Method for identifying whether authenticity of news is changed or not and related equipment

Abstract

The invention provides a method and related equipment for identifying whether the authenticity of news is changed or not, and relates to the technical field of news authenticity identification. Each piece of news in an obtained target news set is judged to obtain clarified news; extracting clarified data from the clarified news by utilizing the constructed content extraction cue word; taking the clarified data as retrieval content and / or a text of the clarified news to generate a retrieval vector, and retrieving news related to the retrieval content and / or the retrieval vector from the historical news to obtain a candidate news set; analyzing the content consistency and content conflict between all news in the candidate news set and the clarified news, judging whether news corresponding to the clarified news exists in the candidate news set or not, if the news corresponding to the clarified news exists in the candidate news set, considering that the authenticity of the news corresponding to the clarified news is changed, and otherwise, judging that the authenticity of the news corresponding to the clarified news is changed. Otherwise, considering that no news with changed authenticity exists in the candidate news set; and the authenticity identification capability of the news is improved.

Description

Technical Field

[0001] This invention relates to the field of news authenticity identification technology, and in particular to a method and related equipment for identifying whether the authenticity of news has been altered. Background Technology

[0002] In the information age, the Internet has become the main way for people to obtain information. Although multimedia information such as audio and video is developing rapidly, reading online news and articles is still an indispensable choice for most netizens.

[0003] Among the vast amount of internet news, there are often some news items with questionable authenticity. These news items can disturb readers and may even cause economic losses. Therefore, verifying the authenticity of news is an important task in the governance of the online environment.

[0004] Current mainstream fake news detection methods all aim to detect deliberately fabricated news. For example, existing technologies disclose a fake news detection method and system based on time interval and knowledge fusion. This method extracts features from news images and text separately, obtains external features from an external knowledge base, and considers the time difference between the occurrence and publication of the news event. It then integrates all features to determine whether the news is fake. Other existing technologies provide a multimodal fake information detection method that combines domain information. This method uses a large model to obtain domain information about the news and integrates textual and image features from the domain information into the classification to improve the discrimination effect. There are also publicly disclosed fake news detection inference methods and systems based on large models and pre-training. These methods extract image, text, and speech features separately, and then enhance them using a multimodal feature enhancement module after attention fusion to capture subtle differences and potential manipulation signs in the news, thereby improving the discrimination effect.

[0005] The aforementioned methods for detecting fake news have some ability to detect deliberately fabricated news, but they are completely incapable of identifying a specific type of news. When the authors of these news stories obtained certain information, that information was true, and they had no intention of deliberately fabricating facts when writing the news content. Apart from the core content, the supporting evidence in the news is also true. At the time of publication, this type of news is entirely true. However, as time goes on, the relevant parties involved in the news clarify or refute it, pointing out that the core content of the news is not true. At this point, this type of news becomes fake news. News that undergoes clarification or refutation will be referred to as "clarification news." Currently, all fake news detection methods determine the authenticity of news by looking for contradictions and errors within the news, and they are only capable of identifying news that is deliberately fabricated or distorts facts. Summary of the Invention

[0006] This invention provides a method and related equipment for identifying whether the authenticity of news has changed, with the aim of improving the ability to identify the authenticity of news.

[0007] To achieve the above objectives, the present invention provides a method for identifying whether the authenticity of news has been altered, comprising:

[0008] Step 1: Obtain the target news collection;

[0009] Step 2: Analyze each news item in the target news collection to obtain clarification news;

[0010] Step 3: Extract clarification data from the clarification news using the constructed content extraction prompts;

[0011] Step 4: Using the clarified data as the search content and / or using the text of the clarified news to generate a search vector, retrieve news related to the search content and / or search vector from historical news to obtain a candidate news set;

[0012] Step 5: Analyze the consistency and conflict of content between all news items in the candidate news set and the clarification news. Determine whether there is news in the candidate news set that corresponds to the clarification news. If there is news in the candidate news set that corresponds to the clarification news, it is considered that the authenticity of the news corresponding to the clarification news has changed. Otherwise, it is considered that there is no news in the candidate news set whose authenticity has changed.

[0013] Furthermore, step 2 includes:

[0014] Extract the title and first sentence of each news item in the target news collection as the content of the analysis;

[0015] Construct a negative feature set and use the negative features in the negative feature set to replace the analysis content;

[0016] Construct a positive feature set, use the positive features in the positive feature set to match the analyzed content, and use the news that successfully matches the analyzed content as clarification news.

