An AI-based online rumor recognition system

Abstract

The present invention relates to the technical field of network rumor recognition, and in particular to a network rumor recognition system based on artificial intelligence, including: a data collection module for collecting news texts; a text analysis module for analyzing the characteristics of news texts according to preset keywords and the number of typos in the news texts; a user analysis module for analyzing the behavioral characteristics of users; a propagation analysis module for analyzing the propagation characteristics of news texts; an identification module for identifying news texts, and also for analyzing the source credibility of news texts according to the sources of the news texts, and also for analyzing the time consistency of news texts according to the release time of the news texts; a storage module for storing the recognition results of news texts and outputting them to users. The present invention effectively improves the accuracy of network rumor recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of network rumor recognition, and particularly to a network rumor recognition system based on artificial intelligence. Background Art

[0002] With the development of the Internet and social media, various types of information are spreading at an extremely fast speed, but a large amount of false information and network rumors are mixed in. The wide spread of misleading information not only affects public opinion, but may also have a negative impact on social stability and the mental health of individuals. Therefore, it is particularly important to develop an efficient and accurate network rumor recognition system.

[0003] Chinese Patent Publication No. CN112231562A discloses a network rumor recognition method and system. The method includes: obtaining a text feature matrix according to multiple texts containing rumor information; constructing a propagation graph structure, where the nodes in the graph structure are multiple texts, and the adjacency matrix in the graph structure is the forwarding and commenting relationship of rumor information among multiple texts; constructing a graph convolutional neural network model; the input of the graph convolutional neural network model is the text feature matrix and the adjacency matrix, and the output of the graph convolutional neural network model is a rumor feature matrix; training a neural network model according to the rumor feature matrix to obtain a rumor recognition model; and recognizing network rumors according to the rumor recognition model. The present invention trains a graph convolutional neural network model according to the forwarding and commenting relationship of rumors among multiple texts, and trains a neural network model according to the rumor feature matrix, effectively capturing the widely spread and scattered propagation characteristics of rumor information, and can effectively identify rumor information. It can be seen that when this solution identifies network rumors, it does not comprehensively analyze text emotional expression, social network user behavior, and data source, and there is a problem of low accuracy in network rumor recognition. Summary of the Invention

[0004] The purpose of the present invention is to provide a network rumor recognition system based on artificial intelligence to solve at least one of the problems existing in the prior art.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] A network rumor recognition system based on artificial intelligence includes:

[0007] A data collection module for collecting news texts;

[0008] A text analysis module for analyzing news text features according to the number of typos in the news text and preset keywords;

[0009] A user analysis module for analyzing user behavior characteristics according to the number of comments, likes, shares, and replies of the news text publisher within the first monitoring period;

[0010] A dissemination analysis module, which is used to analyze the dissemination speed of news texts according to the total number of shares of news texts within the second monitoring period, and analyze the dissemination characteristics of news texts according to the dissemination speed of news texts and the number of people who have disseminated the news texts within the second monitoring period;

[0011] An identification module, which is used to identify news texts according to the analysis results of the characteristics of news texts, the analysis results of user behavior characteristics, and the analysis results of the dissemination characteristics of news texts to obtain the identification results of news texts, and is also used to adjust the identification process of news texts according to the number of rumor texts historically disseminated by the news text publisher and the release time of the news text;

[0012] A storage module, which is used to store the identification results of news texts and output them to users.

[0013] Further, the text analysis module includes a text analysis unit, which is used to match preset keywords with news texts. If the preset keywords match the news texts successfully, the preset keywords are used as the rumor keywords of the news texts, and the number of occurrences of the i-th rumor keyword in the news texts is counted and denoted as fi;

[0014] The text analysis unit is also used to perform word segmentation on news texts and set the sentiment intensity value of the j-th word in the news texts as Dj;

[0015] The text analysis unit is also used to count the number of typos in news texts and denote it as B.

[0016] Further, the text analysis module also includes a feature extraction unit, which calculates the text features of news texts according to the number of occurrences fi of each rumor keyword in the news texts, the sentiment intensity value Dj of each word in the news texts, and the number of typos B in the news texts to obtain the features WT of the news texts.

