Intelligent calibration news content verification method

Abstract

The invention relates to the technical field of news content verification, in particular to an intelligently calibrated news content verification method, which comprises the following steps of: dividing news content to obtain a plurality of target texts; determining a semantic analysis result at a position adjacent to the target text, and obtaining a plurality of target keywords from the semantic analysis result; performing similarity analysis on the target keywords, determining the similarity of the target keywords, checking the emotional tendency of the target keywords, and setting word use probability density corresponding to the target keywords; based on the word use probability density of the target keyword, verifying vocabularies in the target text to obtain a first verification result of news verification; vocabularies matched with the first verification result are obtained from the news content, and a second verification result is generated; comparing vocabularies in the first verification result with vocabularies in the second verification result, and determining a difference result of the news content before and after modification; and the news verification efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of news content verification, and specifically, to an intelligent calibration method for news content verification. Background Art

[0002] With the development of NLP technology, pre-trained models such as BERT and Word2Vec have been able to achieve excellent performance in various tasks. These models can capture the complex relationships between words, providing strong support for semantic analysis and sentiment analysis. News content verification is first processed based on a keyword matching filtering method. The principle is as follows: First, prepare a keyword library, which records some pre-selected keywords. When judging a news content, if there is one or more keywords that can match the text content in the news content, that is, a keyword appears in the text, then it is judged whether this web page belongs to the information that needs to be controlled. If so, this news content is filtered; otherwise, it means that this web page is not the target to be filtered. However, this processing will ignore the low-frequency texts that appear in the news content and the replacement words corresponding to these low-frequency texts, resulting in easy processing errors when calibrating and verifying low-frequency words.

[0003] For example, Chinese Patent Publication No. CN114943285A discloses an intelligent audit system for Internet news content data. The system includes: a segmentation unit configured to first perform a first semantic analysis on the news content data to be audited to establish each paragraph with independent semantics in the news content data to be audited, perform paragraph division, and then perform segmentation according to the result of the paragraph division to obtain a plurality of segmented contents; a keyword extraction unit configured to perform a second semantic analysis on each segmented content to establish the keywords of each segmented content. The present invention realizes the intelligence of news content data audit by semantically segmenting the news content to be audited, then converting it into an image for similarity analysis, finding the keywords of the news content to be audited, and performing sensitivity analysis on these keywords, and has the advantages of high accuracy and high efficiency.

[0004] The prior art mainly completes the audit of news content after performing sensitive analysis on the text. However, when there are some low-frequency words with few occurrences and the paraphrases in certain sentence patterns and paragraphs change, it is difficult to make adaptive adjustments to these problems, resulting in a reduction in the efficiency of news audit processing. Summary of the Invention

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: an intelligent calibration method for news content verification, including: S1, obtaining the news content to be verified from the information library, dividing the news content, and obtaining a plurality of target texts.

[0006] S2. Segment and recognize the target text, determine the semantic analysis results at adjacent positions of the target text, and obtain multiple target keywords from the semantic analysis results.

[0007] S3. Conduct similarity analysis on the target keywords, determine the similarity of the target keywords, check the sentiment tendency of the target keywords, and set the word probability density corresponding to the target keywords.

[0008] S4. Based on the word probability density of the target keywords, verify the words in the target text to obtain the first verification result of news verification.

[0009] S5. Based on the obtained first verification result, obtain the words in the news content that match the first verification result and generate the second verification result.

[0010] S6. Compare the words in the first verification result and the second verification result, determine the difference result of the news content before and after modification, and judge whether the news content verification is completed according to the difference result.

[0011] The beneficial effects of the present invention are as follows: First, by dividing, segmenting and recognizing, and semantically analyzing the news content, the present invention ensures that each target text can be analyzed in detail. By constructing a news similarity matrix, the system can capture the internal connections between news contents, thereby more accurately identifying key information.

[0012] Second, by considering the word probability density of words, and combining the semantic similarity score and sentiment tendency score, the present invention makes the evaluation of target keywords more comprehensive. This multi-dimensional evaluation method can better reflect the actual meaning and role of words in a specific context.

[0013] Third, based on the word probability density and similarity metric, the system can automatically select the optimal first verification result and further generate the second verification result. This method ensures the scientificity and rationality of the verification process and provides data support for subsequent improvement at the same time.

[0014] Fourth, by comparing the first verification result and the second verification result, calculating the difference in similarity scores, type classification, and domain relevance index before and after word replacement, the system can generate a detailed difference result report. This not only helps editors understand the effect of modification but also provides a reference for future optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention will be further described below with reference to the drawings and embodiments.