[0017] Furthermore, step 3 includes:

[0018] The clarification news was pieced together to create content extraction prompts;

[0019] Input the extracted prompts into the large language model to generate clarification content for the clarification news;

[0020] Extract the main content, relevant subjects, relevant locations, and relevant figures from the clarification content of the news clarification to obtain clarification data.

[0021] Furthermore, step 4 includes:

[0022] Using clarification data as the search content and / or generating search vectors from the text of clarification news, and using discrete search methods and / or vector search methods, retrieve news related to the search content and / or search vectors from historical news to obtain a candidate news set.

[0023] Furthermore, using discrete retrieval methods and / or vector retrieval methods, news related to the search content and / or search vector is retrieved from historical news, resulting in a candidate news set, including:

[0024] Generate search results based on the clarified data;

[0025] The headline of the clarification news is concatenated with the body text of the clarification news to create a vector retrieval text.

[0026] Use a text embedding model to transform vector retrieval text into retrieval vectors;

[0027] By using search terms and / or search vectors, news related to the search terms and / or search vectors can be retrieved from historical news, resulting in a candidate news set.

[0028] Furthermore, prior to step 5, the process includes using relevant subjects, locations, and figures from the clarified data as key information to perform preliminary filtering of the news in the candidate newsletter:

[0029] The candidate news set is filtered to extract news from all relevant subjects, including those with clarified data, to obtain the first filtering result;

[0030] The first filter result is used to filter out news from all relevant locations that include the clarified data, resulting in the second filter result;

[0031] The second filtering result is used to filter out news items that contain at least one related number in the clarification data that is of the same order of magnitude and has the same highest digit, thus obtaining the third filtering result.

[0032] Connect the headline of the clarification news with the first paragraph of the clarification news article as the first text embedded content;

[0033] For each news item in the third filtering result, the news title is connected to the first paragraph of the news article to form the second text embedded content;

[0034] The text embedding model is used to transform the first text embedding content and the second text embedding content into a first text embedding vector and a second text embedding vector.

[0035] Calculate the similarity between the first text embedding vector and the second text embedding vector, and perform preliminary filtering on the news in the third filtering result based on the preset similarity interval.

[0036] Furthermore, analyzing the consistency and conflict between the content of all news items in the candidate news set and the clarifying news item determines whether there is a corresponding news item in the candidate news set that corresponds to the clarifying news item, including:

[0037] Construct content consistency detection prompts, which include the title and body of clarifying news, and the title and body of news after preliminary filtering;

[0038] The content consistency detection prompts are input into a large language model for analysis to obtain the first analysis result;

[0039] Construct content conflict detection prompts, including the headlines and body text of clarifying news, and the headlines and body text of news after preliminary filtering;

[0040] The content conflict detection prompts are input into a large language model for analysis to obtain the second analysis result;

[0041] Based on the results of the first and second analyses, determine whether there is a news item in the candidate news set that corresponds to the clarification news.

[0042] The present invention also provides an apparatus for identifying whether the authenticity of news has been altered, comprising:

[0043] The acquisition module is used to acquire the target news collection;

[0044] The discrimination module is used to discriminate each news item in the target news collection and obtain clarified news.

[0045] The extraction module is used to extract clarification data from clarification news using constructed content extraction prompts;

[0046] The retrieval module is used to generate retrieval vectors using clarification data as retrieval content and / or text of clarification news, and to retrieve news related to the retrieval content and / or retrieval vectors from historical news to obtain a candidate news set.

[0047] The analysis module is used to analyze the consistency and content conflict between all news in the candidate news set and the clarification news, and to determine whether there is a news in the candidate news set that corresponds to the clarification news. If there is a news in the candidate news set that corresponds to the clarification news, it is considered that the authenticity of the news corresponding to the clarification news has changed; otherwise, it is considered that there is no news in the candidate news set whose authenticity has changed.

[0048] The present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for identifying whether the authenticity of news has changed.

[0049] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for identifying whether the authenticity of news has been altered.

