Intelligent scoring method and related device for content copywriting

Abstract

The present application is applicable to the field of data processing technology, and provides an intelligent scoring method and related devices for content copywriting, aiming to reduce the influence of industry fields on the scoring value of content copywriting, improve the efficiency of content copywriting scoring, and realize content copywriting scoring without distinguishing between industry fields. The method of the present application mainly includes: obtaining the target content copywriting click-through rates corresponding to the target content copywriting in X industry fields, and the field click-through rate of each industry field in the X industry fields, where X is a positive integer greater than 0; calculating the comprehensive click-through rate of the target content copywriting, the comprehensive click-through rate is equal to the sum of the product of the target content copywriting click-through rate corresponding to each industry field and the industry field click-through rate of the target content copywriting; determining the scoring value corresponding to the comprehensive click-through rate in the preset scoring standard as the scoring value of the target content copywriting, and the preset scoring standard records the scoring values ​​corresponding to the numerical values ​​of different click-through rates.

Description

Technical Field

[0001] The present application belongs to the field of data processing technology, and in particular, relates to an intelligent scoring method for content copywriting and related devices. Background Art

[0002] With the popularization and development of Internet technology, all industries in society have been connected to the Internet. Some content sharing platforms on the Internet (such as Weibo, Douyin, bilibili, etc.) will collect content copywriting (videos, pictures, texts, audio, etc.) from various industries. Content sharing platforms usually need to score content copywriting in different industry fields and obtain the score of each content copywriting, so as to configure the recommendation strategy of the content copywriting on the content sharing platform according to the score of each content copywriting.

[0003] However, the existing same content copy has a large difference in the scores received in different industry fields, which leads to the need for content sharing platforms to first classify content copy by industry field, and then configure the recommendation strategy of the content copy on the content sharing platform according to the score of the content copy in the subdivided industry field. As the industry fields become more and more subdivided, the workload of the content sharing platform to classify new content copy by industry field is increasing, making the rules for the content sharing platform to score content copy in different industry fields more and more complicated, which in turn leads to the unsatisfactory efficiency of scoring new content copy. That is, the existing technology lacks an intelligent scoring technology solution that can reduce the impact of industry fields on content copy scoring. Summary of the invention

[0004] The purpose of this application is to provide an intelligent scoring method and related devices for content copywriting, aiming to reduce the influence of industry fields on the scoring values ​​of content copywriting, improve the efficiency of scoring content copywriting, and achieve comprehensive scoring of content copywriting regardless of industry fields.

[0005] In a first aspect, the present application provides a method for intelligently scoring content copywriting, comprising:

[0006] Obtaining the target content copy click-through rates corresponding to the target content copy in X industry fields, and the field click-through rate of each industry field in the X industry fields, where X is a positive integer greater than 0;

[0007] Calculating the comprehensive click-through rate of the target content copy, the comprehensive click-through rate is equal to the sum of the product of the click-through rate of the target content copy corresponding to each industry field and the click-through rate of the industry field;

[0008] The scoring value corresponding to the comprehensive click-through rate is determined in the preset scoring standard as the scoring value of the target content copy, and the preset scoring standard records the scoring values ​​corresponding to the numerical values ​​of different click-through rates.

[0009] Optionally, before determining the score corresponding to the comprehensive click-through rate in the preset scoring criteria as the score of the target content copy, the method further includes:

[0010] Obtain historical copywriting in the X industry fields and historical copywriting click-through rates corresponding to the historical copywriting;

[0011] According to the click rate of the historical copy, the historical copy is graded into Y levels to obtain a scoring standard of Y levels, where Y is a positive integer greater than 0;

[0012] The historical documents are scored according to the scoring criteria to obtain a scoring value corresponding to the historical documents.

[0013] Optionally, determining the score corresponding to the comprehensive click-through rate in the preset scoring criteria as the score of the target content copy includes:

[0014] Determine whether the target content copy belongs to the content copy in the historical copy;

[0015] If the target content copy belongs to the content copy in the historical copy, the score corresponding to the historical copy is called as the score of the target copy;

[0016] If the target content copy does not belong to the content copy in the historical copy, the most similar historical copy that is most similar to the target content copy is obtained from the historical copy, and the scoring score corresponding to the most similar historical copy is called as the scoring score of the target content copy.

[0017] Optionally, after obtaining the historical copywriting in the X industry fields and the historical copywriting click-through rates corresponding to the historical copywriting, the method further includes:

[0018] Use the pre-trained model to vectorize the historical documents in the X industry fields to obtain vectorized information corresponding to the historical documents;

[0019] If the target content copy does not follow the content copy in the historical copy, the method further includes:

[0020] Using the pre-trained model to vectorize the target content copy, and obtain vectorized information corresponding to the target content copy;

[0021] Using the vectorized information corresponding to the target content copy, obtaining Z closest target historical copies from the vectorized information corresponding to the historical copies, where Z is a positive integer greater than 0;

[0022] Calculate the average click-through rate of the comprehensive click-through rates corresponding to the Z closest target historical texts;

[0023] The scoring value of the scoring standard corresponding to the average click rate is used as the scoring score of the target content copy.

