Album content detection method, server and computer-readable storage medium

Through the album content detection method on the audio platform, the detection model is used to distinguish and distribute audio content of independence and contextual relevance, which solves the problem of poor distribution quality of the audio platform and improves the user experience.

CN114417053BActive Publication Date: 2024-11-15SHANGHAI ZHENGDA XIMALAYA NETWORK TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210063773.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-20
Publication Date
2024-11-15
Estimated Expiration
2042-01-20

AI Technical Summary

Technical Problem

Audio platforms have poor distribution quality when distributing sounds, resulting in poor user experience and inability to effectively distinguish and recommend independent and context-related audio content.

Method used

Using the album content detection method, the preset detection model predicts the characteristics and probability values ​​of shared information, and determines independence and contextual relevance through the trained pre-trained language model, and then distributes independent sharing information to ensure that the content received by the user can be Listen independently.

Benefits of technology

Improve the distribution quality of the audio platform and ensure that the content received by users is independent or context-sensitive, thereby improving the user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114417053B_ABST
    Figure CN114417053B_ABST
Patent Text Reader

Abstract

The embodiment of the present invention proposes an album content detection method, a server and a computer-readable storage medium, which belong to the field of data processing technology. The method is applied to the server, and the server is communicatively connected with a first client and a second client. The method includes: obtaining an album to be analyzed sent by the first client, and the album to be analyzed includes at least one sharing information, so as to use a preset detection model to predict the characteristics of each sharing information in the album to be analyzed and the probability value of each characteristic, wherein the characteristics include independence and context relevance, and then determine independent sharing information to distribute the independent sharing information to the second client, so that the sharing information received by the user of the second client can be listened to independently, thereby improving the current method of distributing audio texts on audio platforms, which has the problem of poor user experience.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an album content detection method, a server and a computer-readable storage medium. Background Art

[0002] In recent years, with the development of Internet technology and society, the scale of online audio and video users has maintained a continuous growth trend. Driven by policies, economy, technology and social needs, the online audio and video industry, which mainly focuses on online audiobooks, audio live broadcasts, knowledge payment and other business models, has gradually spread to users of all ages.

[0003] On audio platforms, audio and video hosts record online novels, radio dramas, and courses with their own voices and upload them to audio platforms for sharing. The voices uploaded by hosts are generally included in the albums they created. A host can have multiple voice albums of their own, and there are usually multiple voices in an album. The audio platform will distribute the voices uploaded by the hosts, such as recommending them to users. However, the current audio platforms have the problem of poor distribution quality. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide an album content detection method, a server and a computer-readable storage medium, which can improve the current method of distributing sound on audio platforms, which has the problem of poor distribution quality.

[0005] In order to achieve the above objectives, the technical solutions adopted by the embodiments of the present invention are as follows.

[0006] In a first aspect, the present invention provides an album content detection method, which adopts the following technical solution.

[0007] A method for detecting album content is applied to a server, wherein the server is in communication connection with a first client and a second client, and the method comprises:

[0008] Acquire the album to be analyzed sent by the first client, where the album to be analyzed includes at least one piece of sharing information;

[0009] Using a preset detection model, predict the characteristics of each piece of shared information in the album to be analyzed and the probability value of each characteristic; wherein the characteristics include independence and context relevance;

[0010] The sharing information whose probability value of independence is higher than the probability value of context relevance is determined as independent sharing information, so as to distribute the independent sharing information to the second client.

[0011] Furthermore, the method further comprises the step of obtaining a detection model, the step comprising:

[0012] Acquire multiple analyzed albums, label the titles of the shared information in the analyzed albums according to the characteristics to obtain multiple labeled data, and use all the labeled data as a training set, wherein the labeled data includes first label data and second label data, the first label represents context relevance, and the second label represents independence;

[0013] The pre-trained language model is trained using the labeled data, and the trained pre-trained language model is distilled to obtain a detection model.

[0014] Furthermore, the step of marking the sharing information in the analyzed album according to the characteristics to obtain the marked data includes:

[0015] For the analyzed albums that are completely context-related, two pieces of shared information from the analyzed albums are randomly extracted multiple times, the titles of the two pieces of shared information are used as a title pair, and the label of the title pair is set as the first label;

[0016] For the analyzed albums that are locally relevant and whose titles of the shared information include the relevant identifier, the shared information whose title includes the relevant identifier and the shared information whose serial numbers are adjacent are taken as a shared information pair, and the title of the shared information pair is set as the first label of the title pair, and the relevant identifier represents that the characteristic of the shared information is context-related shared information;

[0017] For the analyzed albums whose sharing information is all independent, randomly extracting two pieces of sharing information from the analyzed albums as sharing information pairs multiple times, taking the titles of the sharing information pairs as title pairs, and setting the labels of the title pairs as the second labels;

[0018] All the labeled title pairs are used as annotation data.

