A method and apparatus for opinion verification and data publishing

Abstract

The disclosure provides a viewpoint verification method, a data publishing method and device, wherein the viewpoint verification method comprises: obtaining a viewpoint to be verified, initiating a search based on the viewpoint to be verified, and determining a first document corresponding to the viewpoint to be verified; determining first text content associated with the viewpoint to be verified in the first document; determining a first target document in which target text content in the first text content is located, and determining an associated document of the first target document; determining second text content associated with the viewpoint to be verified in the associated document based on the viewpoint to be verified and the target text content; and verifying the authenticity of the viewpoint to be verified based on the first text content and the second text content.

Description

Technical Field

[0001] This disclosure relates to the field of information processing technology, and more specifically, to a method for verifying opinions, a method for publishing data, and an apparatus. Background Technology

[0002] With the advent of the information age, the types and volume of information disseminated on the internet have become increasingly diverse. However, not all information disseminated online is truthful; the spread of false information can have a significant negative social impact. Therefore, how to verify the authenticity of information has become an urgent technical problem to be solved. Summary of the Invention

[0003] This disclosure provides at least one method for verifying viewpoints, a method for publishing data, and an apparatus.

[0004] In a first aspect, embodiments of this disclosure provide a method for verifying an opinion, including:

[0005] Obtain the viewpoint to be verified, initiate a search based on the viewpoint to be verified, and determine the first document corresponding to the viewpoint to be verified;

[0006] Determine the first text content in the first document that is associated with the viewpoint to be verified;

[0007] Determine the first target document containing the target text content in the first text content, and determine the associated document of the first target document;

[0008] Based on the viewpoint to be verified and the target text content, determine the second text content in the associated document that is associated with the viewpoint to be verified;

[0009] The authenticity of the viewpoint to be verified is verified based on the first text content and the second text content.

[0010] In the above method, a search can be initiated based on the viewpoint to be verified to determine the coarse document (i.e., the first document) related to the viewpoint to be verified. Then, the refined document (i.e., the second document) related to the viewpoint to be verified can be determined through the first document. Then, the authenticity of the viewpoint to be verified can be verified through the text content related to the viewpoint to be verified in the first and second documents. Thus, through multiple rounds of iterative retrieval, a comprehensive verification of the viewpoint to be verified can be achieved, improving the accuracy and efficiency of the verification.

[0011] In one possible implementation, the step of initiating a search based on the viewpoint to be verified and determining the first document corresponding to the viewpoint to be verified includes:

[0012] Identify the keywords contained in the viewpoint to be verified;

[0013] Initiate a search based on the keywords and identify the first document containing the keywords.

[0014] In this way, a preliminary search for relevant documents of the viewpoint to be verified can be achieved, narrowing the scope of verification of the viewpoint and making it easier to verify the authenticity of the viewpoint through the content of the first document.

[0015] In one possible implementation, determining the first text content in the first document associated with the viewpoint to be verified includes:

[0016] For any first document, the viewpoint to be verified, the text content in the first document, and the document title of the first document are encoded based on a pre-trained encoder to determine the first encoding sequence corresponding to the first document;

[0017] The first encoded sequence is input into a pre-trained feature fusion network to determine the first relevance between each text content in the first document and the viewpoint to be verified.

[0018] The first text content is determined based on the first relevance between the text content in each first document and the viewpoint to be verified.

[0019] Here, when determining the first encoding sequence, combining it with the document title of the first document can address the referential issues during the encoding process. Furthermore, this method allows for the selection of first text content related to the viewpoint to be verified from the first document. This first text content can then be used to verify the authenticity of the viewpoint, improving both verification efficiency and accuracy.

[0020] In one possible implementation, determining the first target document containing the target text content within the first text content includes:

[0021] The first text content whose first relevance meets the preset condition is taken as the target text content, and the first target document where the target text content is located is determined;

[0022] The step of determining the associated documents of the first target document includes:

[0023] Determine the associated document corresponding to the hyperlink in the first target document; or,

[0024] Identify associated documents whose relevance to the first target document exceeds a preset value.

[0025] By retrieving related documents, we can more comprehensively identify the text content related to the viewpoint to be verified, solve the multi-hop problem, and improve the accuracy of verifying the viewpoint.

[0026] In one possible implementation, determining the second text content in the associated document that is associated with the viewpoint to be verified, based on the viewpoint to be verified and the target text content, includes:

[0027] For any associated document, the viewpoint to be verified, the target text content, the text content in the associated document, and the document title of the associated document are encoded based on a pre-trained encoder to determine the second encoding sequence corresponding to the associated document;

[0028] The second encoded sequence is input into a pre-trained feature fusion network to determine the second relevance between each text content in the associated document and the viewpoint to be verified.

[0029] The second text content is determined based on the second relevance between the text content in each associated document and the viewpoint to be verified.

[0030] Here, the purpose of adding the target text content when determining the second encoding sequence is to solve the multi-hop problem and make it easier to find the second text content more accurately.

[0031] In one possible implementation, the method further includes training the encoder according to the following method:

[0032] Multiple sets of sample data are obtained, each set of sample data includes sample premises and sample arguments, and each set of sample data has a corresponding label, which is used to represent the relationship between the sample premises and the sample arguments.

[0033] Based on the sample data, the encoder to be trained is pre-trained to obtain the pre-trained encoder;

[0034] The pre-trained encoder is fine-tuned to obtain a trained encoder.

[0035] This implementation method enables pre-training of the encoder. However, the focus of pre-training is to train the encoder to recognize the relationship between a set of data, that is, the relationship between a premise and an argument. In practical applications, a premise often needs to be supported by multiple arguments, so the encoder needs to be fine-tuned.

[0036] In one possible implementation, the pre-training of the encoder to be trained based on the sample data to obtain a pre-trained encoder includes:

[0037] The sample data is input into the encoder to be trained to obtain the third encoding sequence corresponding to the sample data;

[0038] The third encoded sequence is classified based on the first classifier to obtain the predicted relationship corresponding to the sample data;

[0039] Based on the prediction relationship corresponding to the sample data and the label corresponding to the sample data, the encoder to be trained is pre-trained to obtain the pre-trained encoder.