[0024] Optionally, the pre-trained model is a BERT pre-trained model. After obtaining the vectorized information corresponding to the historical text, the method further includes:

[0025] The comprehensive click-through rate corresponding to each historical copy and the vectorized information corresponding to all historical copies are used for joint modeling to obtain Y levels of user preferred copies in the X industry fields and obtain Y levels of scoring standards. The user preferred copy is the content copy whose comprehensive click-through rate falls within a preset threshold range.

[0026] Optionally, the joint modeling using the comprehensive click-through rate corresponding to each historical copy and the vectorized information corresponding to all historical copies includes:

[0027] The BERT pre-trained model is trained for Y levels of classification using the historical text as a feature and the score value corresponding to the historical text as an objective function value to obtain a trained joint training model. The joint training model can recognize and classify the input text and then output a score value of the corresponding level.

[0028] Optionally, if the target content copy does not belong to the copy in the historical copy, the method further includes:

[0029] The target content copy is input into the joint training model to obtain a scoring score corresponding to the target content copy.

[0030] In a second aspect, the present application provides a content copywriting intelligent scoring system, comprising:

[0031] An acquisition unit, used to acquire the click-through rates of the target content copy corresponding to the target content copy in X industry fields, and the field click-through rate of each industry field in the X industry fields, where X is a positive integer greater than 0;

[0032] A calculation unit, used to calculate the comprehensive click-through rate of the target content copy, wherein the comprehensive click-through rate is equal to the sum of the product of the click-through rate of the target content copy corresponding to each industry field and the click-through rate of the industry field;

[0033] A determination unit is used to determine the score corresponding to the comprehensive click-through rate in a preset scoring standard as the score of the target content copy, and the preset scoring standard records the score values ​​corresponding to the numerical values ​​of different click-through rates.

[0034] Optionally, the system further includes:

[0035] The acquisition unit is further used to acquire historical copywriting in the X industry fields and historical copywriting click-through rates corresponding to the historical copywriting;

[0036] A grading unit, configured to grade the historical copy into Y levels according to the click rate of the historical copy, and obtain Y levels of scoring standards, where Y is a positive integer greater than 0;

[0037] The scoring unit is used to score the historical documents according to the scoring criteria to obtain the scoring values ​​corresponding to the historical documents.

[0038] Optionally, when the determining unit determines the score corresponding to the comprehensive click rate in the preset scoring standard as the score of the target content copy, it specifically includes:

[0039] Determine whether the target content copy belongs to the content copy in the historical copy;

[0040] If the target content copy belongs to the content copy in the historical copy, the score corresponding to the historical copy is called as the score of the target copy;

[0041] If the target content copy does not belong to the content copy in the historical copy, the most similar historical copy that is most similar to the target content copy is obtained from the historical copy, and the scoring score corresponding to the most similar historical copy is called as the scoring score of the target content copy.

[0042] Optionally, the system further includes:

[0043] An expression unit, using a pre-trained model to vectorize the historical documents in the X industry fields to obtain vectorized information corresponding to the historical documents;

[0044] The expression unit is further used to use the pre-trained model to vectorize the target content copy to obtain vectorized information corresponding to the target content copy;

[0045] The acquisition unit is further used to use the vectorized information corresponding to the target content copy to acquire Z closest target historical copies from the vectorized information corresponding to the historical copies, where Z is a positive integer greater than 0;

[0046] The calculation unit is further used to calculate the average click-through rate of the comprehensive click-through rates corresponding to the Z closest target historical texts;

[0047] As a unit, it is used to use the score value of the average click rate corresponding to the scoring standard as the score value of the target content copy.

[0048] Optionally, the pre-trained model is a BERT pre-trained model, and the system further includes:

[0049] The joint modeling unit is used to perform joint modeling using the comprehensive click-through rate corresponding to each historical copy and the vectorized information corresponding to all historical copies, to obtain Y levels of user preferred copies in the X industry fields, and to obtain Y levels of scoring standards, wherein the user preferred copies are content copies whose comprehensive click-through rate falls within a preset threshold range.

[0050] Optionally, when the joint modeling unit uses the comprehensive click-through rate corresponding to each historical copy and the vectorized information corresponding to all historical copies for joint modeling, it is specifically used to:

[0051] The BERT pre-trained model is trained for Y levels of classification using the historical text as a feature and the score value corresponding to the historical text as an objective function value to obtain a trained joint training model. The joint training model can recognize and classify the input text and then output a score value of the corresponding level.