[0050] The above-described solution of the present invention has the following beneficial effects:

[0051] This invention identifies each news item in the target news set to obtain clarifying news; it extracts clarifying data from the clarifying news using constructed content extraction prompts; it uses the clarifying data as search content and / or generates search vectors from the text of the clarifying news, retrieving news related to the search content and / or search vectors from historical news to obtain a candidate news set; it analyzes the consistency and content conflict between all news items in the candidate news set and the clarifying news, determining whether there is news in the candidate news set corresponding to the clarifying news. If there is news in the candidate news set corresponding to the clarifying news, it is considered that the authenticity of the news item corresponding to the clarifying news has changed; otherwise, it is considered that there is no news item in the candidate news set whose authenticity has changed. Compared with the prior art, this invention first finds clarifying news and then finds corresponding historical news based on the clarifying news. The computational workload of determining whether a news item is a clarifying news item is relatively small. By first determining whether a news item is a clarifying news item, news items that do not require subsequent calculations can be quickly eliminated. This method can greatly reduce the computational workload. By using the clarifying data to analyze the consistency and content conflict between all news items in the candidate news set and the clarifying news, it is possible to effectively determine whether the candidate news item corresponds to and contradicts the clarifying news, thereby improving the ability to identify the authenticity of news.

[0052] Other beneficial effects of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating an embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of the device structure for identifying whether the authenticity of news has changed in an embodiment of the present invention.

[0055] Figure 3 This is a schematic diagram of the structure of the terminal device in an embodiment of the present invention. Detailed Implementation

[0056] To make the technical problems, solutions, and advantages of this invention clearer, a detailed description will be provided below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0057] In the description of this invention, it should be noted that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0058] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0059] This invention addresses existing problems by providing a method and related equipment for identifying whether the authenticity of news has changed.

[0060] like Figure 1 As shown, an embodiment of the present invention provides a method for identifying whether the authenticity of news has changed, including:

[0061] Step 1: Obtain the target news collection;

[0062] Step 2: Analyze each news item in the target news collection to obtain clarification news;

[0063] Step 3: Extract clarification data from the clarification news using the constructed content extraction prompts;

[0064] Step 4: Using the clarified data as the search content and / or using the text of the clarified news to generate a search vector, retrieve news related to the search content and / or search vector from historical news to obtain a candidate news set;

[0065] Step 5: Analyze the consistency and conflict of content between all news items in the candidate news set and the clarification news. Determine whether there is news in the candidate news set that corresponds to the clarification news. If there is news in the candidate news set that corresponds to the clarification news, it is considered that the authenticity of the news corresponding to the clarification news has changed. Otherwise, it is considered that there is no news in the candidate news set whose authenticity has changed.

[0066] Specifically, step 2 includes:

[0067] Extract the title and first sentence of each news item in the target news collection as the content of the analysis;

[0068] Construct a negative feature set and use the negative features in the negative feature set to replace the analysis content;

[0069] Construct a positive feature set, use the positive features in the positive feature set to match the analyzed content, and use the news that successfully matches the analyzed content as clarification news.

[0070] In this embodiment of the invention, the main text of each news item in the target news collection is segmented according to clause breaks and line breaks to obtain a list of sentences. Clause breaks include periods, question marks, exclamation marks, and ellipses. If there is at least one sentence in the sentence list and the length of the first sentence in the sentence list is not 0, then the title of the input news and the first sentence in the sentence list are concatenated with a period and recorded as the analysis content; otherwise, the title of the input news is recorded as the analysis content.

[0071] Since strings in the negative feature set can cause false matches in subsequent matching, the negative features appearing in the analyzed content are replaced with non-word characters to avoid false matches in the next matching step. One example of replacement is to use regular expressions to avoid the input news being judged as clarification news.

[0072] The constructed positive feature set contains positive features that can express meanings such as "clarification", "debunking", and "denial". The positive features are used to match the analysis content. If the match is successful, it means that the input news is a clarification news; otherwise, it means that the input news is not a clarification news.

[0073] It should be noted that the methods used to replace the analysis content in the embodiments of the present invention include, but are not limited to, regular expressions.

[0074] Specifically, step 3 includes:

[0075] The clarification news was pieced together to create content extraction prompts;

[0076] Input the extracted prompts into the large language model to generate clarification content for the clarification news;

[0077] Extract the main content, relevant subjects, relevant locations, and relevant figures from the clarification content of the news clarification to obtain clarification data.

[0078] In this embodiment of the invention, the content extraction prompts are used to ask the large language model whether the input news denies certain events, the main content of the denial content, the relevant subject, the relevant location, and the relevant numbers. The prompts use a few-sample approach to provide the large language model with some examples for reference, and the output answer is limited to JSON format.