[0040] In one possible implementation, the fine-tuning of the pre-trained encoder to obtain a trained encoder includes:

[0041] Obtain sample viewpoints and corresponding tags for the sample viewpoints, wherein the tags for the sample viewpoints are used to characterize the authenticity of the sample viewpoints;

[0042] Based on the pre-trained encoder, the sample text content associated with the sample viewpoint is determined;

[0043] The sample viewpoint and the associated sample text content are input into the pre-trained encoder to determine the first sample encoding sequence;

[0044] After processing the first sample encoding sequence and the sample viewpoint based on a multi-level attention mechanism, the classification result corresponding to the sample viewpoint is determined based on the second classifier;

[0045] Based on the classification results and labels corresponding to the sample viewpoints, the pre-trained encoder is fine-tuned to obtain a trained encoder.

[0046] By fine-tuning the pre-trained encoder, the encoder can better distinguish the relationship between a sample viewpoint (equivalent to the premise in the pre-training process) and multiple sample text contents (equivalent to the arguments in the pre-training process), resulting in more accurate encoding results.

[0047] In one possible implementation, verifying the authenticity of the viewpoint to be verified based on the first text content and the second text content includes:

[0048] The first text content, the second text content, and the viewpoint to be verified are input into a pre-trained encoder to obtain a fourth encoding sequence;

[0049] The fourth encoding sequence and the viewpoint to be verified are processed based on a multi-level attention mechanism to obtain the fifth encoding sequence;

[0050] The fifth encoded sequence is classified based on a pre-trained second classifier to determine the authenticity of the viewpoint to be verified.

[0051] This method allows for the comprehensive identification of textual content related to the viewpoint to be verified, and the authenticity of the viewpoint can be verified based on the textual content, thereby improving the accuracy of verification and further enhancing information security.

[0052] Secondly, this disclosure also provides a data publishing method, including:

[0053] Get the data to be published;

[0054] The data to be published is taken as a viewpoint to be verified, and the authenticity of the data to be published is detected based on the viewpoint verification method described in the first aspect or any possible implementation of the first aspect.

[0055] If the data to be published is real data, then the data to be published will be published.

[0056] This method allows for the verification of data authenticity before its release, ensuring the publication of accurate data and enhancing the security of information.

[0057] Thirdly, this disclosure also provides a viewpoint verification device, comprising:

[0058] The first acquisition module is used to acquire the viewpoint to be verified, initiate a search based on the viewpoint to be verified, and determine the first document corresponding to the viewpoint to be verified.

[0059] The first determining module is used to determine the first text content in the first document that is associated with the viewpoint to be verified.

[0060] The second determining module is used to determine the first target document where the target text content in the first text content is located, and to determine the associated document of the first target document;

[0061] The third determining module is used to determine, based on the viewpoint to be verified and the target text content, the second text content in the associated document that is associated with the viewpoint to be verified;

[0062] The verification module is used to verify the authenticity of the viewpoint to be verified based on the first text content and the second text content.

[0063] In one possible implementation, when the first acquisition module initiates a search based on the viewpoint to be verified and determines the first document corresponding to the viewpoint to be verified, it is used to:

[0064] Identify the keywords contained in the viewpoint to be verified;

[0065] Initiate a search based on the keywords and identify the first document containing the keywords.

[0066] In one possible implementation, the first determining module, when determining the first text content in the first document associated with the viewpoint to be verified, is configured to:

[0067] For any first document, the viewpoint to be verified, the text content in the first document, and the document title of the first document are encoded based on a pre-trained encoder to determine the first encoding sequence corresponding to the first document;

[0068] The first encoded sequence is input into a pre-trained feature fusion network to determine the first relevance between each text content in the first document and the viewpoint to be verified.

[0069] The first text content is determined based on the first relevance between the text content in each first document and the viewpoint to be verified.

[0070] In one possible implementation, the second determining module, when determining the first target document where the target text content in the first text content is located, is used to:

[0071] The first text content whose first relevance meets the preset condition is taken as the target text content, and the first target document where the target text content is located is determined;

[0072] The second determining module, when determining the associated document of the first target document, is used to:

[0073] Determine the associated document corresponding to the hyperlink in the first target document; or,

[0074] Identify associated documents whose relevance to the first target document exceeds a preset value.

[0075] In one possible implementation, the third determining module, when determining the second text content in the associated document associated with the viewpoint to be verified based on the viewpoint to be verified and the target text content, is used to:

[0076] For any associated document, the viewpoint to be verified, the target text content, the text content in the associated document, and the document title of the associated document are encoded based on a pre-trained encoder to determine the second encoding sequence corresponding to the associated document;

[0077] The second encoded sequence is input into a pre-trained feature fusion network to determine the second relevance between each text content in the associated document and the viewpoint to be verified.

[0078] The second text content is determined based on the second relevance between the text content in each associated document and the viewpoint to be verified.

[0079] In one possible implementation, the apparatus further includes a training module for training the encoder according to the following method:

[0080] Multiple sets of sample data are obtained, each set of sample data includes sample premises and sample arguments, and each set of sample data has a corresponding label, which is used to represent the relationship between the sample premises and the sample arguments.

[0081] Based on the sample data, the encoder to be trained is pre-trained to obtain the pre-trained encoder;

[0082] The pre-trained encoder is fine-tuned to obtain a trained encoder.

[0083] In one possible implementation, the training module, when pre-training the encoder to be trained based on the sample data to obtain a pre-trained encoder, is used to:

[0084] The sample data is input into the encoder to be trained to obtain the third encoding sequence corresponding to the sample data;

[0085] The third encoded sequence is classified based on the first classifier to obtain the predicted relationship corresponding to the sample data;

[0086] Based on the prediction relationship corresponding to the sample data and the label corresponding to the sample data, the encoder to be trained is pre-trained to obtain the pre-trained encoder.

[0087] In one possible implementation, the training module, when fine-tuning the pre-trained encoder to obtain a trained encoder, is used to:

[0088] Obtain sample viewpoints and corresponding tags for the sample viewpoints, wherein the tags for the sample viewpoints are used to characterize the authenticity of the sample viewpoints;

[0089] Based on the pre-trained encoder, the sample text content associated with the sample viewpoint is determined;

[0090] The sample viewpoint and the associated sample text content are input into the pre-trained encoder to determine the first sample encoding sequence;

[0091] After processing the first sample encoding sequence and the sample viewpoint based on a multi-level attention mechanism, the classification result corresponding to the sample viewpoint is determined based on the second classifier;

[0092] Based on the classification results and labels corresponding to the sample viewpoints, the pre-trained encoder is fine-tuned to obtain a trained encoder.