[0019] Furthermore, the method further comprises:

[0020] For the analyzed albums whose sharing information is completely independent, when the titles of the sharing information pairs are taken as title pairs, an offset mark is added to the beginning of the title in each of the title pairs to perform data enhancement.

[0021] Furthermore, the step of using the labeled data to train the pre-trained language model includes:

[0022] Taking each of the annotated data as input and the label of the annotated data as output target, training the NSP task of the pre-trained language model;

[0023] Obtaining a probability value of an output of the pre-trained language model;

[0024] When the probability value output by the pre-trained language model indicates that the characteristic prediction result of the input labeled data does not meet the preset expectation, the labeled data is re-labeled and then put into the training set;

[0025] The preset expectation includes that an absolute difference between a probability value of the first label and a probability value of the second label is greater than a preset value.

[0026] Furthermore, after obtaining the album to be analyzed sent by the first client, and before predicting the characteristics of each piece of sharing information in the album to be analyzed using a preset detection model, the method further includes:

[0027] Convert the Chinese characters expressing numbers in the title of each shared information into Arabic numerals.

[0028] Furthermore, when there is only one piece of sharing information in the album to be analyzed, the step of using a preset detection model to predict the characteristics of each piece of sharing information in the album to be analyzed and the probability value of each feature includes:

[0029] Pair the title of the shared information with the blank title to obtain a title pair;

[0030] The title pair is input into the detection model to obtain the characteristics of the shared information predicted by the detection model and the probability value of each characteristic.

[0031] Furthermore, when the album to be analyzed includes multiple pieces of sharing information, each piece of sharing information has its own sequence number;

[0032] The step of using a preset detection model to predict the characteristics of each piece of sharing information in the album to be analyzed and the probability value of each characteristic includes:

[0033] For each piece of the shared information, the title of the shared information and the title of any shared information whose serial number difference with the shared information is greater than a preset value are taken as a first title pair;

[0034] Inputting the first title pair into the detection model to obtain first characteristic information of the sharing information predicted by the detection model, wherein the first characteristic information includes characteristics and probability values ​​of each characteristic;

[0035] If the first characteristic information indicates that the context is irrelevant, pairing the title of the shared information with the title of the previous shared information and the title of the next shared information respectively to obtain two sets of second title pairs;

[0036] Inputting the second title pair into the detection model to obtain second characteristic information of the sharing information predicted by the detection model, wherein the second characteristic information includes characteristics and probability values ​​of each characteristic;

[0037] If at least one of the second characteristic information indicates context relevance, the characteristics of the shared information and the probability value of each characteristic are obtained according to the second characteristic information indicating context relevance.

[0038] In a second aspect, the present invention provides an album content detection device, which adopts the following technical solution.

[0039] An album content detection device is applied to a server, the server is connected to a first client for communication, and the album content detection device includes a receiving module, a prediction module and a processing module:

[0040] The receiving module is used to obtain the album to be analyzed sent by the first client, where the album to be analyzed includes at least one piece of sharing information;

[0041] The prediction module is used to predict the characteristics of each piece of sharing information in the album to be analyzed and the probability value of each characteristic by using a preset detection model; wherein the characteristics include independence and context-related;

[0042] The processing module is used to determine the sharing information whose independent probability value is higher than the context-related probability value as independent sharing information, so as to distribute the independent sharing information.

[0043] In a third aspect, the present application provides a server that adopts the following technical solution.

[0044] A server includes a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor can execute the machine executable instructions to implement the album content detection method as described in the first aspect.

[0045] In a fourth aspect, the present invention provides a computer-readable storage medium, which adopts the following technical solution.

[0046] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the album content detection method as described in the first aspect.

[0047] In the album content detection method provided by the embodiment of the present invention, the server uses a preset detection model to predict the characteristics of the shared information in the album to be analyzed sent by the first client, and simultaneously obtains the probability value of each characteristic, and based on the probability value, determines the shared information with a higher independent probability value as independent shared information, thereby distributing the independent shared information to the second client, so that the shared information received by the user of the second client can be listened to independently, thereby improving the problem of poor distribution quality existing in the current method of distributing audio texts on audio platforms.

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention 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 creative work.

