Content quality assessment method, model training method and related device

Abstract

In the content quality assessment method, model training method and related devices provided by the present application; the assessment device obtains historical statistical information of media resources and inherent attribute information of media resources, and then uses a pre-trained scoring model to reasonably assess the content quality of the media resources by comprehensively considering the above factors, thereby obtaining a more accurate and reasonable target score for assessing the user's preference for the media resources.

Description

Technical Field

[0001] The present application relates to the field of data processing, and specifically, to a content quality assessment method, a model training method, and related devices. Background Art

[0002] With the rapid development of the Internet and mobile Internet, the number of media resources on Internet platforms has increased dramatically, and there is a large amount of homogeneous content (same title or similar introduction). How to quantify and analyze these media resources and determine their content quality is a very critical and urgent task.

[0003] Currently, when conducting quantitative analysis on media resources, people mostly rely on browsing statistics of media resources. However, research has found that browsing statistics cannot objectively reflect the content quality of media resources and there is a certain lag. Summary of the invention

[0004] In order to overcome at least one deficiency in the prior art, the present application provides a content quality assessment method and a related device, including:

[0005] In a first aspect, the present application provides a content quality assessment method, which is applied to an assessment device, and the method includes:

[0006] Acquire a first feature and a second feature of a media resource, wherein the first feature includes browsing statistics of the media resource, and the second feature includes inherent attribute information of the media resource;

[0007] The first feature and the second feature are input into a pre-trained scoring model to obtain a target score for the media resource, wherein the target score is used to evaluate the user's preference for the media resource.

[0008] In a second aspect, the present application provides a content quality assessment device, which is applied to an assessment device, and the content quality assessment device includes:

[0009] A feature module, used to obtain a first feature and a second feature of a media resource, wherein the first feature includes browsing statistics of the media resource, and the second feature includes inherent attribute information of the media resource;

[0010] A scoring module is used to input the first feature and the second feature into a pre-trained scoring model to obtain a target score for the media resource, wherein the target score is used to evaluate the user's preference for the media resource.

[0011] In a third aspect, the present application provides an evaluation device, which includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the content quality evaluation method is implemented.

[0012] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the content quality assessment method is implemented.

[0013] Compared with the prior art, this application has the following beneficial effects:

[0014] In the content quality assessment method, model training method and related devices provided by the present application; the assessment device obtains historical statistical information of media resources and inherent attribute information of media resources, and then uses a pre-trained scoring model to reasonably assess the content quality of the media resources by comprehensively considering the above factors, thereby obtaining a more accurate and reasonable target score for assessing the user's preference for the media resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A structural block diagram of an evaluation device provided in an embodiment of the present application;

[0017] Figure 2 A flowchart of a content quality assessment method provided in an embodiment of the present application;

[0018] Figure 3 A schematic diagram of the model training method flow provided in the embodiment of the present application;

[0019] Figure 4 A structural block diagram of a content quality assessment device provided in an embodiment of the present application;

[0020] Figure 5 A schematic diagram of the principles of the model training method provided in an embodiment of the present application.

[0021] Icon: 120 - memory; 130 - processor; 140 - communication unit; 201 - feature module; 202 - scoring module. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0024] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0025] Currently, when conducting quantitative analysis on media resources, people mostly rely on browsing statistics of media resources. However, research has found that browsing statistics cannot objectively reflect the content quality of media resources and there is a certain lag.

[0026] For example, browsing statistics may include the historical playback volume, subscription number, click volume, retention rate, etc. of a media resource. Although the above browsing statistics can reflect the quality of a media resource to a certain extent, for example, when the historical playback volume, subscription number, click volume, retention rate, etc. are large, it means that the corresponding media resource has high content quality and is loved by the audience; however, the above browsing statistics can be improved in an unregulated manner, for example, by using a water army to improve the browsing statistics of a media resource.

[0027] In addition, for some newly created media resources, it is difficult to obtain enough browsing statistics in a short period of time, that is, the browsing statistics of media resources in a short period of time do not have objective reference significance. Therefore, relying solely on the browsing statistics of media resources cannot objectively reflect the content quality of media resources.

