer for receiving the feature data, and an output layer for calculating a model output result based on the parameters of the current layer and the output result of the hidden layer, and obtaining and displaying a creative ability evaluation result based on the model output result; The hidden layer includes: a shallow feature representation layer, an explicit feature cross layer and an implicit feature cross layer; The shallow feature representation layer obtains each feature representation result based on the feature data segment mapping corresponding to each preset feature type in the feature data; obtains a shallow representation result based on each feature representation result and the intersection result between the feature representation results; The explicit feature intersection layer obtains an explicit intersection result based on the intersection between partial results corresponding to different preset feature types in the shallow representation results; and The implicit feature cross layer calculates an implicit cross result based on the explicit cross result to serve as the hidden layer output result.

21. The training device according to claim 20, characterized in that The acquisition module includes: A reference score acquisition module is used to obtain the creator's creative ability reference score; A comparison module is used to obtain the creative ability type label of the creator based on the comparison result of the creative ability reference score and the preset threshold.

22. The training device according to claim 21, characterized in that Each of the preset feature types includes at least one of the following: a first feature type related to the original creative behavior of the creator; a second feature type related to the dissemination effect of the creator's work; The third characteristic type is related to the ecological contribution of the network platform where the creator's work is located.

23. The training device according to claim 21, characterized in that The training sample generation module includes: A normalization processing module, used to process the eigenvalues ​​of multiple creators in each eigendimension by a normalization method to obtain corresponding processed eigenvalues; The feature data generating module is used to construct the feature data of each creator based on each processed feature value.

24. The training device according to claim 21, characterized in that The creative ability evaluation model is implemented based on a deep neural network model; and adjusting the parameters of the creative ability evaluation model based on the loss function to complete the training includes: Based on the loss function of this round, the parameters of the creative ability evaluation model are derived and updated; Based on the updated parameters, the next round of loss function is determined until the loss function meets the conditions and the training ends.

25. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it realizes: The method for calculating creative ability information according to any one of claims 1 to 7; or the method for training a creative ability evaluation model according to claims 8 to 12.

26. An electronic device, characterized in that: include: processor; as well as a memory for storing executable instructions of the processor; The processor is configured to execute the following instructions: The method for calculating creative ability information according to any one of claims 1 to 7; or the method for training a creative ability evaluation model according to claims 8 to 12.

27. A network platform based on works, characterized by: include: The electronic device as claimed in claim 26.

Images