[0093] In one possible implementation, the verification module, when verifying the authenticity of the viewpoint to be verified based on the first text content and the second text content, is used to:

[0094] The first text content, the second text content, and the viewpoint to be verified are input into a pre-trained encoder to obtain a fourth encoding sequence;

[0095] The fourth encoding sequence and the viewpoint to be verified are processed based on a multi-level attention mechanism to obtain the fifth encoding sequence;

[0096] The fifth encoded sequence is classified based on a pre-trained second classifier to determine the authenticity of the viewpoint to be verified.

[0097] Fourthly, embodiments of this disclosure also provide a data publishing device, including:

[0098] The second acquisition module is used to acquire the data to be published;

[0099] The detection module is used to take the data to be published as a viewpoint to be verified, and to detect the authenticity of the data to be published based on the viewpoint verification method described in the first aspect or any possible implementation of the first aspect.

[0100] The publishing module is used to publish the data to be published if the data to be published is real data.

[0101] Fifthly, embodiments of this disclosure also provide a computer device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via the bus, and when the machine-readable instructions are executed by the processor, the steps of the first aspect described above, or any possible implementation of the first aspect, or the steps of the second aspect described above are executed.

[0102] In a sixth aspect, embodiments of this disclosure also provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the first aspect described above, or any possible implementation of the first aspect, or performs the steps of the second aspect described above.

[0103] For a description of the effects of the aforementioned viewpoint verification device, data publishing device, computer equipment, and computer-readable storage medium, please refer to the descriptions of the aforementioned viewpoint verification method and data publishing method; they will not be repeated here.

[0104] To make the above-mentioned objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0105] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly described below. These drawings are incorporated in and constitute a part of this specification. They illustrate embodiments conforming to this disclosure and, together with the specification, serve to explain the technical solutions of this disclosure. It should be understood that the following drawings only show some embodiments of this disclosure and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0106] Figure 1 A flowchart of an opinion verification method provided by an embodiment of this disclosure is shown;

[0107] Figure 2 A flowchart of the method for determining the first text content in the opinion verification method provided in this embodiment of the present disclosure is shown;

[0108] Figure 3 A schematic diagram of a first relevance determination method provided by an embodiment of this disclosure is shown;

[0109] Figure 4 A flowchart of the method for determining the first text content in the opinion verification method provided in this embodiment of the present disclosure is shown;

[0110] Figure 5 A schematic diagram of a second relevance determination method provided by an embodiment of this disclosure is shown;

[0111] Figure 6 A flowchart is shown of a method for verifying the authenticity of a viewpoint to be verified in the viewpoint verification method provided in this embodiment of the present disclosure;

[0112] Figure 7 A flowchart illustrating a pre-training method for an encoder provided in an embodiment of this disclosure is shown.

[0113] Figure 8 A flowchart illustrating a method for fine-tuning the pre-trained encoder provided in an embodiment of this disclosure is shown.

[0114] Figure 9 A schematic diagram of the architecture of an opinion verification device provided in an embodiment of this disclosure is shown;

[0115] Figure 10 This diagram illustrates the architecture of a data publishing apparatus provided in an embodiment of the present disclosure.

[0116] Figure 11 A schematic diagram of the structure of a computer device provided in an embodiment of this disclosure is shown. Detailed Implementation

[0117] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. The components of the embodiments of this disclosure described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed disclosure, but merely represents selected embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0118] With the advent of the information age, the types and volume of information disseminated on the internet have become increasingly diverse. However, not all information disseminated online is truthful; the spread of false information can have a significant negative social impact. Therefore, how to verify the authenticity of information has become an urgent technical problem to be solved.

[0119] Based on the above research, this disclosure provides a method for verifying viewpoints, a method for publishing data, and an apparatus. It can initiate a search based on the viewpoint to be verified, determine a coarse document (i.e., a first document) related to the viewpoint to be verified, then determine a refined document (i.e., a second document) related to the viewpoint to be verified through the first document, and then verify the authenticity of the viewpoint to be verified through the text content related to the viewpoint to be verified in the first and second documents. Thus, through multiple rounds of iterative retrieval, a comprehensive verification of the viewpoint to be verified can be achieved, improving the accuracy and efficiency of the verification.

[0120] To facilitate understanding of the embodiments of this disclosure, the technical terms involved in the following embodiments will first be explained.

[0121] Multi-hop problems refer to problems that require multiple viewpoints to prove. For example, if a viewpoint is "the lead actor in movie A is from country B," then verifying this viewpoint requires first determining who the lead actor in movie A is, and then determining which country the lead actor is from.

[0122] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0123] To facilitate understanding of this embodiment, a viewpoint verification method disclosed in this disclosure will first be described in detail. The execution entity of the viewpoint verification method provided in this disclosure is generally a computer device with certain computing capabilities. This computer device may include, for example, a terminal device, a server, or other processing devices. The terminal device may be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, wearable device, etc. In some possible implementations, this viewpoint verification method can be implemented by a processor calling computer-readable instructions stored in memory.

[0124] See Figure 1 The diagram shows a flowchart of a viewpoint verification method provided in an embodiment of this disclosure. The method includes steps 101 to 105, wherein:

[0125] Step 101: Obtain the viewpoint to be verified, initiate a search based on the viewpoint to be verified, and determine the first document corresponding to the viewpoint to be verified.

[0126] Step 102: Determine the first text content in the first document that is associated with the viewpoint to be verified.

[0127] Step 103: Determine the first target document where the target text content in the first text content is located, and determine the associated document of the first target document.

[0128] Step 104: Based on the viewpoint to be verified and the target text content, determine the second text content in the associated document that is associated with the viewpoint to be verified.

[0129] Step 105: Verify the authenticity of the viewpoint to be verified based on the first text content and the second text content.

[0130] The following is a detailed description of the steps described above.

[0131] Regarding step 101,

[0132] In one possible implementation, the viewpoint to be verified can be one whose veracity cannot be determined solely by textual description, such as news articles to be verified. Generally, the veracity of the viewpoint can be determined by searching relevant documents.