[0052] Optionally, the system further includes:

[0053] The scoring unit is also used to input the target content copy into the joint training model to obtain a scoring score corresponding to the target content copy.

[0054] A third aspect of the present application provides a computer device, including:

[0055] Processor, memory, bus, input and output interface, wireless network interface;

[0056] The processor is connected to the memory, the input / output interface, and the wireless network interface via a bus;

[0057] The memory stores a program;

[0058] When the processor executes the program stored in the memory, the content copy intelligent scoring method described in the first aspect is implemented.

[0059] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer storage medium stores instructions, and when the instructions are executed on a computer, the computer executes the content copy intelligent scoring method as described in the first aspect above.

[0060] The fifth aspect of the present application provides a computer program product. When the computer program product is executed on a computer, the computer executes the content copy intelligent scoring method as described in the first aspect.

[0061] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0062] The intelligent scoring method for content copy of the present application obtains the click-through rates of the target content copy corresponding to the target content copy in X industry fields, and the field click-through rate of each industry field in the X industry fields, and then calculates the comprehensive click-through rate of the target content copy, wherein the comprehensive click-through rate is equal to the sum of the product of the click-through rate of the target content copy corresponding to each industry field and the industry field click-through rate, and the comprehensive click-through rate balances the influence of various industry fields on the scoring score of the target content copy, and then determines the scoring score corresponding to the comprehensive click-through rate in the preset scoring standard as the scoring score of the target content copy, wherein the preset scoring standard records the matching scoring scores corresponding to the numerical values ​​of different click-through rates, thereby realizing the scoring of content copy without distinguishing between industry fields, and effectively improving the efficiency of content copy scoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flowchart of an embodiment of the method for intelligently scoring the content of the application is shown below;

[0064] Figure 2 This is a flow chart of another embodiment of the method for intelligently scoring the content of the present application;

[0065] Figure 3 This is a flow chart of another embodiment of the method for intelligently scoring the content of the present application;

[0066] Figure 4 This is a schematic diagram of the structure of an embodiment of the intelligent scoring system for the content of this application;

[0067] Figure 5 This is a schematic diagram of the structure of an embodiment of a computer device of the present application. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0069] See also Figure 1 , an embodiment of the intelligent scoring method for the content of the present application includes:

[0070] 101. Obtain the click-through rates of the target content copy corresponding to the target content copy in X industry fields, and the field click-through rate of each industry field in the X industry fields, where X is a positive integer greater than 0.

[0071] The target content copy of this step can be one or more electronic content copy such as video, picture, text, audio, etc., and the target content copy is not specifically limited here; the industry field of this step can be social, financial, game, e-commerce and other industries that have been connected to the Internet, and the industry field is not specifically limited here; this step can obtain the target content copy click-through rate corresponding to the target content copy of the content sharing platform in each industry field through the Software Development Kit (SDK), and obtain the field click-through rate of each field in the corresponding industry field. Among them, the target content copy click-through rate is equal to the number of clicks on the target content copy by users divided by the number of exposures of the target content copy; the field click-through rate is equal to the number of clicks on the industry field by users divided by the number of exposures of the industry field.

[0072] For example: assuming that the target content copy is message a, X industry fields are equal to 3 industry fields (corresponding to industry field A, industry field B, and industry field C respectively), then this step obtains the message a click-through rate P(a|A) of message a in industry field A, the message a click-through rate P(a|B) of message a in industry field B, and the message a click-through rate P(a|C) of message a in industry field C through the software development toolkit configured on the content sharing platform; and obtains the field A click-through rate P(A) of industry field A, the field B click-through rate P(B) of industry field B, and the field C click-through rate P(C) of industry field C through the software development toolkit configured on the content sharing platform.

[0073] 102. Calculate the comprehensive click-through rate of the target content copy. The comprehensive click-through rate is equal to the product of the click-through rate of the target content copy corresponding to each industry field and the click-through rate of the industry field.

[0074] It is understandable that for the same target content copy, different users have different interests in the target content copy, resulting in different click probabilities for the same target content copy, especially for users classified in different industry fields. In order to balance the impact of various industry fields on the score value of the target content copy, this step can calculate the comprehensive click rate of the target content copy in step 101. For example, combined with the example of step 101, according to the Bayesian formula, it can be obtained:

[0075] P(a)=P(a|A)×P(A)+P(a|B)×P(B)+P(a|C)×P(C)

[0076] Among them, P(a) represents the comprehensive click rate of message a;

[0077] P(a|A) represents the click-through rate of message a in industry field A;

[0078] P(A) represents the click-through rate of field A in industry field A;

[0079] P(a|B) represents the click-through rate of message a in industry field B;

[0080] P(B) represents the click-through rate of domain B in industry domain B;

[0081] P(a|C) represents the click-through rate of message a in industry field C;

[0082] P(C) represents the click-through rate of field C in industry field C.