[0079] It should be noted that, in embodiments of the present invention, any API interface can be used to call the large language model or a locally deployed large language model for answering and content extraction.

[0080] Specifically, step 4 includes:

[0081] Using clarification data as the search content and / or generating search vectors from the text of clarification news, and using discrete search methods and / or vector search methods, retrieve news related to the search content and / or search vectors from historical news to obtain a candidate news set.

[0082] Specifically, using discrete retrieval methods and / or vector retrieval methods, news related to the search content and / or search vector is retrieved from historical news, resulting in a candidate news set, including:

[0083] Generate search results based on the clarified data;

[0084] The headline and part of the body text of the clarifying news are concatenated to form a vector retrieval text;

[0085] Use a text embedding model to transform vector retrieval text into retrieval vectors;

[0086] By using search terms and / or search vectors, news related to the search terms and / or search vectors can be retrieved from historical news, resulting in a candidate news set.

[0087] In this embodiment of the invention, when using the vector retrieval method, the title of the clarification news and the first paragraph with more than 20 characters in the news text are connected by a newline character to form the vector retrieval text.

[0088] In this embodiment of the invention, a news retrieval engine retrieves news related to the search content and / or search vector from historical news, resulting in a candidate news set. The news retrieval engine operates in two modes: news maintenance and news retrieval. The processing flow of each mode is described below:

[0089] The news retrieval engine uses discrete retrieval engines and / or vector retrieval engines, and stores complete news data through a separate news data dictionary. The dictionary data corresponds to the data in the retrieval engine index through news IDs.

[0090] In the news maintenance workflow, the input news ID, title, and body text are inserted into the index of the discrete search engine.

[0091] and / or

[0092] The title of the input news article and the first paragraph with more than 20 characters in the body are connected by a newline character to form the vector index text. The text embedding model is used to convert the vector index text into a vector, and the converted vector is inserted into the index of the vector retrieval engine.

[0093] Finally, the complete content of the input news is recorded in the news data dictionary;

[0094] In the news retrieval workflow, the search terms are input into a discrete search engine, which then retrieves news articles related to the search terms from historical news sources, resulting in a candidate news set.

[0095] and / or

[0096] The search vector is input into the search engine, which retrieves news related to the search vector from historical news, thus obtaining a candidate news set.

[0097] Preferably, before step 5, the process includes preliminary filtering of the candidate newsletter using relevant subjects, locations, and figures from the clarified data as key information:

[0098] The candidate news set is filtered to extract news from all relevant subjects, including those with clarified data, to obtain the first filtering result;

[0099] The first filter result is used to filter out news from all relevant locations that include the clarified data, resulting in the second filter result;

[0100] The second filtering result is used to filter out news items that contain at least one related number in the clarification data that is of the same order of magnitude and has the same highest digit, thus obtaining the third filtering result.

[0101] Connect the headline of the clarification news with the first paragraph of the clarification news article as the first text embedded content;

[0102] For each news item in the third filtering result, the news title is connected to the first paragraph of the news article to form the second text embedded content;

[0103] The text embedding model is used to transform the first text embedding content and the second text embedding content into a first text embedding vector and a second text embedding vector.

[0104] Calculate the similarity between the first text embedding vector and the second text embedding vector, and perform preliminary filtering on the news in the third filtering result based on the preset similarity interval.

[0105] In this embodiment of the invention, the preset similarity range is [ ], The range of values ​​is .

[0106] Specifically, this involves analyzing the consistency and conflict between the content of all news items in the candidate news set and the clarifying news, and determining whether there is a corresponding news item in the candidate news set that corresponds to the clarifying news, including:

[0107] Construct content consistency detection prompts, which include the title and body of clarifying news, and the title and body of news after preliminary filtering;

[0108] The content consistency detection prompts are input into a large language model for analysis to obtain the first analysis result;

[0109] Construct content conflict detection prompts, including the headlines and body text of clarifying news, and the headlines and body text of news after preliminary filtering;

[0110] The content conflict detection prompts are input into a large language model for analysis to obtain the second analysis result;

[0111] Based on the results of the first and second analyses, determine whether there is a news item in the candidate news set that corresponds to the clarification news.

[0112] In this embodiment of the invention, the content consistency detection prompt is used to require the large language model to briefly summarize the main content of the clarifying news and the preliminarily filtered news respectively, and to determine whether the main content of the two news articles describes the same thing; the content conflict detection prompt is used to require the large language model to determine whether the main content of the two news articles conflicts.