[0028] In view of this, this embodiment provides a content quality evaluation method applied to an evaluation device. In this method, the evaluation device obtains historical statistical information of media resources and inherent attribute information of media resources, and then uses a pre-trained scoring model to reasonably evaluate the content quality of the media resources by combining the above factors, thereby obtaining a more accurate and reasonable target score for evaluating the user's preference for the media resource.

[0029] Among them, the evaluation device can be a server, and the type of the server can be, but is not limited to, a Web server, an FTP (File Transfer Protocol) server, a data processing server, etc. In addition, the server can be a single server or a server group. The server group can be centralized or distributed (for example, the server can be a distributed system). In some embodiments, the server can be local or remote relative to the user terminal. In some embodiments, the server can be implemented on a cloud platform; as an example only, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an inter-cloud, a multi-cloud, etc., or any combination thereof. In some embodiments, the server can be implemented on an electronic device having one or more components.

[0030] like Figure 1 As shown in the structural block diagram of the evaluation device provided in this embodiment, the evaluation device may include a memory 120, a processor 130, and a communication unit 140. The memory 120, the processor 130, and the communication unit 140 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0031] The memory 120 may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory 120 is used to store a program, and the processor 130 executes the program after receiving an execution instruction.

[0032] The communication unit 140 is used to send and receive data through a network. The network may include a wired network, a wireless network, a fiber optic network, a telecommunication network, an intranet, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a wide area network (WAN), a public switched telephone network (PSTN), a Bluetooth network, a ZigBee network, or a near field communication (NFC) network, or any combination thereof. In some embodiments, the network may include one or more network access points. For example, the network may include a wired or wireless network access point, such as a base station and / or a network switching node, through which one or more components of the service request processing system may be connected to the network to exchange data and / or information.

[0033] The processor 130 may be an integrated circuit chip having a signal processing capability, and the processor may include one or more processing cores (e.g., a single-core processor or a multi-core processor). By way of example only, the processor may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), an application-specific instruction set processor (ASIP), a graphics processing unit (GPU), a physical processing unit (PPU), a digital signal processor (DSP), a field programmable gate array (FPGA), a programmable logic device (PLD), a controller, a microcontroller unit, a reduced instruction set computer (RISC), or a microprocessor, or any combination thereof.

[0034] Based on the above introduction, Figure 2The flowchart shown in the figure is used to describe the method provided in this embodiment in detail. It should be understood that the operations in the flowchart can be implemented in a non-sequential manner, and steps without logical contextual relationships can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart, or remove one or more operations from the flowchart under the guidance of the content of this application. Figure 2 As shown, the content quality assessment method includes:

[0035] S101A, obtaining a first feature and a second feature of a media resource.

[0036] The first feature includes browsing statistics of the media resource, and the second feature includes inherent attribute information of the media resource; and the inherent attribute information may include attribute information of the media resource and attribute information of the creator corresponding to the media resource.

[0037] It is worth noting that the media resources in this embodiment can be, but are not limited to, videos, pictures, audios, etc. In some implementations, the video can be a short video, a movie, a documentary, a TV series, a news video, etc.; the picture can be a landscape photo, a person photo, an art photo, etc.; the audio can be a music single, an audio novel, an audio story, etc.

[0038] For different types of media resources, those skilled in the art can select corresponding browsing statistics and inherent attribute information as needed. Taking audio novels as an example, the browsing statistics of the novels may include the weekly playback volume, daily playback volume, number of subscriptions, user click volume and other information of the novels.

[0039] The attribute information of the media resources corresponding to the novel may include album type, copyright type, payment type, number of sounds, audio quality, audio duration, etc.

[0040] The album type indicates the category of the novel, such as urban, romance, horror, etc. The copyright type indicates the copyright information of the novel content. The payment type indicates whether the audio novel is free. Whether the audio novel is paid can reflect the content quality of the audio novel to a certain extent. The number of sounds indicates the number of sound sources in the audio novel, including the number of anchors, the number of background music, etc. The audio quality indicates whether the audio novel is clear, complete, the anchor's accent, the audio bit rate, etc.

[0041] The attribute information of the corresponding creator of the novel may include the number of fans of the anchor, the anchor level, and the anchor value. Among them, the anchor value is the reference value provided by the platform when conducting a quantitative evaluation of the anchor according to the anchor's historical data.