[0133] In one possible implementation, when initiating a search based on the viewpoint to be verified and determining the first document corresponding to the viewpoint, for example, the keywords contained in the viewpoint to be verified can be determined first, and then a search can be initiated based on the keywords to determine the first document containing the keywords.

[0134] When determining the keywords contained in the viewpoint to be verified, for example, it can be determined by a word segmentation model (e.g., an N-gram model).

[0135] In this way, a preliminary search for relevant documents of the viewpoint to be verified can be achieved, narrowing the scope of verification of the viewpoint and making it easier to verify the authenticity of the viewpoint through the content of the first document.

[0136] Regarding step 102,

[0137] Not all content in the first document can be used to verify the viewpoint to be verified. Therefore, the first text content that is directly related to the viewpoint to be verified can be found in the first document first.

[0138] In one possible implementation, when determining the first text content in the first document associated with the viewpoint to be verified, it can be done by, for example... Figure 2 The method shown includes the following steps:

[0139] Step 201: For any first document, encode the viewpoint to be verified, the text content in the first document, and the document title of the first document based on a pre-trained encoder to determine the first encoding sequence corresponding to the first document.

[0140] The text content in the first document may be, for example, the sentences in the first document, or the paragraphs in the first document.

[0141] Here, when determining the first encoding sequence, combining it with the document title of the first document can address the referential issues during the encoding process.

[0142] For example, if the viewpoint to be verified is "What is Zhang San's birthday?", and the title of the first document is "The Life of Zhang San", in the sentences of the first document, "Zhang San" may be referred to by the pronoun "he". Therefore, by combining the title of the first document "The Life of Zhang San", it can be determined that "he" in the sentence refers to Zhang San. This allows for a more accurate determination of the first relevance between each text content in the first document and the viewpoint to be verified, meaning that the precision of the first encoding sequence can be higher.

[0143] Step 202: Input the first encoded sequence into a pre-trained feature fusion network to determine the first relevance between each text content in the first document and the viewpoint to be verified.

[0144] The specific flowcharts for steps 201 and 202 can be exemplified as follows: Figure 3 As shown, the relevance X represents the first relevance between the viewpoint to be verified and the Xth text content in the first document, and m represents the number of text contents in the first document.

[0145] Step 203: Determine the first text content based on the first relevance between the text content in each first document and the viewpoint to be verified.

[0146] For example, the N text contents with the highest relevance can be used as the first text content, where N is a positive integer.

[0147] This method allows for the selection of first text content related to the viewpoint to be verified from the first document. Furthermore, the authenticity of the viewpoint can be verified using the first text content, thereby improving verification efficiency and accuracy.

[0148] The specific training process for the feature fusion network and encoder will be described below, and will not be elaborated here.

[0149] Regarding step 103,

[0150] In one possible implementation, the target text content may refer to part or all of the content in the first text content. Specifically, the first text content whose corresponding first relevance meets the preset conditions may be used as the target text content.

[0151] For example, the first text content can be sorted in descending order of its corresponding first relevance, and the first text content with the highest first relevance can be used as the target text content.

[0152] To address the multi-hop issue, documents associated with the first target document may also be related to the viewpoint to be verified, and therefore can be considered as associated documents. When determining associated documents of the first target document, for example, associated documents corresponding to hyperlinks in the first target document can be identified; alternatively, associated documents with a relevance exceeding a preset value can be identified.

[0153] For example, internal links can be added to the first target document, and the documents linked to by these internal links are documents related to the content of the first target document. Alternatively, the relevance between the documents can be determined in advance, and then the documents with a higher relevance to the first target document can be selected as the associated documents.

[0154] Regarding step 104,

[0155] In one possible implementation, when determining the second text content in the associated document that is associated with the viewpoint to be verified based on the viewpoint to be verified and the target text content, for example, it can be done by means of... Figure 4 The method shown includes the following steps:

[0156] Step 401: For any associated document, encode the viewpoint to be verified, the target text content, the text content in the associated document, and the document title of the associated document based on a pre-trained encoder to determine the second encoding sequence corresponding to the associated document.

[0157] Step 402: Input the second encoded sequence into a pre-trained feature fusion network to determine the second relevance between each text content in the associated document and the viewpoint to be verified.

[0158] The specific flowcharts for steps 401 and 402 can be exemplified as follows: Figure 5 As shown, the relevance X represents the second relevance between the viewpoint to be verified and the Xth text content in the associated document, and m represents the number of text contents in the associated document.

[0159] Step 403: Determine the second text content based on the second relevance between the text content in each associated document and the viewpoint to be verified.

[0160] In one possible implementation, the corresponding second relevance scores can be sorted in descending order.

[0161] Here, the method for determining the second text content is basically the same as the method for determining the first text content. The difference is that the target text content is added during the encoding process when determining the second text content.

[0162] The purpose of adding the target text content here is to solve the multi-hop problem and make it easier to find the second text content more accurately.

[0163] For example, if the viewpoint to be verified is "the chairman of the XX Association was born in 1990", and the target text content is "Li Si is the chairman of the XX Association", then when searching for the second text content, combining the target text content "Li Si is the chairman of the XX Association" can more accurately find the second text content associated with the viewpoint to be verified, "the chairman of the XX Association was born in 1990".

[0164] The pre-trained encoder used in step 401 can be the same encoder as the encoder in step 201, and the pre-trained feature fusion network used in step 402 can be the same network as the feature fusion network in step 203.

[0165] By retrieving related documents, we can more comprehensively identify the text content related to the viewpoint to be verified, solve the multi-hop problem, and improve the accuracy of verifying the viewpoint.

[0166] Regarding step 105,

[0167] In one possible implementation, when verifying the authenticity of the viewpoint to be verified based on the first text content and the second text content, for example, it can be done by means of... Figure 6 The verification method shown includes the following steps:

[0168] Step 601: Input the first text content, the second text content, and the viewpoint to be verified into a pre-trained encoder to obtain the fourth encoding sequence.

[0169] Step 602: Process the fourth encoding sequence and the viewpoint to be verified based on a multi-level attention mechanism to obtain the fifth encoding sequence.

[0170] Step 603: Classify the fifth encoded sequence based on the pre-trained second classifier to determine the authenticity of the viewpoint to be verified.