[0113] It should be noted that, in this embodiment of the invention, the false news comes first and then the clarifying news, so the clarifying news should be later than the news in the candidate news set.

[0114] The following specific examples further illustrate the method provided in the embodiments of the present invention:

[0115] Suppose we obtain a news article from a news collection, named News A, which reads as follows.

[0116] Title: Automaker A: Rumors circulating online that Automaker B has halted its cooperation with Automaker A are false.

[0117] Text: Recently, some media outlets reported that Automaker B had suspended its battery cooperation with Automaker A. In response, a representative from Automaker A stated that this information is false and does not reflect the actual situation.

[0118] The news headline and the first sentence of the news article are extracted as the content of the analysis, and the analysis content is as follows:

[0119] Automaker A: Rumors circulating online that Automaker B has halted its cooperation with Automaker A are false. Recently, some media outlets reported that Automaker B had suspended its battery cooperation with Automaker A.

[0120] Negative features are regular expressions, such as: "Internet Traditional Media | Online Media |".

[0121] News item A did not match any negative features, therefore the analysis content remains unchanged after the replacement.

[0122] Positive features are regular expressions, such as: "Debunking | Clarifying | Online Rumors | Widespread | Rumors | False Reports | Spreading Rumors | Seriously Inaccurate | [for](?: Misinterpretation | Media Misreading | False Information) | Not True | [::] False | (?: Deny | Response) [\u4e00-\u9fa5] (?: Rumors / rumors)

[0123] News A matched positive features; News A is a clarification news item.

[0124] Extract key words from the concatenated content of news A as follows:

[0125] "user": You are an expert in reading and understanding news. Based on the given news content, if the article involves denying certain events, extract the main relevant parties, relevant figures, relevant locations, and specific details of the denied event using only the original text. Please return the results in the following format: "Relevant parties of the denied event, relevant figures of the denied event, relevant locations of the denied event, content of the denied event";

[0126] News: Is B car company planning to build a factory in country S? B car company executives deny the rumors online;

[0127] Specific details: On September 18th, the XX Daily, citing sources, reported that Country S is in preliminary talks with Country A's electric vehicle manufacturer, Company B, regarding the establishment of a manufacturing plant in Country S. The report stated that Country S has been extending an olive branch to Company B, attempting to attract it to build a factory locally by offering the right to purchase a certain amount of metals and minerals needed for Company B's electric vehicles from countries like Country G. It is understood that one initiative Country S is considering is providing financing to commodity traders to help them complete a cobalt-copper project in Country G, which could potentially help supply one of Company B's factories. Previously, commodity traders were evaluating their options for the Mutoshi project in Country G amid rising costs and persistently low cobalt prices. Country S has been striving to reduce its economic dependence on oil, and in addition to trying to attract Company B, its sovereign wealth fund is also a major investor in the electric vehicle startup in Country A. In response to the above report, the CEO and management of Company B denied the report in posts on the social media platform X, while Country S's Public Investment Fund declined to comment. Automaker B currently operates six factories globally and is building its seventh in northern M country. In May of this year, B's management stated that the company might finalize the location of its next factory by the end of the year. In August, B expressed interest in establishing a factory in Y country to produce low-cost electric vehicles. Recently, media reports indicated that the president of T country met with B's management in New York, A country, and invited B to establish a factory in T country. B's goal is to sell 20 million vehicles annually by 2030, several times more than the 1.3 million vehicles sold in 2022.

[0128] "assistant": "Car company B, country S, car company B plans to build a factory in country S."

[0129] "user": You are an expert in reading and understanding news. Based on the given news content, if the article involves denying certain events, extract the main relevant parties, relevant figures, relevant locations, and specific details of the denied event using only the original text. Please return the results in the following format: "Relevant parties of the denied event, relevant figures of the denied event, relevant locations of the denied event, content of the denied event";

[0130] News: Automaker A: Rumors circulating online that Automaker B has suspended its cooperation with Automaker A are false.

[0131] Details: Recently, some media outlets reported that automaker B had suspended its battery cooperation with automaker A. In response, a representative from automaker A stated that this information was false and inconsistent with the actual situation.

[0132] At this point, the large model returned the following answer: "Automaker A and Automaker B, Automaker B has suspended its cooperation with Automaker A."

[0133] Using "Automaker B halts cooperation with Automaker A" as the search term, 50 news articles were retrieved from discrete search engines.