[0083] 103. Determine the score corresponding to the comprehensive click-through rate in the preset scoring standard as the score of the target content copy. The preset scoring standard records the matching score corresponding to the numerical values ​​of different click-through rates.

[0084] It is worth noting that the present embodiment needs to preset scoring standards for different CTR values ​​of content copywriting in advance. After obtaining the comprehensive CTR of the target content copywriting in step 102, this step can directly determine the scoring value corresponding to the comprehensive CTR in the preset scoring standard, and use the scoring value as the scoring value of the target content copywriting.

[0085] It can be seen that the content copy intelligent scoring method of the present application uses the comprehensive click-through rate to balance the impact of various industry fields on the scoring value of the target content copy, and then determines the scoring value corresponding to the comprehensive click-through rate in the preset scoring standard as the scoring value of the target content copy. The preset scoring standard records the matching scoring values ​​corresponding to the numerical values ​​of different click-through rates, thereby realizing the scoring of content copy without distinguishing between industry fields, and effectively improving the efficiency of content copy scoring.

[0086] See also Figure 2 Another embodiment of the method for intelligently scoring the content of the present application includes:

[0087] 201. Obtain historical copywriting in X industry fields and the historical copywriting click-through rates corresponding to the historical copywriting.

[0088] Specifically, in order to formulate a more objective preset scoring standard, this embodiment needs to refer to some historical data to formulate the preset scoring standard. This step obtains the historical copywriting of the content sharing platform and the historical copywriting click-through rate corresponding to the historical copywriting through the Software Development Kit (SDK). For example, the data collected and reported by the software development kit of the content sharing platform are summarized by Aurora Big Data, and these data include historical copywriting in X industry fields of the content sharing platform and the historical copywriting click-through rate corresponding to the historical copywriting.

[0089] 202. Use the pre-trained model to vectorize the historical documents in X industry fields to obtain the vectorized information corresponding to the historical documents.

[0090] Specifically, the pre-trained model of this step can be a BERT pre-trained model (Bidirectional Encoder Representation from Transformers). The BERT pre-trained model is a pre-trained model proposed by Google AI Research Institute. As a pre-trained model that has been used in the prior art, the BERT pre-trained model will not be elaborated on here. In this step, the historical documents in various industry fields can be vectorized through the BERT pre-trained model to obtain the vectorized information corresponding to each historical document, so that the pre-trained model knows the information expressed by each historical document itself. It can be understood that this step is only explained by taking the BERT pre-trained model as an example. In practical applications, other pre-trained models can be used according to actual conditions, such as the ELMo pre-trained model, etc. The pre-trained model is not further limited here.

[0091] 203. The historical copywriting is graded into Y levels according to the click-through rate of the historical copywriting to obtain Y levels of scoring standards, where Y is a positive integer greater than 0.

[0092] Specifically, since the numerical value of the click-through rate is often a decimal between 0 and 1, it is generally not reflected as a rating value. Therefore, in this step, the click-through rate is divided into Y levels of rating grades according to certain rules, so that the click-through rate can be mapped to Y levels of rating grades according to the rules, generalizing the rating results of the copywriting to obtain Y levels of rating standards.