[0016] Figure 1 It is a schematic flowchart of a news content verification method with intelligent calibration.

[0017] Figure 2 It is a schematic flowchart of step S1 of an intelligent calibration news content verification method.

[0018] Figure 3 It is a schematic flowchart of step S2 of an intelligent calibration news content verification method.

[0019] Figure 4 It is a schematic flowchart of step S4 of an intelligent calibration news content verification method.

[0020] Figure 5 It is a schematic flowchart of step S5 of an intelligent calibration news content verification method.

[0021] Figure 6 It is a schematic flowchart of step S6 of an intelligent calibration news content verification method. Detailed implementation manners

[0022] The embodiments of the present invention will be described in detail below. The described embodiments are exemplary and are only used to explain the present invention and should not be construed as a limitation to the present invention. For those not specified in the embodiments, the techniques or conditions described in the literature in this field or according to the product specifications are followed.

[0023] Refer to Figure 1 , an intelligent calibration news content verification method, includes: S1, obtaining the news content to be verified from the information library, dividing the news content, and obtaining multiple target texts.

[0024] S2, performing segmentation recognition on the target text, determining the semantic analysis results at adjacent positions of the target text, and obtaining multiple target keywords from the semantic analysis results.

[0025] S3, performing similarity analysis on the target keywords, determining the similarity of the target keywords, checking the sentiment tendency of the target keywords, and setting the word probability density corresponding to the target keywords.

[0026] S4, verifying the vocabulary in the target text based on the word probability density of the target keywords to obtain the first verification result of news verification.

[0027] S5, based on the obtained first verification result, obtaining the vocabulary matching the first verification result from the news content and generating a second verification result.

[0028] S6, comparing the vocabulary in the first verification result with that in the second verification result, determining the difference result of the news content before and after modification, and judging whether the news content verification is completed according to the difference result.

[0029] In an embodiment of the present invention, the steps mainly divide the news content to generate multiple target texts for subsequent processing; when obtaining the target texts, it is necessary to first determine the type of the news to be verified, and extract the target texts related to the current news from the title, subtitle, body paragraphs, quotes, and picture captions in the news. The extracted target texts will be extracted according to a specific size and number of characters, and each target text will contain complete semantic information.

[0030] As Figure 2 shown, step S1 further includes: S11, constructing a news similarity matrix based on the news content to be verified; it is necessary to convert the news content into news vectors and calculate the similarity scores between each pair of news vectors to generate a similarity matrix. When generating news vectors, the Word2Vec and BERT models can be used for processing to obtain the news vectors required at this time. The similarity score represents the cosine similarity between the news vectors at this time; the news vectors output by the Word2Vec and BERT models tend to describe the distribution probabilities of the corresponding words and sentences in the news content in the title, subtitle, body paragraphs, quotes, and picture captions of the news. The cosine similarity calculated using these values will represent the relative distribution between the news vectors.

[0031] S12, obtaining the similar nodes of the news similarity matrix on the news content and obtaining the connection paths of the similar nodes; only when the similarity score between the news vectors is greater than the similarity threshold will they be considered similar nodes; after the similar nodes are identified, all the similar nodes are connected to find the connection paths of the similar nodes. The similarity threshold can adopt the average value of the similarity scores of the news vectors in the historical data. Finally, a part of the texts with similarity scores greater than the average level are selected, and these texts will highlight the impact of the similar parts of the news content on news verification.

[0032] S13, combining the connection paths of the similar nodes according to the weights of each similar node on the connection path to obtain the local similar text of the news, and outputting the paragraphs corresponding to the local similar text as the target text. For each connection path, the similarity scores on the connection path are accumulated as the weight of the connection path, and the similar nodes are sorted according to the weights of the connection paths. The news content segments with high weights are selected as part of the local similar text, and then the coherent paragraphs and sentences corresponding to the local similar text are output to obtain the target text.

[0033] In an embodiment of the present invention, step S2 mainly obtains the semantic analysis results at the adjacent positions of the target text and obtains the target keywords from the semantic analysis results; subsequent processing is carried out according to these obtained target keywords.

[0034] The semantic analysis results at adjacent positions of the target text are based on whether the semantics change between texts or paragraphs that are close to the target text, and whether these semantics are continuous. Then, the required keywords are filtered out from the text corresponding to the semantic analysis results. When obtaining the target keywords, it is necessary to perform a reconstruction analysis on the target text according to the semantic analysis results to obtain multiple target keywords with the smallest loss rate after the reconstruction analysis at this time. Based on these target keywords, the subsequent analysis of the similarity of the target keywords is continued.