[0134] The following vector retrieval text is generated by concatenating the title of news article A and the first sentence of the body text that is longer than 20 characters:

[0135] Automaker A: Rumors circulating online that Automaker B has suspended its cooperation with Automaker A are false: Recently, some media outlets reported that Automaker B has suspended its cooperation with Automaker A on batteries.

[0136] Using a text embedding model, the vector search text is transformed into a search vector. The search vector is then input into a vector search engine, which retrieves 50 news articles.

[0137] The news sets retrieved by the discrete search engine and the news sets retrieved by the vector search engine are merged to obtain a candidate news set with a data volume of 67.

[0138] A news item selected from the candidate news collection is designated as News B:

[0139] Title: Automaker B will no longer use batteries from Automaker A;

[0140] Text: According to media reports, Automaker B has decided to stop using batteries from Automaker A. The report states this is due to a series of fires caused by Automaker A's LFP (lithium iron phosphate) batteries. Automaker B will increase its battery orders from LG Energy Solutions and CATL. Furthermore, media reports indicate that LG Energy Solutions is developing LFP batteries for electric vehicles at Automaker B's request. Automaker A stated "no comment at this time."

[0141] News B contains all the relevant subjects in the clarification data, and the clarification data does not contain any relevant locations or figures, therefore News B is retained;

[0142] Connect the headline of news A with the first paragraph of the main text and embed it as the first text content;

[0143] Connect the headline of news B with the first paragraph of the main text and embed it as the second text content;

[0144] The text embedding model is used to transform the first text embedding content and the second text embedding content into a first text embedding vector and a second text embedding vector.

[0145] The similarity between the first text embedding vector and the second text embedding vector is calculated. The similarity is 0.72. The preset similarity interval is set to [0.6, 0.8]. Since the similarity is within the preset similarity interval, news B is retained.

[0146] Using the title and body of news A, and the title and body of news B, construct the following content consistency check prompts:

[0147] Please read the following two texts and summarize the main content of each text in one sentence. Then, determine whether the main content of the two texts involves the same event, answering with "yes" or "no":

[0148] Text 1: Automaker A: Rumors circulating online that Automaker B has halted its cooperation with Automaker A are false. Recently, some media outlets reported that Automaker B had suspended its battery cooperation with Automaker A. In response, a representative from Automaker A stated that this is false information and does not reflect the actual situation.

[0149] Text 2: Automaker B will no longer use batteries from Automaker A. According to media reports, Automaker B has decided to stop using batteries from Automaker A. The report states that this is due to a series of fires caused by Automaker A's LFP (lithium iron phosphate) batteries. Automaker B will increase its battery orders from LG Energy Solutions and CATL. Furthermore, media reports indicate that LG Energy Solutions is developing LFP batteries for electric vehicles at Automaker B's request. Automaker A stated that it has "no comment at this time."

[0150] Input the content consistency cue words into the large model to obtain the answer. The large model's answer is:

[0151] Text 1 Main content: Car company A denies that the news that car company B has suspended its battery cooperation is false information;

[0152] Text 2 Main content: Media reports indicate that automaker B has decided to stop using automaker A's batteries and switch to other suppliers; automaker A has not commented.

[0153] Does it involve the same thing? Yes;

[0154] The answer indicates that the two texts relate to the same event, therefore news item B is retained.

[0155] Using the headline and body of news A, and the headline and body of news B, construct the following content conflict detection prompts:

[0156] Please read the following two texts and determine whether the main content of text 1 completely or partially negates the main content of text 2. Answer with "yes" or "no".

[0157] Text 1: Automaker A: Rumors circulating online that Automaker B has halted its cooperation with Automaker A are false. Recently, some media outlets reported that Automaker B had suspended its battery cooperation with Automaker A. In response, a representative from Automaker A stated that this is false information and does not reflect the actual situation.

[0158] Text 2: Automaker B will no longer use batteries from Automaker A. According to media reports, Automaker B has decided to stop using batteries from Automaker A. The report states that this is due to a series of fires caused by Automaker A's LFP (lithium iron phosphate) batteries. Automaker B will increase its battery orders from LG Energy Solutions and CATL. Furthermore, media reports indicate that LG Energy Solutions is developing LFP batteries for electric vehicles at Automaker B's request. Automaker A stated that it has "no comment at this time."

[0159] Input the content consistency prompt into the large model to get the answer. The large model's answer is: Yes.