[0050] Figure 1 A block diagram of a sharing system provided by an embodiment of the present invention is shown.

[0051] Figure 2 A block diagram of a server provided by an embodiment of the present invention is shown.

[0052] Figure 3 A flowchart showing some steps of the album content detection method provided by an embodiment of the present invention is shown.

[0053] Figure 4 A flow chart showing another part of the steps of the album content detection method provided by an embodiment of the present invention is shown.

[0054] Figure 5 A flowchart diagram showing another part of the steps of the album content detection method provided in an embodiment of the present invention is shown.

[0055] Figure 6 Shows Figure 5 Schematic diagram of the process of some sub-steps of S201.

[0056] Figure 7 Shows Figure 5 Schematic diagram of the process flow of some sub-steps of S202.

[0057] Figure 8 Shows Figure 3 and Figure 4 Schematic diagram of the process flow of some sub-steps of S103.

[0058] Fig. 9 Shows Figure 3 and Figure 4 A schematic flow chart of another part of the sub-steps of S103.

[0059] Fig.10 A block diagram of the album content detection device provided in this embodiment is shown.

[0060] Icons: 100 - sharing system; 110 - server; 120 - first client; 130 - second client; 140 - memory; 150 - processor; 160 - communication module; 170 - album content detection device; 180 - data acquisition module; 190 - prediction module; 200 - processing module. DETAILED DESCRIPTION

[0061] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0062] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0063] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" 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, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0064] With the development of Internet technology, the scale of online audio and video users has maintained a continuous growth trend. At present, the online audio and video industry, which is mainly based on business models such as online audiobooks, audio live broadcasts, video sharing, and knowledge payment, has gradually spread to users of all ages. On sharing platforms such as audio sharing platforms, audio and video sharing platforms, and knowledge sharing platforms, after the anchor (creator) uploads the work to the sharing platform, the sharing platform will distribute the work uploaded by the anchor, such as recommending it to users.

[0065] For example, on an audio sharing platform, an audio host records a work (online novels, radio dramas, courses, etc.) in the form of sound and then uploads the work to the audio sharing platform.

[0066] The works uploaded by the anchor are usually included in their own albums. Generally, a anchor can have multiple albums, and an album usually contains multiple works. Some of the works uploaded by the anchor are context-related, and some are independent. However, due to reasons such as the time of upload and naming, the audio and video sharing platform cannot directly know which works are independent and which are context-related. Therefore, when distributing works, audio and video sharing platforms, especially audio sharing platforms, indiscriminately distribute the works in the album to users (such as recommendations, response to searches), resulting in poor distribution quality, which in turn affects the user experience.

[0067] Among them, contextual relevance refers to the continuity between the work and other works that have been created and works to be created.

[0068] Based on the above considerations, the present invention provides an album content detection method, a server and a computer-readable storage medium.

[0069] Please refer to Figure 1 , is a block diagram of a sharing system 100, in which the album content detection method provided by the present invention can be applied. The sharing system 100 includes a server 110, a plurality of first clients 120 and a plurality of second clients 130, and an audio and video sharing platform is deployed on the server 110. The server 110 is in communication connection with the plurality of first clients 120 and the plurality of second clients 130.

[0070] The host logs in to the audio and video sharing platform through the first client 120 to upload his own works or browse the works on the audio and video platform. The user logs in to the audio and video platform through the second client 130 to browse various audio and video works. The server 110 is used to implement the album content detection method provided by the present invention.

[0071] The first client 120 and the second client 130 may be, but are not limited to: a personal computer, a tablet computer, a laptop computer, a mobile terminal, a wearable device, etc.

[0072] Please refer to Figure 2 , is a block diagram of the server 110. The server 110 includes a memory 140, a processor 150, and a communication module 160. The memory 140, the processor 150, and the communication module 160 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 via one or more communication buses or signal lines.

[0073] The memory 140 is used to store programs or data. The memory 140 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 programmable read-only memory (EEPROM), etc.

[0074] The processor 150 is used to read / write data, computer programs or machine executable instructions stored in the memory 140 and execute corresponding functions. When the processor 150 executes the computer program or machine executable instructions in the memory 140, the album content detection method provided by the present invention is implemented.

[0075] The communication module 160 is used to establish a communication connection between the server 110 and other communication terminals through the network, and to send and receive data through the network.