[0093] For example: Assuming that the Y-level scoring system is divided into 10 levels, then 10 levels of scoring standards can be determined. At this time, the absolute division rule can be used to divide the historical copy click rate into 10 levels of scoring standards in this step, which is formulated as follows: 0 historical copy click rate corresponds to a historical copy score of 0, 0 to 0.1 (including 0.1) corresponds to a historical copy score of 1, 0.1 to 0.2 (including 0.2) corresponds to a historical copy score of 2, 0.2 to 0.3 (including 0.3) corresponds to a historical copy score of 3, 0.3 to 0.4 (including 0.4) corresponds to a historical copy score of 4, and 0.4 to 0.6 (including 0.6) corresponds to a historical copy score of 5. The historical copywriting score corresponding to the historical copywriting click-through rate between 0.5 and 0.5 (including 0.5) is 5, the historical copywriting score corresponding to the historical copywriting click-through rate between 0.5 and 0.6 (including 0.6) is 6, the historical copywriting score corresponding to the historical copywriting click-through rate between 0.6 and 0.7 (including 0.7) is 7, the historical copywriting score corresponding to the historical copywriting click-through rate between 0.7 and 0.8 (including 0.8) is 8, the historical copywriting score corresponding to the historical copywriting click-through rate between 0.8 and 0.9 (including 0.9) is 9, and the historical copywriting score corresponding to the historical copywriting click-through rate between 0.9 and 1 (including 1) is 10.It is worth noting that the division rule of the historical copy click rate corresponding to the 10 levels of scoring standards in this step can also adopt a relative division rule. First, determine the numerical range of the historical copy click rate (assuming it is between 0 and 0.5), and then divide the numerical range of the historical copy click rate into 10 equal parts, that is, formulate it as follows: 0 historical copy click rate corresponds to a historical copy scoring score of 0, 0 to 0.05 (including 0.05) historical copy click rate corresponds to a historical copy scoring score of 1, 0 to 0.05 (including 0.05) historical copy click rate corresponds to a historical copy scoring score of 1, 0.05 to 0.1 (including 0.1) historical copy click rate corresponds to a historical copy scoring score of 2, 0.1 to 0.1 (including 0.15) historical copy click rate corresponds to a historical copy scoring score of 1. The copywriting score is 3, the historical copywriting click-through rate between 0.15 and 0.2 (including 0.2) corresponds to a historical copywriting score of 4, the historical copywriting click-through rate between 0.2 and 0.25 (including 0.25) corresponds to a historical copywriting score of 5, the historical copywriting click-through rate between 0.25 and 0.3 (including 0.3) corresponds to a historical copywriting score of 6, the historical copywriting click-through rate between 0.3 and 0.35 (including 0.35) corresponds to a historical copywriting score of 7, the historical copywriting click-through rate between 0.35 and 0.4 (including 0.4) corresponds to a historical copywriting score of 8, the historical copywriting click-through rate between 0.4 and 0.45 (including 0.45) corresponds to a historical copywriting score of 9, and the historical copywriting click-through rate between 0.45 and 0.5 (including 0.5) corresponds to a historical copywriting score of 10.

[0094] 204. Score the historical copy according to the scoring criteria to obtain the corresponding scoring value of the historical copy.

[0095] The historical copywriting in step 201 is scored using the scoring criteria in step 203. In order to obtain a more objective scoring value, the historical copywriting can be first scored as follows: Figure 1 In the embodiment, the comprehensive click rate of step 102 is calculated, and then the historical copy is scored according to the scoring standard to obtain the scoring value corresponding to the historical copy.

[0096] 205. Obtain the target content copy click-through rates corresponding to the target content copy in X industry fields, and the field click-through rate of each field in the X industry fields, where X is a positive integer greater than 0.

[0097] This step is performed in the same way as above. Figure 1 Step 101 in the embodiment is similar, and the repeated parts are not repeated here. It should be noted that the target content copy in this step mainly refers to the content copy newly added by the content sharing platform that has not been rated.

[0098] 206. Determine whether the target content copy belongs to the content copy in the historical copy. If the target content copy belongs to the content copy in the historical copy, execute step 207; if the target content copy does not belong to the content copy in the historical copy, execute step 208 or step 209.

[0099] In order to improve the scoring efficiency of the target content copy, this step first determines whether the target content copy in step 205 belongs to the content copy in the historical copy in step 201. If the target content copy belongs to the content copy in the historical copy, it indicates that the target content copy has been scored and graded in step 204, and the scoring score of the historical copy corresponding to the target copy can be directly called as its own scoring score; if the target content copy does not belong to the content copy in the historical copy, it indicates that the target content copy has not been scored and graded in step 204, and the above-mentioned scoring can be performed at this time. Figure 1 In the embodiment, the different scoring strategies of step 103 are specifically shown in step 208 or step 209. In this step, a similarity algorithm can be used to determine whether the target content copy belongs to a historical copy.

[0100] 207. Call the scoring value corresponding to the historical copy as the scoring value of the target copy.

[0101] If step 206 determines that the target content copy belongs to the content copy in the historical copy, it means that the target content copy has been scored and graded in step 204. This step can directly call the scoring value corresponding to the historical copy as the scoring value of the target copy.

[0102] 208. Obtain the most similar historical copy that is most similar to the target content copy in the historical copy, and call the score corresponding to the most similar historical copy as the score of the target content copy.

[0103] In one embodiment, if step 206 determines that the target content copy does not belong to the content copy in the historical copy, it means that the target content copy has not been scored and graded in step 204. This step can also obtain the most similar historical copy that is most similar to the target content copy in the historical copy through a similarity algorithm, and then call the scoring score corresponding to the most similar historical copy as the scoring score of the target content copy.

[0104] 209. Use the pre-trained model to vectorize the target content copy to obtain vectorized information corresponding to the target content copy.

[0105] In another embodiment, if step 206 determines that the target content copy does not belong to the content copy in the historical copy, it means that the target content copy has not been scored and graded in step 204. This step can also use a pre-trained model (such as a BERT pre-trained model) to vectorize the target content copy to obtain the vectorized information corresponding to the target content copy.