[0042] S102A, input the first feature and the second feature into a pre-trained scoring model to obtain a target score for the media resource.

[0043] Among them, the target score is used to evaluate the user's preference for the media resource.

[0044] It should be understood that, since the feature sources of media resources include browsing statistics and inherent attribute information of media content, not only are the feature categories diverse, but the number of features is also relatively large, and there are interactive relationships between different features.

[0045] In the related art, a preset mathematical formula can be used to calculate the target score of a media resource, but it requires artificially constructed features, and due to the high-dimensional interactions between features, it is very difficult to solve the coefficients of these high-dimensional items. In this embodiment, a scoring model based on deep learning is used to learn the low-dimensional and high-dimensional feature interaction correlation information between features, so as to obtain a more accurate and reasonable target score.

[0046] The study also found that the deep learning-based scoring model is a classification model, that is, the scoring model outputs the degree of match between media resources and different quality scores based on the first and second input features. This means that when the degree of match between multiple media resources and the same quality score is the greatest, the multiple media resources have the same target score.

[0047] Exemplarily, it is assumed that the scoring model can output 10 classification probabilities for media resources, which respectively represent the degree of matching with 10 quality scores; wherein the 10 quality scores are 0, 1, 2, 3, 4, 5, 6, 7, 8, and 9 in descending order. However, it should be understood that the content quality of the media content is positively correlated with the size of the quality score. Therefore, based on this embodiment, those skilled in the art can adaptively adjust the number and numerical value of the quality score as needed, without making any creative contribution.

[0048] Assume that there are two media resources, denoted as resource A and resource B. When the evaluation device inputs the first feature and the second feature of resource A into the scoring model, the output 10 classification probabilities are shown in Table 1:

[0049] Table 1

[0050] 0 1 2 3 4 5 6 7 8 9 0.002 0.004 0.001 0.003 0.06 0.52 0.39 0.012 0.005 0.003

[0051] When the evaluation device inputs the first feature and the second feature of resource B into the scoring model, the output 10 classification probabilities are shown in Table 2:

[0052] Table 2

[0053] 0 1 2 3 4 5 6 7 8 9 0.001 0.003 0.004 0.002 0.01 0.73 0.24 0.002 0.003 0.005

[0054] It can be seen that both resource A and resource B are most compatible with the quality score of 5 points. However, this embodiment requires a more granular evaluation of media resources with the same quality score. Therefore, the classification problem needs to be converted into a regression problem.

[0055] As an optional implementation, step S102 may convert the classification problem into a regression problem by the following implementation, thereby obtaining a target score for the media resource:

[0056] S102-1, input the first feature and the second feature into a pre-trained scoring model to obtain a plurality of classification probabilities output by the scoring model.

[0057] The multiple classification probabilities respectively represent the matching degree between the media resources and different quality scores.

[0058] S102-1, obtaining a target score of a media resource according to a plurality of classification probabilities and quality scores corresponding to the plurality of classification probabilities.

[0059] Among them, the basic score accounts for the highest proportion in the target score, and the basic score represents the quality score corresponding to the maximum classification probability among multiple classification probabilities.

[0060] Therefore, through the above implementation, the evaluation device uses the quality score corresponding to the maximum classification probability among all classification probabilities of the media resource as the basic score, and adjusts the basic score to obtain the target score, so that the basic score accounts for the highest proportion of the target score. Since the basic score accounts for the highest proportion of the target score, in the process of converting the classification problem into a regression problem,

[0061] Base scoring dominates goal scoring.

[0062] The adjustment range depends on other classification probabilities other than the maximum classification probability and the corresponding quality scores. Therefore, in an optional implementation, step S102-1 may adjust the basic score by the following implementation:

[0063] S102-1-1, based on multiple classification probabilities, taking the quality score corresponding to the maximum classification probability as the basic score.

[0064] S102-1-2, obtaining an expected score according to the remaining classification probabilities and the quality scores corresponding to the remaining classification probabilities.

[0065] Among them, the remaining classification probability does not include the maximum classification probability.