[0076] It should be understood that Figure 2 The structure shown is only a schematic diagram of the structure of the server 110. The server 110 may also include Figure 2 More or fewer components as shown, or with Figure 2 Different configurations are shown. Figure 2 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0077] In one embodiment, see Figure 3 , is a flow chart of the album content detection method provided by the present invention. In this embodiment, the method is applied to the above Figure 1 and Figure 2 The server 110 in FIG. 1 is used as an example.

[0078] S101, obtaining an album to be analyzed sent by a first client.

[0079] The album to be analyzed includes at least one piece of sharing information. Specifically, sharing information refers to a creative work, which can be an audio work, such as an online novel or radio drama recorded in the form of sound, or an audio-visual work, such as a paid knowledge video recorded in the form of sound + video.

[0080] Whenever the host uploads the sharing information (work) he / she created to an album with existing sharing information or a blank album through the first client 120, the album will become an album to be analyzed. The server 110 receives the album with the new sharing information (new work) uploaded.

[0081] S103, using a preset detection model, predicting the characteristics of each piece of shared information in the album to be analyzed and the probability value of each characteristic.

[0082] Among them, the characteristics include independence and context-dependency.

[0083] S105: Determine the sharing information whose probability value of independence is higher than the probability value of context relevance as independent sharing information, so as to distribute the independent sharing information to the second client.

[0084] Specifically, for each piece of shared information in the album, the server 110 uses a preset detection model to predict the characteristics of each piece of shared information and the probability value of each characteristic. Based on the predicted characteristics and the probability value of each characteristic, the server 110 regards the shared information whose independence probability value is higher than the context relevance probability value as independent shared information.

[0085] For example, for the shared information with the serial number 1 in the album, the detection model may predict the result as follows: 70% is independence and 30% is context relevance. At this time, since the probability value of independence (70%) is greater than the probability value of context relevance (30%), the shared information with the serial number 1 in the album is independent shared information.

[0086] After determining the independent sharing information, the server 110 may distribute the independent sharing information, for example, as daily recommended content, and recommend it to the second client 130 for the user to browse and listen to.

[0087] In the above-mentioned album content detection method, the server 110 uses a preset detection model to predict the characteristics of the shared information in the album to be analyzed sent by the first client 120, and simultaneously obtains the probability value of each characteristic, and based on the probability value, determines the shared information with a higher probability value of independence as independent shared information, thereby distributing the independent shared information to the second client 130, so that the shared information received by the user of the second client 130 can be listened to independently, thereby improving the problem of poor distribution quality existing in the current method of distributing audio texts on audio platforms.

[0088] It should be understood that context-related sharing information may also be packaged and distributed so that users can receive coherent sharing information to enhance the user experience.

[0089] In addition, refer to Figure 4The album content detection method provided by the present invention also includes step S102, which is executed after S101 and before S103.

[0090] S102, converting the Chinese characters expressing numbers in the title of each shared information into Arabic numerals.

[0091] Convert the numbers expressed in Chinese into Arabic numerals to facilitate the prediction work of the subsequent detection model.

[0092] Furthermore, after determining the independent sharing information, the non-independent sharing information is the sharing information with contextual relevance. This contextually relevant text will be referred to as the related sharing information in the following content. After determining the independent sharing information and the related sharing information, the server 110 can collect statistics on the related sharing information so that when the user searches for one of the related sharing information, the other related sharing information can be displayed together. The related sharing information can also be packaged and distributed, for example, packaged recommendation, etc.

[0093] Alternatively, when a user listens to or browses one of the related shared information, other related shared information related to the context will be displayed in the recommendation column.

[0094] And, for the detection model in S103, the present invention also includes a step of obtaining the detection model. Figure 5 , Figure 5 A flowchart of some steps for obtaining a detection model.

[0095] S201, obtaining a plurality of analyzed albums, annotating the titles of the shared information in the analyzed albums according to the characteristics, obtaining a plurality of annotated data, and using all the annotated data as a training set.

[0096] The annotated data includes first label data and second label data, the first label represents context relevance, and the second label represents independence.

[0097] S202: Use the labeled data to train the pre-trained language model, and distill the trained pre-trained language model to obtain a detection model.

[0098] A large number of analyzed albums are labeled to obtain a training set. Based on the training set, the pre-trained language model is trained to the expected requirements. The pre-trained model is then distilled to obtain a detection model. This improves the accuracy of the detection model while increasing the prediction speed of the detection model.

[0099] The method for obtaining the analyzed albums in S201 includes: first obtaining a large number of albums, and analyzing the obtained large number of albums using the following rules. It should be understood that the rules can be determined manually or executed by a machine.