[0106] 210. Use the vectorized information corresponding to the target content copy to obtain Z closest target historical copies from the vectorized information corresponding to the historical copies, where Z is a positive integer greater than 0.

[0107] After the vectorized information corresponding to the target content copy is obtained in step 209, the vectorized information corresponding to the target content copy can be used to obtain the Z closest target historical copies from the vectorized information corresponding to the historical copies. For example, the Faiss framework is used to obtain the 10 historical copies that are closest to the vectorized information corresponding to the target content copy. Since the target content copy is expressed by the vectorization of the pre-trained model before obtaining the closest target historical copy, the accuracy is higher.

[0108] 211. Calculate the average click-through rate of the comprehensive click-through rates corresponding to the Z closest target historical texts.

[0109] For example, referring to the example in step 210, the Z closest target historical copies are equal to the 10 closest target historical copies. Since the historical copies have all been scored in step 204 and obtained corresponding scoring scores, it can be known that the comprehensive click-through rates corresponding to the historical copies have also been calculated and recorded. As part of the historical copies, the target historical copies must also record the comprehensive click-through rates corresponding to the target historical copies. At this time, this step calls and calculates the average click-through rate of the comprehensive click-through rates corresponding to the 10 closest target historical copies.

[0110] 212. The score of the scoring standard corresponding to the average click-through rate is used as the score of the target content copy.

[0111] In order to more objectively reflect the comprehensive click-through rate of the target content copy, this step can score the click-through rate average value obtained in step 211 according to the scoring standard of step 203 to obtain a corresponding scoring score, and use the scoring score of the click-through rate average value corresponding to the scoring standard as the scoring score of the target content copy.

[0112] See also Figure 3 , in the above Figure 2 After step 204, another embodiment of the method for intelligently scoring content of the present application includes:

[0113] 301. Use the historical text as a feature and the score value corresponding to the historical text as the objective function value to perform Y levels of classification training on the Bert pre-training model to obtain a trained joint training model.

[0114] For example, in this step, the pre-trained model can also be trained for 10 levels of classification, so as to obtain a trained joint training model that can achieve 10 classifications, learn 10 levels of user preference copywriting in various industry fields, and obtain 10 levels of scoring standards. The user preference copywriting is the content copywriting whose comprehensive click-through rate falls within the preset threshold range. Specifically, the Bert pre-trained model can be trained for 10 levels of classification using historical copywriting as features and the scoring values ​​corresponding to the historical copywriting as the objective function value.

[0115] 302. Input the target content copy into the joint training model to obtain a scoring value corresponding to the target content copy.

[0116] It is worth noting that after training, the joint training model of step 301 can recognize and classify the input content copy and quickly output the corresponding level of scoring score. This step only requires the target content copy to be input into the joint training model to obtain the scoring score corresponding to the target content copy. It is convenient and fast, and the scoring result can generally be obtained in milliseconds to seconds.

[0117] The above embodiment describes the intelligent scoring method of the content copy of the present application. The following describes the intelligent scoring system of the content copy of the present application. Figure 4 , a content copywriting intelligent scoring system, comprising:

[0118] The acquisition unit 401 is used to acquire the click-through rates of the target content texts respectively corresponding to the target content texts in X industry fields, and the field click-through rate of each industry field in the X industry fields, where X is a positive integer greater than 0;

[0119] A calculation unit 402 is used to calculate the comprehensive click-through rate of the target content copy, where the comprehensive click-through rate is equal to the sum of the product of the click-through rate of the target content copy corresponding to each industry field and the click-through rate of the industry field;

[0120] The determination unit 403 is used to determine the score corresponding to the comprehensive click-through rate in the preset scoring standard as the score of the target content copy, and the preset scoring standard records the matching score corresponding to the numerical values ​​of different click-through rates.

[0121] Optionally, the system further includes:

[0122] The acquisition unit 401 is further used to acquire historical copywriting in the X industry fields and historical copywriting click-through rates corresponding to the historical copywriting;

[0123] A grading unit 404 is used to grade the historical copy into Y levels according to the click rate of the historical copy, and obtain Y levels of scoring standards, where Y is a positive integer greater than 0;

[0124] The scoring unit 405 is used to score the historical documents according to the scoring criteria to obtain the scoring value corresponding to the historical documents.

[0125] Optionally, when the determining unit 403 determines the score corresponding to the comprehensive click rate in the preset scoring standard as the score of the target content copy, it specifically includes:

[0126] Determine whether the target content copy belongs to the content copy in the historical copy;

[0127] If the target content copy belongs to the content copy in the historical copy, the score corresponding to the historical copy is called as the score of the target copy;

[0128] If the target content copy does not belong to the content copy in the historical copy, the most similar historical copy that is most similar to the target content copy is obtained from the historical copy, and the scoring score corresponding to the most similar historical copy is called as the scoring score of the target content copy.