[0171] The encoder described in step 601 can be the same encoder as those in steps 201 and 401; the second classifier in step 603 can be a three-classifier, and its classification results include three categories: real, fake, and undetermined.

[0172] The second classifier can be trained along with the encoder's training process, the specific training process of which will be described below.

[0173] This method allows for the comprehensive identification of textual content related to the viewpoint to be verified, and the authenticity of the viewpoint can be verified based on the textual content, thereby improving the accuracy of verification and further enhancing information security.

[0174] In one possible implementation, to improve verification efficiency, when verifying the authenticity of the viewpoint to be verified based on the first text content and the second text content, verification text content related to the viewpoint to be verified can be filtered from the first text content and the second text content based on the first relevance corresponding to the first text content and the second relevance corresponding to the second text content, and then the authenticity of the viewpoint to be verified can be verified based on the verification text content.

[0175] Specifically, the method for verifying the viewpoint to be verified based on the verification text content is similar to... Figure 6 The method shown is the same, so it will not be repeated here.

[0176] The training process of the encoder and the feature fusion network will be described below. Generally, the training process of the feature fusion network follows that of the encoder. First, we will introduce the training process of the encoder.

[0177] The encoder training process mainly consists of two parts: one is the pre-training process of the encoder, and the other is the fine-tuning of the pre-trained encoder. The encoder after fine-tuning is the trained encoder.

[0178] I. Encoder pre-training.

[0179] See Figure 7 The diagram shown is a flowchart illustrating a pre-training method for an encoder provided in this embodiment of the present disclosure, which includes the following steps:

[0180] Step 701: Obtain multiple sets of sample data, wherein each set of sample data contains sample premises and sample arguments, and each set of sample data has a corresponding label, which is used to represent the relationship between the sample premises and the sample arguments.

[0181] Here, the sample arguments are used to demonstrate whether the sample premise is true, and the labels corresponding to the sample data may include any one of "contains", "opposes", or "neutral".

[0182] For example, if the premise of a sample is "A man is swimming in the river" and the evidence is "The man is sitting in the room," then the label for this set of sample data is "opposing." If the premise of a sample is "Zhang San is the chairman of Association A" and the evidence is "Li Si is the chairman of Association B," then the label for this set of sample data is "neutral." If the premise of a sample is "A man is swimming in the river" and the evidence is "There is a man in the river," then the label for this set of sample data is "inclusive."

[0183] Step 702: Input the sample data into the encoder to be trained to obtain the third encoding sequence corresponding to the sample data.

[0184] Step 703: Classify the third encoded sequence based on the first classifier to obtain the predicted relationship corresponding to the sample data.

[0185] It should be noted that the first classifier in step 703 is different from the second classifier in step 603. The input of the second classifier in step 603 is the fifth encoding sequence, which is the encoding sequence after further processing of the encoder output, while the input of the first classifier in step 703 is the third encoding sequence, which is the encoding sequence directly output by the encoder.

[0186] The first classifier in step 703 is also a three-classifier, and the classification result includes any one of "contains", "opposes", and "neutral", that is, the predicted relationship is any one of "contains", "opposes", and "neutral".

[0187] Step 704: Based on the prediction relationship corresponding to the sample data and the label corresponding to the sample data, pre-train the encoder to be trained to obtain the pre-trained encoder.

[0188] In one possible implementation, when pre-training the encoder to be trained based on the prediction relation corresponding to the sample data and the label corresponding to the sample data, the loss value (e.g., the cross-entropy loss value) during the current training process can be calculated based on the prediction relation corresponding to the sample data and the label corresponding to the sample data, and if the training cutoff condition is not met, the process returns to step 701.

[0189] This implementation method enables pre-training of the encoder. However, the focus of pre-training is to train the encoder to recognize the relationship between a set of data, that is, the relationship between a premise and an argument. In practical applications, a premise often needs to be supported by multiple arguments, so the encoder needs to be fine-tuned.

[0190] II. Encoder fine-tuning.

[0191] See Figure 8 The diagram shown is a flowchart illustrating a method for fine-tuning a pre-trained encoder according to an embodiment of this disclosure, comprising the following steps:

[0192] Step 801: Obtain sample viewpoints and corresponding tags for the sample viewpoints. The tags for the sample viewpoints are used to characterize the authenticity of the sample viewpoints.

[0193] Step 802: Based on the pre-trained encoder, determine the sample text content associated with the sample viewpoint.

[0194] Step 803: Input the sample viewpoint and the sample text content associated with the sample viewpoint into the pre-trained encoder to determine the first sample encoding sequence.

[0195] Step 804: After processing the first sample encoding sequence and the sample viewpoint based on the multi-level attention mechanism, determine the classification result corresponding to the sample viewpoint based on the second classifier.

[0196] Step 805: Based on the classification results and labels corresponding to the sample viewpoints, fine-tune the pre-trained encoder to obtain a trained encoder.

[0197] In step 802 above, when determining the sample text content associated with the sample viewpoint based on the pre-trained encoder, it can be similar to the steps for determining the first text content and the second text content described above.

[0198] Specifically, it may include the following steps:

[0199] Step a1: Determine the sample keywords contained in the sample viewpoint, and then initiate a search based on the sample keywords to determine the first associated sample document corresponding to the sample viewpoint;

[0200] Step a2: Based on the pre-trained encoder, determine the first sample text content in the first associated sample document that corresponds to the sample viewpoint (this step is similar to step 102).

[0201] Step a3: Determine the target sample text content in the first sample text content, and the target sample document where the target sample text content is located;

[0202] Step a4: Based on the pre-trained encoder, determine the second sample text content in the target sample document that corresponds to the sample viewpoint (this step is similar to step 104).

[0203] The first sample text content and the second sample text content are the sample text content associated with the sample viewpoint.

[0204] In step 805 above, when fine-tuning the pre-trained encoder based on the classification result and the label corresponding to the sample viewpoint, for example, the loss value (e.g., cross-entropy loss value) in this fine-tuning process can be calculated based on the classification result and the label corresponding to the sample viewpoint, and if the training cutoff condition is not met, the process returns to step 801.

[0205] While fine-tuning the pre-trained encoder, the second classifier can also be adjusted to train the second classifier synchronously.