[0100] First, if there are Chinese numbers in the title of each shared information in each album, the Chinese numbers in the title will be converted into Arabic numerals.

[0101] For example, thirty four is converted to 34.

[0102] Second, for each album, delete the date in the title of each shared information.

[0103] Third, the shared information including the relevant identifier in the shared information title is determined to be context-relevant.

[0104] Relevant identifiers include numbers or special texts enclosed in brackets such as "()", "[]", etc. Special texts include "finished", "end", etc. For example, if the title includes relevant identifiers such as "(1)", "[1]", "《1》", "[finished]", "[end]", and continuous prompt information, the shared information is determined to be contextually relevant.

[0105] Fourth, shared information containing continuous prompt information in the title is judged to be contextually relevant.

[0106] The continuous prompt information includes Lecture 1, Lecture XX, Episode 1, Episode XX, Chapter 1, Chapter XX, Chapter 1, Chapter XX, etc. It should be noted that the continuous prompt information is a type of related identification.

[0107] Fifth, if numbers appear in the titles of the shared information in the album, and the consecutive numbers exceed 30, or the consecutive numbers exceed 60% of the number of shared information in the album, then all the shared information in the album is determined to be context-related.

[0108] That is, if the album hits the fifth rule, the album is considered to be completely contextually relevant. If the album hits the third or fourth rule, but not the fifth rule, the album is considered to be partially relevant.

[0109] Based on the above rules, the characteristics of the shared information in the album can be analyzed to obtain the analyzed album, and then the annotation data can be obtained based on the analyzed album.

[0110] It should be understood that the above-mentioned rules are merely examples of some methods for determining whether the shared information in an album is completely contextually relevant, partially relevant, and the characteristics of the shared information, and are not the only determination method.

[0111] In addition, each shared information in the album has its own serial number. For example, if there are 30 shared information in the album, the serial numbers of the shared information are from 1 to 30.

[0112] Optionally, refer to Figure 6 , Figure 6 This is a flowchart of some sub-steps of the above S201. The following steps are used to annotate the sharing information in the analyzed album according to the characteristics to obtain the annotated data.

[0113] S201-1, for the analyzed album that is completely context-related, randomly extract two pieces of shared information from the analyzed album multiple times, use the titles of the two pieces of shared information as a title pair, and set the label of the title pair as the first label.

[0114] Among them, the analyzed albums with completely relevant context refer to the albums that hit the fifth rule mentioned above.

[0115] S201-2, for the analyzed albums that are locally relevant and whose titles of shared information include relevant identifiers, the shared information whose titles include relevant identifiers and the shared information whose serial numbers are adjacent are taken as a shared information pair, and the title of the shared information pair is set as the first label of the title pair.

[0116] The relevant identifier represents that the shared information is context-related shared information.

[0117] For example, an album includes 10 sharing information, among which the sharing information with sequence number 3 hits the third rule mentioned above, that is, the title of the sharing information with sequence number 3 includes the relevant identifier, and the album does not hit the fifth rule, then the sharing information with sequence number 3 and the sharing information with sequence number 4 are taken as a sharing information pair, or the sharing information with sequence number 3 and the sharing information with sequence number 2 are taken as a sharing information pair, and then a title pair is obtained, and the label of the title pair is set as the first label.

[0118] S201-3, for the analyzed albums whose sharing information is all independent, randomly extract two pieces of sharing information from the analyzed albums as sharing information pairs multiple times, use the titles of the sharing information pairs as title pairs, and set the labels of the title pairs as the second labels.

[0119] Specifically, the first label can be 1, and the second label can be 0. 0 represents independence, and 1 represents context relevance.

[0120] S201-4, taking all the labeled title pairs as annotation data.

[0121] The above steps S201 - 1 to S201 - 3 may be performed simultaneously or sequentially, and the order may be reversed.

[0122] And further, with respect to S201-3, the following method is also included: for the analyzed albums whose sharing information is completely independent, when the titles of the sharing information pairs are taken as title pairs, an offset mark is added to the beginning of the titles in each title pair to perform data enhancement.

[0123] Specifically, if all the sharing information in the album is independent, any two titles of the sharing information are randomly selected multiple times as a title pair, an offset mark is added to the beginning of each title of the title pair, and the label of the title pair is set to 0.

[0124] Among them, the offset mark includes texts of similar mark sequences such as "(1)", "No. 1" and "1," etc.

[0125] By using the above method, data annotation is performed to obtain a training set. Furthermore, by adding offset identifiers to the titles of the title pairs obtained from completely independent albums, the impact of shared information in the training set that contains similar identifier text but is actually irrelevant to the context can be offset, thereby improving the prediction accuracy of the detection model.