[0017] Further, the user analysis module is used to analyze user behavior characteristics according to the number of comments Ct, the number of likes R, the number of shares L, and the number of replies S of the news text publisher within the first monitoring period to obtain the behavior characteristics HT of users.

[0018] Further, the dissemination analysis module includes a speed analysis unit, which is used to analyze the dissemination speed of news texts according to the total number of shares fx of news texts within the second monitoring period to obtain the dissemination speed Ve of news texts.

[0019] Further, the propagation analysis module further includes a construction unit configured to construct a propagator coefficient based on the number of people who have propagated the news text during the second monitoring period. The construction results of the propagator coefficient include P1, P2, and P3.

[0020] Further, the propagation analysis module further includes a propagation analysis unit configured to analyze the propagation characteristics of the news text based on the propagation speed Ve of the news text, the number of people zc who have propagated the news text during the second monitoring period, and the construction results of the propagator coefficient, so as to obtain the propagation characteristics CT of the news text.

[0021] Further, the recognition module includes a recognition unit configured to recognize the news text based on the analysis results of the characteristics of the news text, the analysis results of the user behavior characteristics, and the analysis results of the propagation characteristics of the news text, where:

[0022] If γ1×WT + γ2×HT + γ3×CT ≤ u0, the recognition unit determines that the news text is a normal news text;

[0023] If γ1×WT + γ2×HT + γ3×CT > u0, the recognition unit determines that the news text is a news text at risk of being a rumor;

[0024] Where γ1 is the text weight, γ2 is the behavior weight, γ3 is the propagation weight, and u0 is a preset abnormal coefficient.

[0025] Further, the recognition module further includes a source analysis unit configured to construct a source abnormal coefficient of the news text based on the number ys of rumor texts historically propagated by the news text publisher, and adjust the recognition process of the news text according to the source abnormal coefficient, and set the adjusted preset abnormal coefficient as u1.

[0026] Further, the recognition further includes a consistency analysis unit configured to analyze the consistency parameter YZ of the news text based on the number of days t0 since the news text was published and the preset number of days t1, and compare the consistency parameter YZ of the news text with the preset consistency threshold yz0. If the consistency parameter of the news text does not meet the threshold, the preset adjustment ratio is set as η1.

[0027] The beneficial effects of the present invention are as follows: Through multi-level feature analysis, including text features, user behavior, dissemination features, and source credibility, network rumors can be accurately identified, improving the accuracy and efficiency of judgment. The introduction of the user analysis and dissemination analysis modules helps better evaluate the dissemination risk of information and provides strong support for decision-making. Finally, the storage and feedback functions of the system ensure that users can obtain the identification results in a timely manner, improving the user experience and trust. In summary, the present invention provides an efficient and accurate solution for rumor identification, contributing to maintaining a healthy and authentic network environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0029] Figure 1 It is a schematic structural diagram of the network rumor identification system based on artificial intelligence in this embodiment.

[0030] Figure 2 It is a schematic structural diagram of the text analysis module in this embodiment.

[0031] Figure 3 It is a schematic structural diagram of the dissemination analysis module in this embodiment.

[0032] Figure 4 It is a schematic structural diagram of the identification module in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] In order to more clearly illustrate the present invention, the following further describes the present invention in conjunction with the preferred embodiments and the drawings. Similar components in the drawings are represented by the same reference numerals. Those skilled in the art should understand that the following specific description is illustrative rather than restrictive, and should not limit the protection scope of the present invention.

[0034] It should be noted that although terms such as first, second, and third may be used in the embodiments of the present application for description, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, without departing from the scope of the embodiments of the present application, the first can also be referred to as the second, and similarly, the second can also be referred to as the first.

[0035] Please refer to Figure 1 as shown, which is a schematic structural diagram of the network rumor identification system based on artificial intelligence in this embodiment. The system includes

[0036] A data collection module for collecting news texts. It can be understood that in this embodiment, the collection method of news texts is not specifically limited, and those skilled in the art can freely set it as long as it meets the requirements of the news text collection method. Among them, it can be collected from news websites and social media through web crawler technology or third-party API interfaces.