[0035] When performing segmentation and recognition on the target text, the target text is segmented in the form of short sentences, and the semantic analysis results at adjacent positions of the target text are calculated. The semantic analysis results tend to identify the sentence structure of the target text and the understanding of the overall grammar by individual words to capture the semantic relationships of relevant words at adjacent positions of the target text.

[0036] Such as Figure 3 shown, step S2 further includes: S21, after segmenting the target text, obtain the sentence structure corresponding to the target text.

[0037] S22, identify the adjacent words of the target text under the corresponding sentence structure, and perform semantic recognition on the adjacent words according to the adjacent positions of the target text to obtain the semantic analysis results at adjacent positions of the target text; the semantic analysis performed at this time is to identify the meanings of the target text at adjacent positions, and use these target texts with recognized meanings as the semantic analysis results at this time.

[0038] S23, perform a reconstruction analysis on the target text according to the semantic analysis results, and output the target keywords with the smallest loss rate after the reconstruction analysis. The reconstruction analysis is to shuffle and reorganize the target text to determine whether the semantics of these texts after reorganization are the same as the original text. By calculating the cosine similarity between the target texts after reorganization and comparing the cosine similarity with the preset similarity threshold, the loss rate at this time is obtained. When the loss rate is the smallest, it means that the information fidelity between the target keyword set and the original text is the highest after the target text is reorganized, that is, as much semantic information of the original text as possible is retained, while unnecessary redundancy or errors are reduced.

[0039] In one embodiment of the present invention, step S3 is inclined to process the similarity score and sentiment tendency score of the target keyword to set the word usage probability density corresponding to the target keyword. The similarity score of the target keyword is consistent with the similarity score calculated above, and both are calculated using the pre-similarity. After obtaining the similarity score, the sentiment tendency of the target keyword is identified, and the word usage probability density is set according to the values of the target keyword similarity score and sentiment tendency that appear in each word. The sentiment tendency can adopt the technology of the sentiment dictionary, comparing the target keyword with the sentiment dictionary to obtain a sentiment tendency score, which is used as the subsequent word usage probability density for setting the target keyword.

[0040] For example, each target keyword has a similarity score and a sentiment tendency score. The relative ratio of multiple target keywords before and after the currently identified target keyword is used to obtain the word usage probability density of the current target keyword.

[0041] Therefore, the word usage probability density can be expressed as follows: obtain the context information corresponding to the target keyword, and obtain the occurrence probability of the target keyword under the corresponding context information; calculate the similarity score and sentiment tendency score of the target keyword, and combine the occurrence probability of the target keyword under the corresponding context information with the similarity score and sentiment tendency score of the target keyword to obtain the word usage probability density of the target keyword.

[0042] Among them, P′(w i |C) represents the word usage probability density of the i-th target keyword, n represents the number of target keywords, the value range of i is from 1 to n, k represents the front and back offset value of the current target keyword, which is used to represent the relative situation of the previous several words of the currently selected target keyword to identify the current target keyword, and the value of k will be set to 2 or 3 according to requirements; j represents the number of the target keyword, the value range of j is from i - k to i + k. When i - k is less than 1, the corresponding number is not selected, and the value of j starts from 1. When i + k is greater than n, i + k is equal to n; P(w i |C) represents the occurrence probability of the i-th target keyword under the corresponding context information, C represents the context information corresponding to the target keyword in the news content; S(w j ,w i ) represents the similarity score between the i-th word and the j-th word, E(w j ) represents the sentiment tendency score of the j-th word, α represents the weight coefficient of similarity, and β represents the weight coefficient of sentiment tendency. The value range of the weight coefficients for similarity and sentiment tendency is set between 0 and 1. For example, in the order of similarity and sentiment tendency, these two weight coefficients are set to 0.7 and 0.3.

[0043] In one embodiment of the present invention, step S4 mainly relies on the word probability density of the target keyword and uses this word probability density to search for relevant words in the target text, thereby obtaining the first verification result of news verification.