[0129] Optionally, the system further includes:

[0130] The expression unit 406 uses a pre-trained model to vectorize the historical documents in the X industry fields to obtain vectorized information corresponding to the historical documents;

[0131] The expression unit 406 is further used to use the pre-trained model to vectorize the target content text to obtain vectorized information corresponding to the target content text;

[0132] The acquisition unit 401 is further configured to use the vectorized information corresponding to the target content copy to acquire Z closest target historical copies from the vectorized information corresponding to the historical copies, where Z is a positive integer greater than 0;

[0133] The calculation unit 402 is further used to calculate the average click-through rate of the comprehensive click-through rates corresponding to the Z closest target historical texts;

[0134] As unit 407, it is used to use the score value of the average click rate corresponding to the scoring standard as the score value of the target content copy.

[0135] Optionally, the pre-trained model is a BERT pre-trained model, and the system further includes:

[0136] The joint modeling unit 408 is used to perform joint modeling using the comprehensive click-through rate corresponding to each historical copy and the vectorized information corresponding to all historical copies, to obtain Y levels of user preferred copies in the X industry fields, and to obtain Y levels of scoring standards, wherein the user preferred copy is a content copy whose comprehensive click-through rate falls within a preset threshold range.

[0137] Optionally, when the joint modeling unit 408 uses the comprehensive click-through rate corresponding to each historical copy and the vectorized information corresponding to all historical copies for joint modeling, it is specifically used to:

[0138] The BERT pre-trained model is trained for Y levels of classification using the historical text as a feature and the score value corresponding to the historical text as an objective function value to obtain a trained joint training model. The joint training model can recognize and classify the input text and then output a score value of the corresponding level.

[0139] Optionally, the system further includes:

[0140] The scoring unit 405 is further configured to input the target content text into the joint training model to obtain a scoring value corresponding to the target content text.

[0141] The operations performed by the content copywriting intelligent scoring system in the present embodiment of the application are similar to those described above. Figure 1 , Figure 2 as well as Figure 3 The operations performed in the embodiment are similar and will not be described in detail here.

[0142] The following describes the computer device of the embodiment of the present application. Figure 5 , an embodiment of the computer device in the embodiment of the present application includes:

[0143] The computer device 500 may include one or more processors (central processing units, CPU) 501 and a memory 502, in which one or more application programs or data are stored. The memory 502 is a volatile storage or a persistent storage. The program stored in the memory 502 may include one or more modules, and each module may include a series of instruction operations in the computer device. Furthermore, the processor 501 may be configured to communicate with the memory 502 to execute a series of instruction operations in the memory 502 on the computer device 500. The computer device 500 may also include one or more wireless network interfaces 503, one or more input and output interfaces 504, and / or one or more operating systems, such as Windows Server, Mac OS, Unix, Linux, FreeBSD, etc. The processor 501 may execute the aforementioned Figures 1 to 3 The operations performed in the embodiment will not be described in detail here.

[0144] In several embodiments provided in the embodiments of the present application, those skilled in the art should understand that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the unit is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0145] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.

[0146] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions or improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for intelligently scoring content copywriting, characterized in that: include: Obtaining the target content copy click-through rates corresponding to the target content copy in X industry fields, and the field click-through rate of each industry field in the X industry fields, where X is a positive integer greater than 0; Calculating the comprehensive click-through rate of the target content copy, the comprehensive click-through rate is equal to the sum of the product of the click-through rate of the target content copy corresponding to each industry field and the click-through rate of the industry field; Determine the score corresponding to the comprehensive click-through rate in the preset scoring standard as the score of the target content copy, wherein the preset scoring standard records the score corresponding to the numerical values ​​of different click-through rates; Before determining the score corresponding to the comprehensive click-through rate in the preset scoring criteria as the score of the target content copy, the method further includes: Obtain historical copywriting in the X industry fields and historical copywriting click-through rates corresponding to the historical copywriting; According to the click rate of the historical copy, the historical copy is graded into Y levels to obtain a scoring standard of Y levels, where Y is a positive integer greater than 0; Scoring the historical copy according to the scoring criteria to obtain a scoring value corresponding to the historical copy; Determining the score corresponding to the comprehensive click rate in the preset scoring criteria as the score of the target content copy includes: Determine whether the target content copy belongs to the content copy in the historical copy; If the target content copy belongs to the content copy in the historical copy, the score corresponding to the historical copy is called as the score of the target copy; If the target content copy does not belong to the content copy in the historical copy, then obtaining the most similar historical copy that is most similar to the target content copy in the historical copy, and calling the score corresponding to the most similar historical copy as the score of the target content copy; After obtaining the historical copywriting in the X industry fields and the historical copywriting click-through rates corresponding to the historical copywriting, the method further includes: Use the pre-trained model to vectorize the historical documents in the X industry fields to obtain vectorized information corresponding to the historical documents; If the target content copy does not follow the content copy in the historical copy, the method further includes: Using the pre-trained model to vectorize the target content copy, and obtain vectorized information corresponding to the target content copy; Using the vectorized information corresponding to the target content copy, obtaining Z closest target historical copies from the vectorized information corresponding to the historical copies, where Z is a positive integer greater than 0; Calculate the average click-through rate of the comprehensive click-through rates corresponding to the Z closest target historical texts; The scoring value of the scoring standard corresponding to the average click rate is used as the scoring score of the target content copy.