[0126] After obtaining the training set, for S202, refer to Figure 7 , which is a flowchart of some sub-steps of step S202, and the pre-trained language model is trained using labeled data through the following sub-steps.

[0127] S202-1, taking each annotated data as input and the label of the annotated data as output target, training the NSP task of the pre-trained language model.

[0128] The pre-trained language model may be a BERT model, or may be other pre-trained language models.

[0129] It should be understood that although the output target of the pre-trained language model is the label of the annotated data, the actual output is the feature probability value, namely the independence probability value and the context relevance probability value.

[0130] For example, if the label of the annotated data (i.e., title pair) is 1 (i.e., context relevance), then the probability value of the corresponding output target actually being context relevance is greater than the probability value of independence.

[0131] S202-2, obtaining a probability value of an output of the pre-trained language model.

[0132] Specifically, the output probability value includes an independence probability value and a context-dependency probability value.

[0133] S202-3, when the probability value output by the pre-trained language model indicates that the characteristic prediction result of the input labeled data does not meet the preset expectation, the labeled data is re-labeled and then put into the training set.

[0134] The preset expectation includes that the absolute difference between the probability value of the first label and the probability value of the second label is greater than the preset value. The preset value may be 0.1 (i.e., 10%). That is, the preset expectation may be that the absolute difference between the probability value of independence and the probability value of context relevance is greater than 0.1 (10%).

[0135] Specifically, for a certain annotated data, if the independence probability value and the context relevance probability value of the number predicted by the pre-trained language model are less than the preset expectation, the annotated data is re-annotated and then put into the training set.

[0136] After the re-labeled data is put back into the training set, the pre-trained language model will relearn the rules of the labeled data.

[0137] Through steps S202-1 to S202-3, while the model is being trained, the pre-trained language model is actively learned, that is, step S202-3 is executed, thereby improving the prediction accuracy of the pre-trained language model.

[0138] Based on the data in the training set, the pre-trained language model is trained back and forth. When the accuracy of the pre-trained language model reaches the preset accuracy, the model training is completed, and then the trained pre-trained language model is distilled to obtain the detection model.

[0139] Among them, the preset accuracy can be set according to actual needs.

[0140] Specifically, for the model distillation in S202, when the pre-trained language model is a BERT model, the model has 12 layers of parameters, so the trained pre-trained language model also has 12 layers of parameters. Therefore, in the distillation process, the 12 layers of parameters can be distilled to 3 layers of parameters to improve the model prediction speed.

[0141] It should be noted that since the labeled data in the training set includes data that hits the third and fourth rules above (i.e., the title pairs of shared information in the album), the trained detection model has learned the third and fourth rules above, so the detection model supports the prediction of a single piece of shared information. That is, it supports the feature prediction of shared information in an album that contains only one piece of shared information.

[0142] Specifically, when there is only one sharing information in the album to be analyzed, refer to Figure 8 , is a flowchart of some sub-steps of the above step S103, including the following steps.

[0143] S103-0, pair the title of the shared information with the blank title to obtain a title pair.

[0144] S103-1, input the title pair into the detection model to obtain the characteristics of the shared information predicted by the detection model and the probability value of each characteristic.

[0145] The characteristics and the probability values ​​of each characteristic include an independence probability value and a context-dependency probability value.

[0146] When there is only one piece of shared information in the album, since the detection model has learned the third and fourth rules, the title of the shared information and the empty title are used as a title pair, and the title pair is input into the detection model to obtain the characteristics of the shared information and the probability value of the characteristics predicted by the detection model.

[0147] When the album to be analyzed contains multiple sharing messages, refer to Fig. 9 , is a flowchart of another partial sub-step of the above S103, which realizes the prediction of the characteristics and characteristic probability values ​​of multiple pieces of sharing information of the album to be analyzed. It should be understood that the sharing information in the album to be analyzed has its own serial number.

[0148] S103-2: For each piece of shared information, the title of the shared information and the title of any shared information whose serial number difference with the shared information is greater than a preset value are taken as a first title pair.

[0149] Specifically, the preset value can be adjusted according to the actual situation of the shared information in the album. For example, the preset value can be any natural number that is not 0.

[0150] S103-3, input the first title pair into the detection model to obtain the first characteristic information of the shared information predicted by the detection model.

[0151] The first characteristic information includes characteristics and probability values ​​of each characteristic, that is, the first characteristic information includes independence probability values ​​and context relevance probability values.