[0044] In the first verification result, preference is given to feedback testing of the target text based on the obtained word probability density. For example, the target text is compared with the word corresponding to the minimum word probability density in the target keyword, and the group of words with the highest similarity to this word in the target text is selected and set as the first verification result. The purpose of finding the first verification result in this way is to discover the words with the lowest word probability density in the news text, which have a very low probability of appearing in the article. Then, after finding whether there are words very similar to it, the word with the lowest word probability density and the word with the highest similarity to this word are marked to determine whether this word needs to be replaced in subsequent verification to make the text more appropriate and accurate.

[0045] Such as Figure 4 shown, step S4 includes: S41, comparing the target keyword corresponding to the minimum word probability density with the target text to obtain the text similarity measure between the target keyword corresponding to the minimum word probability density and the target text; the text similarity measure refers to calculating the cosine similarity between the target text and this word when the word probability density of the target keyword takes the minimum value, as the text similarity measure at this time.

[0046] S42, when the text similarity measure takes the maximum value, the corresponding target text is used as the first verification result.

[0047] When obtaining the first verification result, a loop check method will be adopted to continuously check the target keyword with the lowest word probability density and the words corresponding to the target keyword. After the check is completed, the corresponding words will be removed, and the loop detection will continue until the words in the news content are all checked.

[0048] In one embodiment of the present invention, step S5 prefers to obtain words in the news content that match the first verification result and select the second verification result. Compared with the first verification result, the second verification result takes into account whether there is content in the news content that has a high domain relevance to the words in the first verification result. The domain relevance will be set according to the number of occurrences of the words existing in the first verification result in the news content. Finally, an index corresponding to the domain relevance is obtained, and this index is used to select the second verification result corresponding to each word in the first verification result. The word corresponding to the maximum value of the domain relevance for each word in the first verification result is set as the second verification result.

[0049] Such as Figure 5As shown in the figure, step S5 includes: S51, obtaining the occurrence probability of the first verification result in the news content and determining the occurrence probability of the words other than the first verification result in the news content.

[0050] S52, taking the occurrence probability of the first verification result in the news content and the occurrence probability of the words other than the first verification result in the news content as basic conditions, constructing a scatter plot. The abscissa of the scatter plot is the corresponding occurrence probability value in the basic conditions, and the ordinate is set as the weight of the corresponding word in the news content in the basic conditions. At this time, the weight will set a weight score according to the words in the first verification result and the words in the news content, and this score will be used as the weight at this time according to the usage frequency of the corresponding words in the historical data.

[0051] S53, dividing the coordinate intervals of the scatter plot, determining the probability that the scatter points corresponding to the first verification result in the scatter plot fall into the coordinate intervals and the probability of the number of scatter points falling into the coordinate intervals; the probability of the number of scatter points indicates how many scatter points may exist in the coordinate interval where the scatter point corresponding to the word of the first verification result falls when regarded as scatter point division; setting the consistency score of the first verification result.

[0052] S54, adjusting the coordinate intervals according to the corresponding ratio, determining the probability change value that the scatter points corresponding to the first verification result fall into the coordinate intervals under the corresponding adjustment ratio, and setting the variability score of the first verification result.

[0053] S55, setting the domain relevance index of the first verification result according to the consistency score and variability score of the first verification result, and setting the news content corresponding to the maximum value of the domain relevance index as the second verification result.

[0054] The consistency score is expressed as: Among them, χ 1 represents the consistency score, P b represents the probability that the scatter points corresponding to the first verification result fall into the coordinate intervals, and P np represents the probability of the number of scatter points that the scatter points corresponding to the first verification result fall into the coordinate intervals.

[0055] The variability score is expressed as Among them, χ 2 represents the variability score, AP represents the adjustment ratio, P(b|AP) represents the probability change value that the scatter points corresponding to the first verification result fall into the coordinate intervals under the corresponding adjustment ratio, e represents the exponential constant, and λ represents the attenuation coefficient. The attenuation coefficient is set to 1.679 at this time to control the influence of the probability change value by the adjustment ratio.

[0056] The finally set domain relevance index is expressed as: Among them, χ fin represents the domain relevance index, and π represents the pi.

[0057] At this time, it is to judge the news content existing in the corresponding coordinate interval in the scenario where the domain relevance index can take the maximum value, and these news contents will be output as the second verification result.

[0058] In an embodiment of the present invention, step S6 determines the differences in the content of the first verification result and the second verification result before and after the modification of the news content to judge whether the news content verification is completed.

[0059] When comparing the first verification result and the second verification result, mainly compare the words existing in the first verification result and the second verification result in the form of similarity scores, etc., determine whether there is a positive correlation among these verified words, and what form of difference level the difference values between the corresponding scores of these words present. These difference levels will be output as the difference results, and finally know the specific verification situation of the current news content.