[0037] Please continue to refer to Figure 1 As shown, the system further includes:

[0038] A text analysis module, which is connected to the data collection module. The text analysis module is used to analyze the news text features according to the preset keywords and the number of typos in the news text.

[0039] Specifically, in this embodiment, the setting of the preset keywords is not specifically limited, and those skilled in the art can freely set it as long as it meets the requirements of the preset keyword setting. Among them, the preset keywords can be set as "breaking, urgent, shocking, breaking news, truth, secret, action, cover-up, decision-making insider, danger, alert, threat, fraud, health warning", etc.

[0040] Please refer to Figure 2 As shown, the text analysis module includes:

[0041] A text analysis unit, which is used to match the preset keywords with the news text. If the preset keywords match the news text successfully, the preset keywords will be used as the rumor keywords of the news text. If the preset keywords do not match the news text successfully, the preset keywords will not be used as the rumor keywords of the news text, and the number of occurrences of the i-th rumor keyword in the news text will be counted and denoted as fi;

[0042] The text analysis unit is also used to perform word segmentation on the news text and set the sentiment intensity value of the j-th word in the news text as Dj;

[0043] The text analysis unit is also used to count the number of typos in the news text and denote it as B. By analyzing the preset keywords and the number of typos, this module can quickly identify possible rumor text features and promote preliminary screening. This method can not only capture keywords in a timely manner but also perform automated evaluation based on the quality of the text, effectively improving the efficiency of rumor recognition.

[0044] Specifically, in this embodiment, there are no specific limitations on the number of occurrences of each rumor keyword in the news text, the way of word segmentation of the news text, the way of assigning emotional intensity values to each word in the news text, and the way of obtaining the number of typos. Those skilled in the art can set them freely, as long as the settings meet the requirements for the number of occurrences of each rumor keyword in the news text, the way of word segmentation of the news text, the way of assigning emotional intensity values to each word in the news text, and the way of obtaining the number of typos. Among them, the word frequency can be counted through SnowNLP, the news text can be segmented through the Jieba word segmentation tool, the emotional values of each word in the news text can be assigned through the Harbin Institute of Technology emotional dictionary, and the number of typos in the news text can be identified through Hunspell; in this embodiment, there are no specific limitations on the matching method, and those skilled in the art can set it freely, as long as the matching requirements are met, among which, matching can be based on strings.

[0045] Please continue to refer to Figure 2 As shown, the text analysis module further includes:

[0046] A feature extraction unit, which is connected to the text analysis unit. The feature extraction unit calculates the text feature WT of the news text according to the number of occurrences of each rumor keyword in the news text, the emotional intensity value of each word in the news text, and the number of typos in the news text, and sets:

[0047] ; By comprehensively calculating the text feature WT based on multi-dimensional features such as keyword occurrence frequency, emotional intensity, and the number of typos, the judgment ability and accuracy of the system for news texts are improved, thereby improving the accuracy of online rumor recognition. Among them, w1 is the keyword weight, w2 is the emotional weight, w3 is the typo weight, w1 + w2 + w3 = 1, I is the number of rumor keywords in the news text, N is the total number of words in the news text, and NZ is the total number of characters in the news text.

[0048] It can be understood that in this embodiment, there are no specific limitations on the way of obtaining the total number of words (N) and the total number of characters (NZ) of the news text. Those skilled in the art can set them freely, as long as the requirements for obtaining the total number of words (N) and the total number of characters (NZ) of the news text, which can usually be achieved through the string processing function in the programming language, are met. Among them, it can be obtained through the string processing function in the programming language.

[0049] Please continue to refer to Figure 1 As shown, the system further includes:

[0050] A user analysis module, which is connected to the text analysis module. The user analysis module is used to analyze the user behavior characteristics based on the number of comments Ct, the number of likes R, the number of shares L, and the number of replies S of the news text publisher within the first monitoring period, so as to effectively evaluate the user behavior characteristics. Among them:

[0051] The user behavior characteristics analysis module sets the user behavior characteristics as HT, and sets HT = α×Ct / T1 + β×(R + L + S) / T1, where T1 is the duration of the first monitoring period, α is the comment weight, β is the interaction weight, and α + β = 1. It can be understood that in this embodiment, the settings of the comment weight and the interaction weight are not specifically limited. Those skilled in the art can freely set them as long as they meet the setting requirements of the comment weight and the interaction weight. Among them, the best value of α is 0.4, and the best value of β is 0.6. By analyzing the interaction behaviors (comments, likes, shares, etc.) of the news publisher within the first monitoring period, this module can effectively evaluate the user behavior characteristics HT and reflect the potential rumor spreading risk. The user behavior analysis integrates social interactions into the rumor recognition process, improving the practicality of the system.