2. The content copywriting intelligent scoring method according to claim 1 is characterized in that: The pre-trained model is a BERT pre-trained model. After obtaining the vectorized information corresponding to the historical text, the method further includes: The comprehensive click-through rate corresponding to each historical copy and the vectorized information corresponding to all historical copies are used for joint modeling to obtain Y levels of user preferred copies in the X industry fields and obtain Y levels of scoring standards. The user preferred copy is the content copy whose comprehensive click-through rate falls within a preset threshold range.

3. The content copywriting intelligent scoring method according to claim 2 is characterized in that: The joint modeling using the comprehensive click rate corresponding to each historical copy and the vectorized information corresponding to all historical copies includes: The BERT pre-trained model is trained for Y levels of classification using the historical text as a feature and the score value corresponding to the historical text as an objective function value to obtain a trained joint training model. The joint training model can recognize and classify the input text and then output a score value of the corresponding level.

4. The method for intelligently scoring content copywriting according to claim 3 is characterized in that: If the target content copy does not belong to the copy in the historical copy, the method further includes: The target content copy is input into the joint training model to obtain a scoring score corresponding to the target content copy.

5. A content copywriting intelligent scoring system, characterized in that: include: An acquisition unit, used to acquire the click-through rates of the target content copy corresponding to the target content copy in X industry fields, and the field click-through rate of each industry field in the X industry fields, where X is a positive integer greater than 0; A calculation unit, used to calculate the comprehensive click-through rate of the target content copy, wherein the comprehensive click-through rate is equal to the sum of the product of the click-through rate of the target content copy corresponding to each industry field and the click-through rate of the industry field; A determination unit, configured to determine a score corresponding to the comprehensive click-through rate in a preset scoring standard as a score of the target content copy, wherein the preset scoring standard records the score corresponding to the numerical values ​​of different click-through rates; The system further comprises: The acquisition unit is further used to acquire historical copywriting in the X industry fields and historical copywriting click-through rates corresponding to the historical copywriting; A grading unit, configured to grade the historical copy into Y levels according to the click rate of the historical copy, and obtain Y levels of scoring standards, where Y is a positive integer greater than 0; A scoring unit, used to score the historical copy according to the scoring criteria to obtain a scoring value corresponding to the historical copy; When the determining unit determines the score corresponding to the comprehensive click rate in the preset scoring standard as the score of the target content copy, it specifically includes: Determine whether the target content copy belongs to the content copy in the historical copy; If the target content copy belongs to the content copy in the historical copy, the score corresponding to the historical copy is called as the score of the target copy; If the target content copy does not belong to the content copy in the historical copy, then obtaining the most similar historical copy that is most similar to the target content copy in the historical copy, and calling the score corresponding to the most similar historical copy as the score of the target content copy; The system further comprises: An expression unit, using a pre-trained model to vectorize the historical documents in the X industry fields to obtain vectorized information corresponding to the historical documents; The expression unit is further used to use the pre-trained model to vectorize the target content copy to obtain vectorized information corresponding to the target content copy; The acquisition unit is further used to use the vectorized information corresponding to the target content copy to acquire Z closest target historical copies from the vectorized information corresponding to the historical copies, where Z is a positive integer greater than 0; The calculation unit is further used to calculate the average click-through rate of the comprehensive click-through rates corresponding to the Z closest target historical texts; As a unit, it is used to use the score value of the average click rate corresponding to the scoring standard as the score value of the target content copy.

6. A computer device, characterized in that: include: Processor, memory, bus, input and output interface, wireless network interface; The processor is connected to the memory, the input / output interface, and the wireless network interface via a bus; The memory stores a program; When the processor executes the program stored in the memory, the content copy intelligent scoring method described in any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that: The computer storage medium stores instructions, and when the instructions are executed on a computer, the computer executes the content copy intelligent scoring method described in any one of claims 1 to 4.

Images