[0160] Therefore, it can be seen that in Text 1, car company A explicitly denies that the news that "car company B has stopped cooperating with car company A" is false information, while the main content of Text 2 is that the media reported that car company B decided to stop using car company A's batteries. Therefore, the content of Text 1 directly refutes the core information of Text 2.

[0161] The answer indicates that the main content of the two texts conflicts, therefore news A confirms that news B is false, and the veracity of news B has changed.

[0162] This invention, in its embodiments, identifies each news item in the target news set to obtain clarifying news; it extracts clarifying data from the clarifying news using constructed content extraction prompts; it uses the clarifying data as retrieval content and / or generates retrieval vectors from the text of the clarifying news, retrieving news related to the retrieval content and / or retrieval vectors from historical news to obtain a candidate news set; it analyzes the consistency and content conflict between all news items in the candidate news set and the clarifying news, determining whether there is news in the candidate news set corresponding to the clarifying news. If there is news in the candidate news set corresponding to the clarifying news, it is considered that the authenticity of the news item corresponding to the clarifying news has changed; otherwise, it is considered that there is no news item in the candidate news set whose authenticity has changed. Compared with the prior art, this invention first searches for clarifying news and then searches for corresponding historical news based on the clarifying news. The computational workload of determining whether a news item is a clarifying news item is relatively small. By first determining whether a news item is a clarifying news item, news items that do not require subsequent calculations can be quickly eliminated. This method can greatly reduce the computational workload. By using the clarifying data to analyze the consistency and content conflict between all news items in the candidate news set and the clarifying news, it is possible to effectively determine whether the candidate news item corresponds to and contradicts the clarifying news, thereby improving the ability to identify the authenticity of news.

[0163] Corresponding to the method for identifying whether the authenticity of news has changed as described in the above embodiments, such as Figure 2 As shown, this embodiment of the invention also provides a device 100 for identifying whether the authenticity of news has changed. The device 100 for identifying whether the authenticity of news has changed includes:

[0164] Module 101 is used to acquire the target news collection;

[0165] The discrimination module 102 is used to discriminate each news item in the target news collection to obtain the clarified news.

[0166] Extraction module 103 is used to extract clarification data from clarification news using constructed content extraction prompts;

[0167] The retrieval module 104 is used to generate a retrieval vector with the clarification data as the retrieval content and / or with the text of the clarification news, and to retrieve news related to the retrieval content and / or retrieval vector from historical news to obtain a candidate news set.

[0168] Analysis module 105 is used to analyze the consistency and content conflict between all news in the candidate news set and the clarification news, and to determine whether there is news in the candidate news set that corresponds to the clarification news. If there is news in the candidate news set that corresponds to the clarification news, it is considered that the authenticity of the news corresponding to the clarification news has changed; otherwise, it is considered that there is no news in the candidate news set whose authenticity has changed.

[0169] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0170] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0171] This invention also provides a terminal device, such as... Figure 3 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 3The diagram shows only one processor, a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100. When the processor D100 executes the computer program D102, it implements the above-described method for identifying whether the authenticity of news has changed.

[0172] The terminal device D10 can be a desktop computer, laptop, handheld computer, server, server cluster, or cloud server, etc. This terminal device may include, but is not limited to, a processor D100 and a memory D101. Those skilled in the art will understand that... Figure 3 This is merely an example of terminal device D10 and does not constitute a limitation on terminal device D10. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0173] The processor D100 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0174] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the terminal device D10. Furthermore, the memory D101 may include both internal and external storage units of the terminal device D10. The memory D101 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory D101 can also be used to temporarily store data that has been output or will be output.

[0175] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0176] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0177] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for identifying whether the authenticity of news has changed.

[0178] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a building device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0179] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for identifying whether the authenticity of news has changed, characterized in that, include: Step 1, obtain the target news collection; Step 2: Analyze each news item in the target news collection to obtain clarification news; Step 3: Extract clarification data from the clarification news using the constructed content extraction prompts; Step 4: Using the clarified data as the search content and / or using the text of the clarified news to generate a search vector, retrieve news related to the search content and / or the search vector from historical news to obtain a candidate news set; Step 5: Analyze the consistency and content conflict between all news items in the candidate news set and the clarification news, and determine whether there is any news item in the candidate news set that corresponds to the clarification news. If there is a news item in the candidate news set that corresponds to the clarification news, it is considered that the authenticity of the news item corresponding to the clarification news has changed; otherwise, it is considered that there is no news item in the candidate news set whose authenticity has changed.