[0052] It can be understood that in this implementation, the acquisition methods of the number of comments Ct, the number of likes R, the number of shares L, and the number of replies S of the news text publisher within the first monitoring period are not specifically limited. Those skilled in the art can freely set them as long as they meet the acquisition requirements of the number of comments Ct, the number of likes R, the number of shares L, and the number of replies S of the news text publisher within the first monitoring period. Among them, the number of comments Ct, the number of likes R, the number of shares L, and the number of replies S of the news text publisher within the first monitoring period can be obtained through the API of social media or news websites. In this embodiment, the setting of the first monitoring period is not specifically limited. Those skilled in the art can freely set it as long as it meets the setting requirements of the first monitoring period. Among them, the 7 days before the news text publisher publishes can be used as the first monitoring period.

[0053] Please continue to refer to Figure 1 As shown, the system further includes:

[0054] A propagation analysis module, which is connected to the user analysis module. The propagation analysis module is used to analyze the propagation characteristics of the news text based on the total number of shares and the number of people who spread the news text within the second monitoring period.

[0055] Please refer to Figure 3 As shown, the propagation analysis module includes:

[0056] A speed analysis unit is used to analyze the propagation speed Ve of news texts based on the total number of shares fx of news texts within the second monitoring period, so as to promptly identify the propagation trend of information. It is set that Ve = fx / T2, where T2 is the duration of the second monitoring period; by analyzing the relationship between the total number of shares of news texts and time within the specified monitoring period, the propagation speed Ve is calculated. A high propagation speed usually means that the content has been concerned and shared by a large number of users in a short time, which may indicate that the content is an emergency or information with a high degree of attention. Through this analysis, the system can promptly identify the propagation trend of information and provide a basis for rumor identification.

[0057] Please continue to refer to Figure 3 As shown, the propagation analysis module further includes:

[0058] A construction unit, which is connected to the speed analysis unit. The construction unit is used to construct a propagator coefficient based on the number of propagators of news texts within the second monitoring period, so as to effectively distinguish the credibility of different propagators and reduce the propagation risk of false information; if cb(k) ≤ q1, the propagation analysis unit sets the propagator coefficient of the kth propagator to P1; if q1 < cb(k) ≤ q2, the propagation analysis unit sets the propagator coefficient of the kth propagator to P2; if cb(k) > q2, the propagation analysis unit sets the propagator coefficient of the kth propagator to P3; constructing the propagator coefficient based on the historical behavior statistics of propagators can evaluate the trust level of propagators in the current propagation chain according to their previous propagation records, further enhancing the robustness and effectiveness of the system;

[0059] Among them, cb(k) is the number of historical rumor - spreading texts of the kth propagator of the news text, q1 is the first preset quantity, q2 is the second preset quantity, P1 is the first preset coefficient, P2 is the second preset coefficient, P3 is the third preset coefficient, and P1 < P2 < P3; it can be understood that in this embodiment, no specific limitations are imposed on the preset quantities and preset coefficients, and those skilled in the art can freely set them as long as they meet the setting requirements of the preset quantities and preset coefficients. Among them, the best value of q1 is 3, the best value of q2 is 5, the best value of P1 is 1, the best value of P2 is 1.2, and the best value of P3 is 1.5.

[0060] Specifically, in this embodiment, there is no specific limitation on the acquisition methods of the total number of shares, the number of disseminators, and the number of rumor texts previously disseminated by the disseminators of the news text within the second monitoring period. Those skilled in the art can freely set them, as long as the acquisition requirements of the total number of shares, the number of disseminators, and the number of rumor texts previously disseminated by the disseminators of the news text within the second monitoring period are met. Among them, it can be obtained through the API of social media or news websites. In this embodiment, there is no specific limitation on the setting of the second monitoring period. Those skilled in the art can freely set it, as long as the setting requirements of the second monitoring period are met. Among them, the second monitoring period can be set as a time period of 5 days after the news text is released.