[0060] When comparing the words in the first verification result and the second verification result, first judge whether the words in the first verification result and the second result can match, and then, according to their corresponding mapping relationships, classify the corresponding contents in the original sentences of these words to determine what type of text content these words should be regarded as after matching. These contents will reflect the situation of the current news content during verification, and at the same time obtain the performance forms of this content before and after the news modification, and take the difference between the corresponding similarity scores before and after the news modification as the difference result at this time. Then, according to this difference result and the corresponding type of text content, output these contents to understand whether the verification work for the news content is completed.

[0061] As Figure 6 shown, the implementation method of step S6 is: S61, construct the corresponding word mapping relationship between the first verification result and the second verification result to obtain the mapping comparison table corresponding to the first verification result and the second verification result. The word mapping relationship is the mapping relationship of the corresponding words in the first verification result and the second verification result at their original positions in the news content.

[0062] When constructing the mapping comparison table, a table will be generated according to the positions, original contents, modified contents, and word usage probability densities of the corresponding words in the first verification result and the second verification result. This table will display all the words in the first verification result and the second verification result to judge whether there are corresponding differences during modification and verification.

[0063] S62. Calculate the similarity score and the difference in similarity scores of the corresponding words in the first verification result and the second verification result in the news content before and after modification. The similarity score here is the difference in the similarity scores of the corresponding word pairs in the first verification result and the second verification result before and after modification to determine the difference in similarity generated at this time.

[0064] S63. According to the mapping comparison table corresponding to the first verification result and the second verification result, identify the literal interpretations of the news content before and after modification and set type classifications. The type classifications here will mark the purposes of the modifications made to the modified text. For example, classifications set for the meanings expressed by the text such as reducing emotional descriptions and increasing text resonance.

[0065] S64. Calculate the difference in the domain relevance indicators of the corresponding words in the first verification result and the second verification result before and after modification. According to the difference in similarity scores, type classifications, and domain relevance indicators of the corresponding words in the first verification result and the second verification result before and after modification, evaluate the corresponding words in the first verification result and the second verification result to generate a difference result.

[0066] When evaluating in step S64, use the difference in similarity scores, type classifications, and domain relevance indicators as the content of the evaluation. Convert the type classifications into the occurrence probabilities of the corresponding type classifications. After normalizing the difference in similarity scores and domain relevance indicators to enhance the impact caused by the differences, perform a weighted sum of the occurrence probabilities of the type classifications and the normalized difference in similarity scores and domain relevance indicators as the content for evaluating the corresponding words in the first verification result and the second verification result. Then, combine the score of this weighted sum, the mapping comparison table, type classifications, the difference in similarity scores, and the domain relevance indicators into a difference result for output.

[0067] Upon receiving this difference result, the external terminal can understand the specific situation of the current news content and the corresponding content before and after modification, facilitating subsequent review and auxiliary modification.

[0068] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations within the scope of the present invention and still be covered by the protection scope of the present invention.

Claims

1. An intelligently calibrated news content verification method, characterized in that: include: S1, obtaining news content to be verified from the information database, dividing the news content, and obtaining multiple target texts; S2, segmenting and recognizing the target text, determining the semantic analysis results at adjacent positions of the target text, and obtaining multiple target keywords from the semantic analysis results; S3, performing similarity analysis on the target keywords, determining the similarity of the target keywords, checking the sentiment tendency of the target keywords, and setting the word probability density corresponding to the target keywords; S4, based on the word probability density of the target keyword, verify the words in the target text to obtain the first verification result of the news verification; S5, based on the obtained first verification result, obtaining a word matching the first verification result from the news content, and generating a second verification result; S6, comparing the words in the first verification result with those in the second verification result, determining the difference between the news content before and after the modification, and judging whether the news content has been verified based on the difference.

2. According to claim 1, a method for verifying news content with intelligent calibration is characterized in that: Step S1 also includes: S11, constructing a news similarity matrix based on the news content to be verified; S12, obtaining similar nodes of the news similarity matrix on the news content, and obtaining connection paths of the similar nodes; S13, combining the connection paths of similar nodes according to the weight of each similar node on the connection paths to obtain local similar texts of the news, and outputting the paragraphs corresponding to the local similar texts as target texts.