2. The method for identifying whether the authenticity of news has changed according to claim 1, characterized in that, Step 2 includes: The title and first sentence of each news item in the target news collection are extracted as the analysis content; Construct a negative feature set, and use the negative features in the negative feature set to replace the analysis content; A positive feature set is constructed, and the positive features in the positive feature set are used to match the analyzed content. News that successfully match the analyzed content is designated as clarification news.

3. The method for identifying whether the authenticity of news has changed according to claim 2, characterized in that, Step 3 includes: The clarification news was pieced together to form content extraction prompts; Extract prompt words from the content and input them into a large language model to generate the clarification content of the clarification news; Extract the main content, relevant subjects, relevant locations, and relevant figures from the clarification content of the clarification news to obtain clarification data.

4. The method for identifying whether the authenticity of news has changed according to claim 3, characterized in that, Step 4 includes: Using the clarified data as the search content and / or generating a search vector from the text of the clarified news, a discrete search method and / or a vector search method are used to retrieve news related to the search content and / or the search vector from historical news, thus obtaining a candidate news set.

5. The method for identifying whether the authenticity of news has changed according to claim 4, characterized in that, Using discrete retrieval methods and / or vector retrieval methods, news related to the search content and / or the search vector is retrieved from historical news to obtain a candidate news set, including: Generate search results based on the clarified data; The title of the clarification news is concatenated with the text of the body of the clarification news to form a vector retrieval text; The text retrieval vector is transformed into a retrieval vector using a text embedding model. By using the search content and / or the search vector, news related to the search content and / or the search vector is retrieved from historical news, resulting in a candidate news set.

6. The method for identifying whether the authenticity of news has changed according to claim 4, characterized in that, Before step 5, the process includes using relevant subjects, locations, and figures from the clarified data as key information to perform preliminary filtering of the news in the candidate news set: The candidate news set is filtered to extract news from all relevant subjects that include the clarification data, resulting in a first filtering result; The first filtering result is used to filter out news from all relevant locations in the clarified data to obtain the second filtering result; The second filtering result is used to filter out news items that contain numbers that are of the same order of magnitude and have the same highest digit as at least one related number in the clarified data, thus obtaining the third filtering result. The headline of the clarification news is connected to the first paragraph of the body of the clarification news, which is used as the first text embedded content; For each news item in the third filtering result, the news title is connected to the first paragraph of the news article to form the second text embedded content; The first text embedding content and the second text embedding content are transformed into a first text embedding vector and a second text embedding vector using a text embedding model. Calculate the similarity between the first text embedding vector and the second text embedding vector, and perform preliminary filtering on the news in the third filtering result based on the preset similarity interval.

7. The method for identifying whether the authenticity of news has changed according to claim 6, characterized in that, Analyze the consistency and content conflict between all news items in the candidate news set and the clarification news, and determine whether there is a news item in the candidate news set corresponding to the clarification news, including: Construct content consistency detection prompts, which include the title and body of the clarification news, and the title and body of the news after preliminary filtering; The content consistency detection prompts are input into a large language model for analysis to obtain the first analysis result. Construct the content conflict detection prompts, which include the title and body of the clarification news, and the title and body of the news after preliminary filtering; The content conflict detection prompts are input into a large language model for analysis to obtain a second analysis result. Based on the first analysis result and the second analysis result, it is determined whether there is a news item in the candidate news set that corresponds to the clarification news.

8. A device for identifying whether the authenticity of news has been altered, characterized in that, include: The acquisition module is used to acquire the target news collection; The discrimination module is used to discriminate each news item in the target news collection to obtain clarification news; The extraction module is used to extract clarification data from the clarification news using constructed content extraction prompts; The retrieval module is used to generate a retrieval vector using the clarification data as retrieval content and / or the text of the clarification news, and to retrieve news related to the retrieval content and / or retrieval vector from historical news to obtain a candidate news set. The analysis module is used to analyze the consistency and content conflict between all news in the candidate news set and the clarification news, and to determine whether there is news in the candidate news set that corresponds to the clarification news. If there is news in the candidate news set that corresponds to the clarification news, it is considered that the authenticity of the news corresponding to the clarification news has changed; otherwise, it is considered that there is no news in the candidate news set whose authenticity has changed.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for identifying whether the authenticity of news has changed as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for identifying whether the authenticity of news has changed as described in any one of claims 1 to 7.

Images