[0206] By fine-tuning the pre-trained encoder, the encoder can better distinguish the relationship between a sample viewpoint (equivalent to the premise in the pre-training process) and multiple sample text contents (equivalent to the arguments in the pre-training process), resulting in more accurate encoding results.

[0207] III. Training process of feature fusion network.

[0208] The feature fusion network consists of a K-layer multilayer perceptron (MLP) and an activation function, such as a sigmoid function, and K can be 2.

[0209] Generally, the training process of the feature fusion network is carried out after the pre-trained encoder has completed fine-tuning.

[0210] For example, the training process of the feature fusion network may include the following steps:

[0211] Step b1: Obtain sample documents and sample viewpoints, as well as the tags corresponding to the sample documents. The tags corresponding to the sample documents are used to indicate whether each text content in the sample documents is evidence corresponding to the sample viewpoint, that is, whether it can be used to verify the authenticity of the sample viewpoint.

[0212] Step b2: Input the sample viewpoint, the text content of each sample document, and the title of the sample document into the fine-tuned encoder to determine the second sample encoding sequence.

[0213] Step b3: Input the second sample encoding sequence into the feature fusion network to be trained to determine the sample relevance between each text content in the sample document and the sample viewpoint.

[0214] Specifically, the second sample encoding sequence includes an embedded representation of the sample viewpoint, an embedded representation of each text content in the sample document, and an embedded representation of the title of the sample document.

[0215] After inputting the second sample encoded sequence into the feature fusion network to be trained, the various embedding representations can be fused. For example, the fusion can be performed using the following formula:

[0216] E ts =W a [E t E s E t -E s E t ⊙E s (1)

[0217] Among them, E t E represents the embedded representation of the sample title. s E represents the embedded representation of the text content in the sample document. ts W represents the first fusion feature that combines the features of the sample title and text content. a These are the parameters to be trained in the feature fusion network, used to align the embedded representation.

[0218] For each text content, the fusion feature between the text content and the sample title is calculated using the above formula (1).

[0219] E cts =W a '[E c E ts E c -E ts E c ⊙E ts (2)

[0220] Among them, E c E represents the embedding representation of the sample viewpoint. cts W represents the second fusion feature, which integrates features from the sample title, text content, and sample viewpoint. a ' represents the parameters to be trained in the feature fusion network, used to align the embedded representation.

[0221] After obtaining the second fusion feature, the sample relevance can be calculated, for example, using the following formula:

[0222]

[0223] in, This represents the correlation between samples, Sigmoid is the activation function, and MLP stands for Multilayer Perceptron.

[0224] Step b4: Train the feature fusion network to be trained based on the sample relevance and the tags corresponding to the sample documents.

[0225] In one possible implementation, when training the feature fusion network to be trained based on the sample relevance and the labels corresponding to the sample documents, the loss value during the current training process can be calculated first based on the sample relevance and the labels corresponding to the sample documents, and then the process can return to step b1 if the training cutoff condition is not met.

[0226] For example, when calculating the loss value, it can be calculated using the following formula:

[0227]

[0228] Among them, L E The loss value is represented by m, which represents the number of samples in the batch, and n is the number of samples in the batch. i Indicates the text content number, y ij y represents the tag of the j-th text content in the i-th document. y is 1 when the j-th text content is evidence corresponding to the sample viewpoint, and y is 0 when the j-th text content is not evidence corresponding to the sample viewpoint.

[0229] In the opinion verification method provided in the above embodiments, a search can be initiated based on the opinion to be verified to determine a coarse document (i.e., a first document) related to the opinion to be verified. Then, a refined document (i.e., a second document) related to the opinion to be verified can be determined through the first document. The authenticity of the opinion to be verified can then be verified through the text content related to the opinion to be verified in the first and second documents. Thus, through multiple rounds of iterative retrieval, a comprehensive verification of the opinion to be verified can be achieved, improving the accuracy and efficiency of the verification.

[0230] Based on the same concept, this disclosure also provides a data publishing method, including the following steps:

[0231] Step c1: Obtain the data to be published.

[0232] Step c2: Treat the data to be published as a viewpoint to be verified, and detect the authenticity of the data to be published based on the above embodiments.

[0233] Step c3: If the data to be published is real data, then publish the data to be published.

[0234] This method allows for the verification of data authenticity before its release, ensuring the publication of accurate data and enhancing the security of information.

[0235] Alternatively, based on the viewpoint verification method provided in this disclosure, the authenticity of data published on various platforms can be verified, and the verification results of each published data can be marked on the corresponding data (or marked on the preview information of the corresponding data), thereby enabling security reminders to users.

[0236] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0237] Based on the same inventive concept, this disclosure also provides an opinion verification device corresponding to the opinion verification method. Since the principle of the device in this disclosure for solving the problem is similar to that of the opinion verification method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0238] Reference Figure 9 The diagram shown is an architectural schematic of an opinion verification device provided in an embodiment of this disclosure. The device includes: a first acquisition module 901, a first determination module 902, a second determination module 903, a third determination module 904, a verification module 905, and a training module 906; wherein,

[0239] The first acquisition module 901 is used to acquire the viewpoint to be verified, initiate a search based on the viewpoint to be verified, and determine the first document corresponding to the viewpoint to be verified.

[0240] The first determining module 902 is used to determine the first text content in the first document that is associated with the viewpoint to be verified.

[0241] The second determining module 903 is used to determine the first target document where the target text content in the first text content is located, and to determine the associated document of the first target document;

[0242] The third determining module 904 is used to determine the second text content in the associated document that is associated with the viewpoint to be verified, based on the viewpoint to be verified and the target text content;

[0243] The verification module 905 is used to verify the authenticity of the viewpoint to be verified based on the first text content and the second text content.

[0244] In one possible implementation, the first acquisition module 901, when initiating a search based on the viewpoint to be verified and determining the first document corresponding to the viewpoint to be verified, is used to:

[0245] Identify the keywords contained in the viewpoint to be verified;

[0246] Initiate a search based on the keywords and identify the first document containing the keywords.

[0247] In one possible implementation, the first determining module 902, when determining the first text content in the first document associated with the viewpoint to be verified, is used to:

[0248] For any first document, the viewpoint to be verified, the text content in the first document, and the document title of the first document are encoded based on a pre-trained encoder to determine the first encoding sequence corresponding to the first document;

[0249] The first encoded sequence is input into a pre-trained feature fusion network to determine the first relevance between each text content in the first document and the viewpoint to be verified.