3. The intelligently calibrated news content verification method according to claim 1 is characterized in that: Step S2 also includes: S21, after segmenting the target text, obtaining the sentence structure corresponding to the target text; S22, identifying adjacent words in the target text under the corresponding sentence structure, and performing semantic recognition on the adjacent words according to adjacent positions of the target text to obtain semantic analysis results at adjacent positions of the target text; S23, reconstructing and analyzing the target text according to the semantic analysis result, and outputting the target keyword with the smallest loss rate after the reconstruction and analysis.

4. The intelligently calibrated news content verification method according to claim 1, characterized in that: The word probability density can be expressed as follows: obtain the context information corresponding to the target keyword to obtain the probability of occurrence of the target keyword under the corresponding context information; calculate the similarity score and sentiment tendency score of the target keyword, combine the probability of occurrence of the target keyword under the corresponding context information with the similarity score and sentiment tendency score of the target keyword, and obtain the word probability density of the target keyword.

5. The intelligently calibrated news content verification method according to claim 4 is characterized in that: The word probability density is expressed as: Among them, P′(w i |C) represents the word probability density of the i-th target keyword, n represents the number of target keywords, i ranges from 1 to n, k represents the previous and next offset values ​​of the current target keyword, and j represents the number of the target keyword; P(w i |C) represents the probability of occurrence of the i-th target keyword in the corresponding context information, and C represents the context information corresponding to the target keyword in the news content; S(w j ,w i ) represents the similarity score between the i-th word and the j-th word, E(w j ) represents the sentiment tendency score of the jth word, α represents the weight coefficient of similarity, and β represents the weight coefficient of sentiment tendency.

6. The intelligently calibrated news content verification method according to claim 1, characterized in that: Step S4 includes: S41, comparing the target keyword corresponding to the minimum word probability density with the target text to obtain a text similarity measure between the target keyword corresponding to the minimum word probability density and the target text; S42, when the text similarity measure reaches the maximum value, the corresponding target text is used as the first verification result.

7. The intelligently calibrated news content verification method according to claim 1, characterized in that: Step S5 includes: S51, obtaining the occurrence probability of the first verification result in the news content, and determining the occurrence probability of words other than the first verification result in the news content; S52, constructing a scatter plot using the probability of occurrence of the first verification result in the news content and the probability of occurrence of words other than the first verification result in the news content as basic conditions; S53, dividing the scatter point coordinate graph into coordinate intervals, determining the probability that the scatter points corresponding to the first verification result in the scatter point coordinate graph fall into the coordinate interval and the probability of the number of scatter points falling into the coordinate interval, and setting the consistency score of the first verification result; S54, adjusting the coordinate interval according to a corresponding ratio, determining a probability change value of the scattered points corresponding to the first verification result falling into the coordinate interval under the corresponding adjustment ratio, and setting a variability score of the first verification result; S55, according to the consistency score and variability score of the first verification result, the domain relevance index of the first verification result is set, and the news content corresponding to the maximum value of the domain relevance index is set as the second verification result.

8. The intelligently calibrated news content verification method according to claim 7, characterized in that: The consistency score is expressed as: Among them, χ1 represents the consistency score, P b It represents the probability that the scattered points corresponding to the first verification result fall into the coordinate interval, P np Indicates the probability of the number of scattered points corresponding to the first verification result falling within the coordinate interval; The variability score is expressed as: Among them, χ2 represents the variability score, AP represents the adjustment ratio, P(b|AP) represents the probability change value of the scattered points corresponding to the first verification result falling into the coordinate interval under the corresponding adjustment ratio, e represents the exponential constant, and λ represents the attenuation coefficient.

9. The intelligently calibrated news content verification method according to claim 8, characterized in that: The domain relevance index is expressed as: Among them, χ fin represents the field relevance index, and π represents pi.

10. The intelligently calibrated news content verification method according to claim 7, characterized in that: Step S6 includes: S61, constructing a vocabulary mapping relationship between the first verification result and the second verification result, and obtaining a mapping comparison table between the first verification result and the second verification result; S62, calculating the similarity score and the similarity score difference of the corresponding words in the first verification result and the second verification result before and after the news content is modified; S63, identifying the textual interpretation of the news content before and after the modification according to the mapping comparison table corresponding to the first verification result and the second verification result, and setting the type classification; S64, calculate the difference in domain relevance index between the corresponding words in the first verification result and the second verification result before and after modification, evaluate the corresponding words in the first verification result and the second verification result according to the difference in similarity scores, type classification and domain relevance index between the corresponding words in the first verification result and the second verification result, and generate a differential result.

Images