[0152] S103-4, determine whether the first characteristic information indicates context relevance. If yes, execute S103-5, otherwise execute S103-6.

[0153] Specifically, when the probability value of context dependency is greater than the probability value of independence, it indicates that the context is dependent.

[0154] S103-5, obtaining the characteristics of the shared information and the probability value of each characteristic according to the first characteristic information.

[0155] S103-6, pairing the title of the shared information with the title of the previous shared information and the title of the next shared information respectively, to obtain two sets of second title pairs.

[0156] For example, if the serial number of the shared information is 10 and the patent to be analyzed includes 30 pieces of shared information, in the first characteristic information corresponding to the first title pair, if the probability value of independence is greater than the probability value of context relevance, it indicates that the context is irrelevant. At this time, a group of second title pairs is obtained based on the shared information and the shared information with serial number 9, and another group of second title pairs is obtained based on the shared information and the shared information with serial number 11.

[0157] Sharing information S103-7, inputting the second title pair into the detection model, and obtaining the second characteristic information of the shared information predicted by the detection model.

[0158] The second characteristic information includes characteristics and probability values ​​of each characteristic, that is, the second characteristic information includes independence probability values ​​and context relevance probability values.

[0159] S103-8, determine whether at least one second characteristic information indicates context relevance. If yes, execute S103-9, otherwise, execute S103-10.

[0160] Specifically, when the probability value of the context correlation in the second characteristic information is greater than the independence probability value, it indicates that the context is correlated.

[0161] S103-9, obtaining the characteristics of the shared information and the probability value of each characteristic according to the second characteristic information indicating context relevance.

[0162] Specifically, the second feature information indicating context relevance is used as the characteristic of the shared information and the probability value of the characteristic.

[0163] S103-10, obtaining the characteristics of the shared information and the probability value of each characteristic according to any second characteristic information or the first characteristic information.

[0164] In addition, S103 - 5 and S103 - 10 may also directly determine that the shared information is independent shared information.

[0165] Through the above S103-2 to S103-10, the characteristics of the album to be analyzed containing multiple pieces of sharing information can be predicted. In addition, when the prediction result shows that the context is irrelevant, the second title pair is paired again for prediction to improve the accuracy of the prediction of the characteristics of the sharing information.

[0166] Through S103-0 to S103-1, the characteristic detection of a single piece of sharing information can be realized, that is, it is applicable to the situation where there is only one piece of sharing information in the album; through S103-2 to S103-10, it is applicable to the situation where there is more than one piece of sharing information in the album, so it has a wider applicability.

[0167] It should be understood that although Figure 3-Figure 9The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 3-Figure 9 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0168] In order to execute the corresponding steps in the above embodiments and various possible methods, an implementation method of the album content detection device 170 is given below. Optionally, the album content detection device 170 can adopt the above Figure 2 For further information, please refer to Fig.10 , Fig.10 This is a functional module diagram of an album content detection device 170 provided in an embodiment of the present invention. It should be noted that the basic principle and technical effects of the album content detection device 170 provided in this embodiment are the same as those of the above-mentioned embodiments. For the sake of brief description, for parts not mentioned in this embodiment, reference can be made to the corresponding contents in the above-mentioned embodiments. The album content detection device 170 is applied to a server 110, which is in communication connection with a first client 120 and a second client 130. The album content detection device 170 includes a data acquisition module 180, a prediction module 190 and a processing module 200.

[0169] The data acquisition module 180 is used to acquire the album to be analyzed sent by the first client 120 .

[0170] The album to be analyzed includes at least one piece of sharing information.

[0171] The prediction module 190 is used to predict the characteristics of each piece of sharing information in the album to be analyzed and the probability value of each characteristic by using a preset detection model.

[0172] Among them, the characteristics include independence and context-dependency.

[0173] The processing module 200 is configured to determine the sharing information whose probability value of independence is higher than the probability value of context relevance as independent sharing information, so as to distribute the independent sharing information to the second client 130 .

[0174] Optionally, the above modules can be stored in the form of software or firmware. Figure 2The memory 140 shown in FIG. 140 may be fixed in the operating system (OS) of the server 110 and may be Figure 2 Meanwhile, the data and program codes required for executing the above modules may be stored in the memory 140.

[0175] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods 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 invention. 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 the 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.

[0176] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0177] 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 invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for 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 invention. 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, etc., which can store program codes.