[0066] S102-1-3, taking the sum of the basic score and the expected score as the target score.

[0067] Therefore, in the above implementation, the scoring device calculates the expected score according to the remaining classification probabilities and the quality scores corresponding to the remaining classification probabilities, so as to adjust the basic score.

[0068] The purpose of this embodiment is to evaluate media resources with the same quality score at a smaller granularity, which means that the final target score needs to be improved to a certain extent based on the basic score, but the adjustment range is related to the distribution of the remaining classification probabilities. Specifically, the greater the probability of matching with a larger quality score, the greater the adjustment range.

[0069] For example, continue to take resource A and resource B in the above example as an example. It can be seen from Table 1 and Table 2 in the above embodiment that resource A and resource B have the highest matching degree with the quality score of 5 points respectively. Therefore, when adjusting the basic scores of resource A and resource B, the classification probabilities corresponding to 6 points, 7 points, 8 points, and 9 points are regarded as the remaining classification probabilities. The larger the sum of the remaining classification probabilities, the greater the adjustment of the basic scores.

[0070] Among them, for resource A, the sum of the remaining classification probabilities corresponding to 6 points, 7 points, 8 points, and 9 points is 0.39+0.012+0.005+0.003=0.41.

[0071] For resource B, the sum of the remaining classification probabilities corresponding to 6 points, 7 points, 8 points, and 9 points is 0.24+0.002+0.003+0.005=0.25.

[0072] Since the sum of the remaining classification probabilities of resource A is greater than the sum of the classification probabilities of resource B, it means that the content quality of resource A is better than the content quality of resource B to a certain extent. Therefore, it is necessary to provide a calculation method so that the target score of resource A is greater than the target score of resource B.

[0073] To achieve this purpose, this embodiment uses different methods to calculate the expected score of the media resource according to different quality scores corresponding to the maximum classification probability. The specific optional implementation methods are as follows:

[0074] If the quality score corresponding to the maximum classification probability is the maximum quality score, the expected score E is calculated by the following expression:

[0075] E=p max *(1-f(S))

[0076]

[0077] If the quality score corresponding to the maximum classification probability is not the maximum quality score, the expected score E is calculated by the following expression:

[0078] E=f(S)

[0079]

[0080] In the formula, i represents the sequence number after all quality scores are arranged in ascending order, n represents the number of quality scores, f(S) represents the operation of S through the hyperbolic tangent function, and p max represents the maximum classification probability, s i represents the i-th quality score, s max represents the maximum quality score, p i Represents the classification probability corresponding to the i-th total quality score.

[0081] In order to make the purpose and advantages of the above calculation method clearer to those skilled in the art, the above quality score of 0-9 is used as an example for explanation. According to the above calculation method of the expected score, if the basic score corresponding to the maximum classification probability is 9 points, since 9 points is the maximum quality score, the classification probability matching 0-8 points is regarded as the remaining classification probability, and the larger the sum of the remaining classification probabilities, the smaller the expected score.

[0082] When the basic score corresponding to the maximum classification probability is not 9 points, but any one of 0-8 points; the classification probability corresponding to the quality score greater than the basic score is regarded as the remaining classification probability, and the larger the sum of the remaining classification probabilities, the larger the expected score.

[0083] For example, assuming that the basic score corresponding to the maximum classification probability is 5 points, the quality score of 6-9 points is greater than the basic score. If the sum of the remaining classification probabilities matching 6-9 points is larger, the corresponding expected score is larger.

[0084] In addition, the expression of the hyperbolic tangent function used in this embodiment is:

[0085]

[0086] Based on the same inventive concept as the content quality assessment method, this embodiment also provides a related device, including:

[0087] This embodiment also provides a content quality assessment device, which is applied to an assessment device. The content quality assessment device includes at least one functional module that can be stored in a memory in the form of software. Figure 3 As shown, from a functional perspective, the content quality assessment device may include:

[0088] The feature module 201 is used to obtain a first feature and a second feature of a media resource, wherein the first feature includes browsing statistics of the media resource, and the second feature includes inherent attribute information of the media resource.

[0089] In this embodiment, the feature module 201 is used to implement Figure 2 For a detailed description of the feature module 201, please refer to the detailed description of step S101A.