[0061] Please continue to refer to Figure 3 as shown, the propagation analysis module further includes:

[0062] A propagation analysis unit, which is connected to the construction unit. The propagation analysis unit is used to analyze the propagation characteristics CT of the news text based on the propagation speed Ve of the news text and the construction results of the number of disseminators zc and the disseminator coefficient of the news text within the second monitoring period. It is set that:

[0063] ;

[0064] Among them, VT is the preset propagation speed, CB is the preset propagation quantity threshold, and P(k) is the disseminator coefficient of the k-th disseminator of the news text. By integrating the results of propagation speed analysis and disseminator coefficient, the propagation characteristics of the news text are deeply analyzed to help judge the influence of the news content and potential public opinion risks, so as to provide a scientific basis for further rumor identification.

[0065] It can be understood that in this embodiment, there is no specific limitation on the setting of the preset propagation speed and the preset propagation quantity threshold. Those skilled in the art can freely set them, as long as the setting requirements of the preset propagation speed and the preset propagation quantity threshold are met. Among them, the best value of VT is 200 times per day, and the best value of CB is 1000.

[0066] Please continue to refer to Figure 1 as shown, the system further includes:

[0067] An identification module, which is connected to the propagation analysis module. The identification module is used to identify the news text based on the analysis results of the characteristics of the news text, the analysis results of user behavior characteristics, and the analysis results of the propagation characteristics of the news text to obtain the identification result of the news text, and is also used to adjust the identification process of the news text according to the number of rumor texts previously disseminated by the news text publisher and the release time of the news text.

[0068] Please refer to Figure 4 as shown, the identification module includes:

[0069] An identification unit, which is used to identify news texts based on the analysis results of the characteristics of news texts, the analysis results of user behavior characteristics, and the analysis results of the dissemination characteristics of news texts, where:

[0070] If γ1×WT + γ2×HT + γ3×CT ≤ u0, the identification unit determines that the news text is a normal news text;

[0071] If γ1×WT + γ2×HT + γ3×CT > u0, the identification unit determines that the news text is a news text with rumor risk;

[0072] Where γ1 is the text weight, γ2 is the behavior weight, γ3 is the dissemination weight, γ1 + γ2 + γ3 = 1, and u0 is a preset abnormal coefficient; by comprehensively analyzing the text characteristics, user behavior characteristics, and dissemination characteristics of news texts, the final determination of news texts is realized, so as to improve the accuracy of network rumor identification.

[0073] It can be understood that in this embodiment, the settings of each weight and the preset abnormal coefficient are not specifically limited, and those skilled in the art can freely set them as long as the settings of each weight and the preset abnormal coefficient meet the requirements. Among them, the best value of γ1 is 0.4, the best value of γ2 is 0.3, the best value of γ3 is 0.3, and the best value of u0 is 0.46.

[0074] Please continue to refer to Figure 4 As shown, the identification module further includes:

[0075] A source analysis unit, which is connected to the identification unit. The source analysis unit constructs the source abnormal coefficient of the news text according to the number ys of historical rumor texts disseminated by the news text publisher, and adjusts the identification process of the news text according to the source abnormal coefficient, where:

[0076] If ys = 0, the source analysis unit sets the source abnormal coefficient of the news text to 0;

[0077] If ys > 0, the source analysis unit sets the source abnormal coefficient of the news text to LY, and sets LY = η×lg(ys + 1), where η is a preset adjustment ratio, 0 < η < 0.4;

[0078] The source analysis unit sets the adjusted preset abnormal coefficient to u1, and sets u1 = u0 - LY; by analyzing the number of historical rumor texts disseminated by the publisher, the source abnormal coefficient is constructed, so that the system can conduct risk assessment on the source of news texts, so as to improve the accuracy of network rumor identification.