[0250] The first text content is determined based on the first relevance between the text content in each first document and the viewpoint to be verified.

[0251] In one possible implementation, the second determining module 903, when determining the first target document where the target text content in the first text content is located, is used to:

[0252] The first text content whose first relevance meets the preset condition is taken as the target text content, and the first target document where the target text content is located is determined;

[0253] The second determining module 903, when determining the associated document of the first target document, is used to:

[0254] Determine the associated document corresponding to the hyperlink in the first target document; or,

[0255] Identify associated documents whose relevance to the first target document exceeds a preset value.

[0256] In one possible implementation, the third determining module 904, when determining the second text content in the associated document associated with the viewpoint to be verified based on the viewpoint to be verified and the target text content, is used to:

[0257] For any associated document, the viewpoint to be verified, the target text content, the text content in the associated document, and the document title of the associated document are encoded based on a pre-trained encoder to determine the second encoding sequence corresponding to the associated document;

[0258] The second encoded sequence is input into a pre-trained feature fusion network to determine the second relevance between each text content in the associated document and the viewpoint to be verified.

[0259] The second text content is determined based on the second relevance between the text content in each associated document and the viewpoint to be verified.

[0260] In one possible implementation, the apparatus further includes a training module 906 for training the encoder according to the following method:

[0261] Multiple sets of sample data are obtained, each set of sample data includes sample premises and sample arguments, and each set of sample data has a corresponding label, which is used to represent the relationship between the sample premises and the sample arguments.

[0262] Based on the sample data, the encoder to be trained is pre-trained to obtain the pre-trained encoder;

[0263] The pre-trained encoder is fine-tuned to obtain a trained encoder.

[0264] In one possible implementation, the training module 906, when pre-training the encoder to be trained based on the sample data to obtain a pre-trained encoder, is used to:

[0265] The sample data is input into the encoder to be trained to obtain the third encoding sequence corresponding to the sample data;

[0266] The third encoded sequence is classified based on the first classifier to obtain the predicted relationship corresponding to the sample data;

[0267] Based on the prediction relationship corresponding to the sample data and the label corresponding to the sample data, the encoder to be trained is pre-trained to obtain the pre-trained encoder.

[0268] In one possible implementation, the training module 906, when fine-tuning the pre-trained encoder to obtain a trained encoder, is used to:

[0269] Obtain sample viewpoints and corresponding tags for the sample viewpoints, wherein the tags for the sample viewpoints are used to characterize the authenticity of the sample viewpoints;

[0270] Based on the pre-trained encoder, the sample text content associated with the sample viewpoint is determined;

[0271] The sample viewpoint and the associated sample text content are input into the pre-trained encoder to determine the first sample encoding sequence;

[0272] After processing the first sample encoding sequence and the sample viewpoint based on a multi-level attention mechanism, the classification result corresponding to the sample viewpoint is determined based on the second classifier;

[0273] Based on the classification results and labels corresponding to the sample viewpoints, the pre-trained encoder is fine-tuned to obtain a trained encoder.

[0274] In one possible implementation, the verification module 905, when verifying the authenticity of the viewpoint to be verified based on the first text content and the second text content, is used to:

[0275] The first text content, the second text content, and the viewpoint to be verified are input into a pre-trained encoder to obtain a fourth encoding sequence;

[0276] The fourth encoding sequence and the viewpoint to be verified are processed based on a multi-level attention mechanism to obtain the fifth encoding sequence;

[0277] The fifth encoded sequence is classified based on a pre-trained second classifier to determine the authenticity of the viewpoint to be verified.

[0278] Based on the same inventive concept, this disclosure also provides a data publishing device corresponding to the data publishing method. Since the principle of the device in this disclosure for solving the problem is similar to the data publishing method described above in this disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0279] Reference Figure 10 The diagram shown is an architectural schematic of a data publishing device provided in an embodiment of this disclosure. The device includes: a second acquisition module 1001, a detection module 1002, and a publishing module 1003; wherein,

[0280] The second acquisition module 1001 is used to acquire data to be published.

[0281] The detection module 1002 is used to take the data to be published as a viewpoint to be verified, and to detect the authenticity of the data to be published based on the viewpoint verification method described in the first aspect or any possible implementation of the first aspect.

[0282] The publishing module 1003 is used to publish the data to be published when the data to be published is real data.

[0283] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0284] Based on the same technical concept, this disclosure also provides a computer device. (See also...) Figure 11 The diagram shows the structure of a computer device 1100 provided in this embodiment of the present disclosure, including a processor 1101, a memory 1102, and a bus 1103. The memory 1102 stores execution instructions and includes main memory 11021 and external memory 11022. The main memory 11021, also called internal memory, is used to temporarily store computational data in the processor 1101 and data exchanged with external memory 11022 such as a hard disk. The processor 1101 exchanges data with the external memory 11022 through the main memory 11021. When the computer device 1100 is running, the processor 1101 and the memory 1102 communicate through the bus 1103, causing the processor 1101 to execute the following instructions:

[0285] Obtain the viewpoint to be verified, initiate a search based on the viewpoint to be verified, and determine the first document corresponding to the viewpoint to be verified;

[0286] Determine the first text content in the first document that is associated with the viewpoint to be verified;

[0287] Determine the first target document containing the target text content in the first text content, and determine the associated document of the first target document;

[0288] Based on the viewpoint to be verified and the target text content, determine the second text content in the associated document that is associated with the viewpoint to be verified;

[0289] The authenticity of the viewpoint to be verified is verified based on the first text content and the second text content.

[0290] Alternatively, processor 1101 can execute the following instructions:

[0291] Get the data to be published;

[0292] The data to be published is used as a viewpoint to be verified, and the authenticity of the data to be published is detected based on the viewpoint verification method described in the above embodiments;

[0293] If the data to be published is real data, then the data to be published will be published.

[0294] This disclosure also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs the steps of the viewpoint verification method and data publishing method described in the above-described method embodiments. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0295] This disclosure also provides a computer program product carrying program code. The program code includes instructions that can be used to execute the steps of the viewpoint verification method and data publishing method described in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.