[0090] The scoring module 202 is used to input the first feature and the second feature into a pre-trained scoring model to obtain a target score for the media resource, wherein the target score is used to evaluate the user's preference for the media resource.

[0091] In this embodiment, the scoring module 202 is used to implement Figure 2 For the detailed description of the feature module 201, please refer to the detailed description of step S102A.

[0092] It should be understood that in some embodiments, the content quality assessment device may further include other functional modules or sub-modules for implementing other steps or sub-steps of the content quality assessment method. In other implementations, the feature module 201 and the scoring module 202 may also be used to implement other steps or sub-steps of the content quality assessment method.

[0093] This embodiment further provides an evaluation device, which includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, the content quality evaluation method is implemented.

[0094] This embodiment further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the content quality assessment method is implemented.

[0095] This embodiment also provides a model training method applied to a model training device, which is used to train a neural network model to obtain the above-mentioned evaluation model. Figure 4 As shown, the model training method includes:

[0096] S101B, obtaining a first sample feature and a second sample feature of a sample resource.

[0097] The first sample feature includes browsing statistics of the sample resource, and the second sample feature includes inherent attribute information of the sample resource;

[0098] S102B, input the first sample feature and the second sample feature into a neural network model for training to obtain a scoring model.

[0099] That is to say, this embodiment uses the first sample feature including browsing statistical information of the sample resource and the second sample feature including inherent attribute information of the sample resource during model training, and trains the neural model to explore the interactive relationship between the features, so that the trained scoring model can make a more objective and reasonable evaluation of the content quality of the media resources.

[0100] As an optional implementation, the neural network model is a DeepFM model. It should be understood that the DeepFM algorithm is increasingly being used because it effectively combines the advantages of factorization machines and neural networks in feature learning to extract both low-order and high-order combination features.

[0101] The DeepFM model includes two parts, FM and DNN. The FM part is used to extract the first-order features and the second-order features formed by the combination of the first-order features, and the DNN part is used to extract the high-order features formed by the full connection of the input first-order features. Since the DeepFM model is a prior art in the field, its specific structure will not be described in detail in this embodiment. The following focuses on a specific example to introduce its training process in detail.

[0102] Assuming that the sample resource is an audio novel album, the album is automatically labeled first. The label is used to define an integer quality score with a score range of 0-9. The corresponding labels are integers 0-9, where 0 corresponds to a quality score of 0 and 9 corresponds to a quality score of 9.

[0103] The model training device selects 15 features, including album browsing statistics and inherent attribute information, and converts the 15 features into feature vectors in vector form through feature engineering to determine the dimension of each feature.

[0104] Then, the model training device inputs the feature vector into the DeepFM model. Since the DNN part and the FM part share the input, the DeepFM model processes the input feature vector separately, outputs two output vectors, adds the two output vectors, and inputs the sum vector into a fully connected layer. Finally, the classification probability of each of the 10 quality scores is obtained, and the target score of the album is calculated based on the classification probability of each of the 10 quality scores.

[0105] like Figure 5 As shown, the detailed training process can include the following steps:

[0106] (1) Define the characteristic fields of the tag, that is, select 6 fields to classify the album. The tag identifier includes integers 0-9, which are used to mark 10 categories.

[0107] (2) Define the input feature fields and convert the album features into discrete variables (also known as index) through feature engineering.

[0108] Among them, continuous features are divided by quantiles, and discrete features are divided by clustering methods. The so-called continuous features refer to features that are difficult to classify, such as the number of album plays, the number of album collections, etc. Taking the number of album plays as an example, this embodiment divides it by quantiles, and the number of plays from 0 to 1000 is marked with the number 1, the number of plays from 1001 to 10000 is marked with the number 2, and the number of plays from 10001 to 100000 is marked with the number 3; the number of plays above 100000 is marked with the number 4.

[0109] The discrete features shown in the figure indicate that the features themselves have category attributes. For example, the type of album, the payment type of the album, etc. Taking the type of album as an example, urban albums are identified by the number 1; romantic albums are identified by the number 2; and horror albums are identified by the number 3.