[0178] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for detecting album content, characterized in that: Applied to a server, the server being in communication connection with a first client and a second client, the method comprising: Acquire the album to be analyzed sent by the first client, where the album to be analyzed includes at least one piece of sharing information; Using a preset detection model, predict the characteristics of each piece of shared information in the album to be analyzed and the probability value of each characteristic; wherein the characteristics include independence and context relevance; Determine the sharing information whose probability value of independence is higher than the probability value of context relevance as independent sharing information, and distribute the independent sharing information to the second client; When the album to be analyzed includes multiple pieces of sharing information, each piece of sharing information has its own serial number; The step of using a preset detection model to predict the characteristics of each piece of sharing information in the album to be analyzed and the probability value of each characteristic includes: For each piece of the shared information, the title of the shared information and the title of any shared information whose serial number difference with the shared information is greater than a preset value are taken as a first title pair; Inputting the first title pair into the detection model to obtain first characteristic information of the sharing information predicted by the detection model, wherein the first characteristic information includes characteristics and probability values ​​of each characteristic; If the first characteristic information indicates that the context is irrelevant, pairing the title of the shared information with the title of the previous shared information and the title of the next shared information respectively to obtain two sets of second title pairs; Inputting the second title pair into the detection model to obtain second characteristic information of the sharing information predicted by the detection model, wherein the second characteristic information includes characteristics and probability values ​​of each characteristic; If at least one of the second characteristic information indicates context relevance, the characteristics of the shared information and the probability value of each characteristic are obtained according to the second characteristic information indicating context relevance.

2. The album content detection method according to claim 1, characterized in that: The method further comprises the step of obtaining a detection model, the step comprising: Acquire multiple analyzed albums, label the titles of the shared information in the analyzed albums according to the characteristics to obtain multiple labeled data, and use all the labeled data as a training set, wherein the labeled data includes first label data and second label data, the first label represents context relevance, and the second label represents independence; The pre-trained language model is trained using the labeled data, and the trained pre-trained language model is distilled to obtain a detection model.

3. The album content detection method according to claim 2, characterized in that: The step of marking the sharing information in the analyzed album according to the characteristics to obtain the marked data includes: For the analyzed albums that are completely context-related, two pieces of shared information from the analyzed albums are randomly extracted multiple times, the titles of the two pieces of shared information are used as a title pair, and the label of the title pair is set as the first label; For the analyzed albums that are locally relevant and whose titles of the shared information include the relevant identifier, the shared information whose title includes the relevant identifier and the shared information whose serial numbers are adjacent are taken as a shared information pair, and the title of the shared information pair is set as the first label of the title pair, and the relevant identifier represents that the feature of the shared information is context-related shared information; For the analyzed albums whose sharing information is all independent, randomly extracting two pieces of sharing information from the analyzed albums as sharing information pairs multiple times, taking the titles of the sharing information pairs as title pairs, and setting the labels of the title pairs as the second labels; All the labeled title pairs are used as annotation data.

4. The album content detection method according to claim 3, characterized in that: The method further comprises: For the analyzed albums whose sharing information is completely independent, when the titles of the sharing information pairs are taken as title pairs, an offset mark is added to the beginning of the title in each of the title pairs to perform data enhancement.

5. The album content detection method according to claim 2, characterized in that: The step of using the labeled data to train the pre-trained language model includes: Taking each of the annotated data as input and the label of the annotated data as output target, training the NSP task of the pre-trained language model; Obtaining a probability value of an output of the pre-trained language model; When the probability value output by the pre-trained language model indicates that the characteristic prediction result of the input labeled data does not meet the preset expectation, the labeled data is re-labeled and then put into the training set; The preset expectation includes that an absolute difference between a probability value of the first label and a probability value of the second label is greater than a preset value.

6. The album content detection method according to claim 1, characterized in that: After obtaining the album to be analyzed sent by the first client, and before using a preset detection model to predict the characteristics of each piece of sharing information in the album to be analyzed, the method further includes: Convert the Chinese characters expressing numbers in the title of each shared information into Arabic numerals.

7. The album content detection method according to any one of claims 1 to 6, characterized in that: When there is only one piece of sharing information in the album to be analyzed, the step of using a preset detection model to predict the characteristics of each piece of sharing information in the album to be analyzed and the probability value of each feature includes: Pair the title of the shared information with the blank title to obtain a title pair; The title pair is input into the detection model to obtain the characteristics of the shared information predicted by the detection model and the probability value of each characteristic.

8. A server, characterized in that: It comprises a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor can execute the machine executable instructions to implement the album content detection method as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the album content detection method according to any one of claims 1 to 7 is implemented.