[0296] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0297] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this disclosure, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0298] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0299] In addition, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0300] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0301] Finally, it should be noted that the above-described embodiments are merely specific implementations of this disclosure, used to illustrate the technical solutions of this disclosure, and not to limit it. The protection scope of this disclosure is not limited thereto. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this disclosure. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be covered within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the protection scope of the claims.

Claims

1. A method for verifying an opinion, characterized in that, include: Obtain the viewpoint to be verified, initiate a search based on the viewpoint to be verified, and determine the first document corresponding to the viewpoint to be verified; Determine the first text content in the first document that is associated with the viewpoint to be verified; Determine the first target document containing the target text content in the first text content, and determine the associated document of the first target document; Based on the viewpoint to be verified and the target text content, determine the second text content in the associated document that is associated with the viewpoint to be verified; The authenticity of the viewpoint to be verified is verified based on the first text content and the second text content.

2. The method according to claim 1, characterized in that, The step of initiating a search based on the viewpoint to be verified, and determining the first document corresponding to the viewpoint to be verified, includes: Identify the keywords contained in the viewpoint to be verified; Initiate a search based on the keywords and identify the first document containing the keywords.

3. The method according to claim 1, characterized in that, Determining the first text content in the first document associated with the viewpoint to be verified includes: For any first document, the viewpoint to be verified, the text content in the first document, and the document title of the first document are encoded based on a pre-trained encoder to determine the first encoding sequence corresponding to the first document; The first encoded sequence is input into a pre-trained feature fusion network to determine the first relevance between each text content in the first document and the viewpoint to be verified. The first text content is determined based on the first relevance between the text content in each first document and the viewpoint to be verified.

4. The method according to claim 3, characterized in that, Determining the first target document containing the target text content within the first text content includes: The first text content whose first relevance meets the preset condition is taken as the target text content, and the first target document where the target text content is located is determined; The step of determining the associated documents of the first target document includes: Determine the associated document corresponding to the hyperlink in the first target document; or, Identify associated documents whose relevance to the first target document exceeds a preset value.

5. The method according to claim 1, characterized in that, The step of determining the second text content in the associated document that is associated with the viewpoint to be verified based on the viewpoint to be verified and the target text content includes: For any associated document, the viewpoint to be verified, the target text content, the text content in the associated document, and the document title of the associated document are encoded based on a pre-trained encoder to determine the second encoding sequence corresponding to the associated document; The second encoded sequence is input into a pre-trained feature fusion network to determine the second relevance between each text content in the associated document and the viewpoint to be verified. The second text content is determined based on the second relevance between the text content in each associated document and the viewpoint to be verified.

6. The method according to claim 3 or 5, characterized in that, The method further includes training the encoder according to the following method: Multiple sets of sample data are obtained, each set of sample data includes sample premises and sample arguments, and each set of sample data has a corresponding label, which is used to represent the relationship between the sample premises and the sample arguments. Based on the sample data, the encoder to be trained is pre-trained to obtain the pre-trained encoder; The pre-trained encoder is fine-tuned to obtain a trained encoder.

7. The method according to claim 6, characterized in that, The step of pre-training the encoder to be trained based on the sample data to obtain a pre-trained encoder includes: The sample data is input into the encoder to be trained to obtain the third encoding sequence corresponding to the sample data; The third encoded sequence is classified based on the first classifier to obtain the predicted relationship corresponding to the sample data; Based on the prediction relationship corresponding to the sample data and the label corresponding to the sample data, the encoder to be trained is pre-trained to obtain the pre-trained encoder.

8. The method according to claim 7, characterized in that, The step of fine-tuning the pre-trained encoder to obtain a trained encoder includes: Obtain sample viewpoints and corresponding tags for the sample viewpoints, wherein the tags for the sample viewpoints are used to characterize the authenticity of the sample viewpoints; Based on the pre-trained encoder, the sample text content associated with the sample viewpoint is determined; The sample viewpoint and the associated sample text content are input into the pre-trained encoder to determine the first sample encoding sequence; After processing the first sample encoding sequence and the sample viewpoint based on a multi-level attention mechanism, the classification result corresponding to the sample viewpoint is determined based on the second classifier; Based on the classification results and labels corresponding to the sample viewpoints, the pre-trained encoder is fine-tuned to obtain a trained encoder.

9. The method according to claim 1, characterized in that, The verification of the authenticity of the viewpoint to be verified based on the first text content and the second text content includes: The first text content, the second text content, and the viewpoint to be verified are input into a pre-trained encoder to obtain a fourth encoding sequence; The fourth encoding sequence and the viewpoint to be verified are processed based on a multi-level attention mechanism to obtain the fifth encoding sequence; The fifth encoded sequence is classified based on a pre-trained second classifier to determine the authenticity of the viewpoint to be verified.

10. A data publishing method, characterized in that, include: Get the data to be published; The data to be published is taken as a viewpoint to be verified, and the authenticity of the data to be published is detected based on the viewpoint verification method described in any one of claims 1 to 9; If the data to be published is real data, then the data to be published will be published.

11. A viewpoint verification device, characterized in that, include: The first acquisition module is used to acquire the viewpoint to be verified, initiate a search based on the viewpoint to be verified, and determine the first document corresponding to the viewpoint to be verified. The first determining module is used to determine the first text content in the first document that is associated with the viewpoint to be verified. The second determining module is used to determine the first target document where the target text content in the first text content is located, and to determine the associated document of the first target document; The third determining module is used to determine, based on the viewpoint to be verified and the target text content, the second text content in the associated document that is associated with the viewpoint to be verified; The verification module is used to verify the authenticity of the viewpoint to be verified based on the first text content and the second text content.

12. A data publishing device, characterized in that, include: The second acquisition module is used to acquire the data to be published; The detection module is used to take the data to be published as a viewpoint to be verified, and to detect the authenticity of the data to be published based on the viewpoint verification method described in any one of claims 1 to 9. The publishing module is used to publish the data to be published if the data to be published is real data.

13. A computer device, characterized in that, include: The computer device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the view verification method as described in any one of claims 1 to 9, or the steps of the data publishing method as described in claim 10.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the viewpoint verification method as described in any one of claims 1 to 9, or the steps of the data publishing method as described in claim 10.

Images