[0110] (3) All datasets are divided into training set, validation set and test set. Then, the discrete variables of the training set are mapped and calculated. The DNN part and the FM part are used to encode the two vectors respectively. The vector weights are summed to obtain the final output sum vector.

[0111] (4) The summed vector is input into the fully connected layer of the next layer, and the summed vector is classified to obtain 10 classification probabilities of the album, which respectively represent the degree of match with the 10 quality scores of 0-9.

[0112] (5) The cross entropy loss function is used as the objective function to calculate the functional loss between the predicted probability and the true probability of the quality score. Then, the gradient descent method is used to minimize the objective function to train the model.

[0113] During training, the training set is used to calculate the loss function, the validation set is used to adjust the model's hyperparameters and verify the model's fitting ability, and the test set is used to evaluate the generalization ability of the final model.

[0114] (6) The trained DeepFM model is used as a scoring model to predict the classification probability of an album with different quality scores. Then, the quality score corresponding to the maximum classification probability is used as the basic score, and the expected score is calculated using the remaining classification probabilities. The two are added together to obtain the target score of the album.

[0115] It should be noted that the terms "first", "second", "third", etc. are only used to distinguish descriptions and cannot be understood as indicating or implying relative importance. In addition, the terms "include", "comprises", or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article, or device. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, method, article, or device including the element.

[0116] It should also be understood that the device and method disclosed in this embodiment can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0117] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0118] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. 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 described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0119] The above are only various implementations of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A content quality assessment method, characterized in that: Applied to an evaluation device, the method comprises: Acquire a first feature and a second feature of a media resource, wherein the first feature includes browsing statistics of the media resource, and the second feature includes inherent attribute information of the media resource; Inputting the first feature and the second feature into a pre-trained scoring model to obtain a plurality of classification probabilities output by the scoring model, wherein the plurality of classification probabilities respectively represent the matching degree between the media resource and different quality scores; According to the multiple classification probabilities, taking the quality score corresponding to the maximum classification probability as the basic score; Obtaining an expected score according to the remaining classification probabilities and the quality scores corresponding to the remaining classification probabilities, wherein the remaining classification probabilities do not include the maximum classification probability; The sum of the basic score and the expected score is taken as a target score, wherein the target score is used to evaluate the user's preference for the media resource.

2. The content quality assessment method according to claim 1, characterized in that: The obtaining the expected score according to the remaining classification probabilities and the quality scores corresponding to the remaining classification probabilities respectively includes: If the quality score corresponding to the maximum classification probability is the maximum quality score, the expected score E is calculated by the following expression: E=p max *(1-f(S)) If the quality score corresponding to the maximum classification probability is not the maximum quality score, the expected score E is calculated by the following expression: E=f(S) In the formula, i represents the sequence number after all quality scores are arranged in ascending order, n represents the number of quality scores, f(S) represents the operation of S through the hyperbolic tangent function, and p max represents the maximum classification probability, s i represents the i-th quality score, s max represents the maximum quality score, p i Represents the classification probability corresponding to the i-th total quality score.

3. The content quality assessment method according to claim 1, characterized in that: The inherent attribute information includes attribute information of the media resource and attribute information of the creator corresponding to the media resource.

4. A content quality assessment device, characterized in that: Applied to an evaluation device, the content quality evaluation device comprises: A feature module, used to obtain a first feature and a second feature of a media resource, wherein the first feature includes browsing statistics of the media resource, and the second feature includes inherent attribute information of the media resource; A scoring module, configured to input the first feature and the second feature into a pre-trained scoring model to obtain a plurality of classification probabilities output by the scoring model, wherein the plurality of classification probabilities respectively represent a matching degree between the media resource and different quality scores; According to the multiple classification probabilities, taking the quality score corresponding to the maximum classification probability as the basic score; Obtaining an expected score according to the remaining classification probabilities and the quality scores corresponding to the remaining classification probabilities, wherein the remaining classification probabilities do not include the maximum classification probability; The sum of the basic score and the expected score is taken as a target score, wherein the target score is used to evaluate the user's preference for the media resource.

5. An evaluation device, characterized in that The evaluation device includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the content quality evaluation method according to any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the content quality assessment method according to any one of claims 1 to 3 is implemented.

Images