[0079] Specifically, this embodiment does not specifically limit the method for obtaining the number of historical rumor texts spread by the news text publisher. Those skilled in the art can freely set it as long as the requirements for obtaining the number of historical rumor texts spread by the news text publisher are met. Among them, it can be obtained through the API of social media or news websites.

[0080] It can be understood that in this embodiment, the setting of the preset adjustment ratio is not specifically limited. Those skilled in the art can freely set it as long as the requirements for setting the preset adjustment ratio are met. Among them, the best value of η is 0.3.

[0081] Please continue to refer to Figure 4 as shown, the recognition further includes:

[0082] A consistency analysis unit, which is connected to the source analysis unit. The consistency analysis unit analyzes the consistency parameter YZ of the news text according to the release days t0 of the news text and the preset release days t1, and sets: YZ = 1 - 1 / (1 + e t1-t0 ), where e is the natural logarithm;

[0083] The consistency analysis unit compares the consistency parameter YZ of the news text with the preset consistency threshold yz0, and adjusts the construction process of the source anomaly coefficient according to the comparison result, where:

[0084] If YZ ≤ yz0, the consistency analysis unit determines that the consistency parameter of the news text meets the threshold and does not make adjustments;

[0085] If YZ > yz0, the consistency analysis unit determines that the consistency parameter of the news text does not meet the threshold, and sets the preset adjustment ratio to η1, and sets η1 = η × {1 + exp[3 × (YZ - yz0) / (YZ + yz0) - 3]}; Evaluating the consistency parameter YZ of the news text based on the release time helps to determine whether the information is an online rumor, thus providing an important basis for rumor recognition and further improving the accuracy of online rumor recognition.

[0086] Specifically, this embodiment does not specifically limit the method for obtaining the release days of the news text. Those skilled in the art can freely set it as long as the requirements for obtaining the release days of the news text are met. Among them, it can be obtained through the API of social media or news websites.

[0087] It can be understood that in this embodiment, the setting of the preset consistency threshold and the preset release days is not specifically limited. Those skilled in the art can freely set it as long as the requirements for setting the preset consistency threshold and the preset release days are met. Among them, the best value of t1 is 5 days, and the best value of yz0 is 0.5.

[0088] Please continue to refer toFigure 1 As shown, the system further includes:

[0089] A storage module, which is connected to the recognition module. The storage module is used to store the recognition results of news texts and output them to users.

[0090] Specifically, the network rumor recognition system based on artificial intelligence in this embodiment is applied to the rumor recognition of social media and network platforms. By comprehensively analyzing multi-dimensional information such as text features, user behaviors, propagation features, and information sources, it not only improves the accuracy and efficiency of rumor recognition, but also helps to build a healthier and safer network environment.

[0091] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation manners here. Any obvious changes or modifications derived from the technical solutions of the present invention still fall within the protection scope of the present invention.

Claims

1. An artificial intelligence-based network rumor recognition system, characterized in that, Including: A data acquisition module for acquiring news texts; A text analysis module for analyzing the characteristics of news texts based on the number of typos in the news texts and preset keywords; A user analysis module for analyzing the user behavior characteristics based on the number of comments, likes, shares, and replies of the news text publisher within the first monitoring period; A dissemination analysis module for analyzing the dissemination speed of news texts based on the total number of shares of news texts within the second monitoring period, and analyzing the dissemination characteristics of news texts based on the dissemination speed of news texts and the number of people who have disseminated the news texts within the second monitoring period; An identification module for identifying news texts based on the analysis results of the characteristics of news texts, the analysis results of user behavior characteristics, and the analysis results of the dissemination characteristics of news texts to obtain the identification results of news texts, and also for adjusting the identification process of news texts based on the number of historical rumor texts disseminated by the news text publisher and the publication time of the news text; A storage module for storing the identification results of news texts and outputting them to users; The text analysis module includes a text analysis unit, which is used to match the preset keywords with the news text. If the preset keywords match the news text successfully, the preset keywords are used as the rumor keywords of the news text, and the number of occurrences of the i-th rumor keyword in the news text is counted and denoted as fi; The text analysis unit is also used to segment the news text and set the sentiment intensity value of the j-th word in the news text as Dj; The text analysis unit is also used to count the number of typos in the news text and denote it as B; The text analysis module also includes a feature extraction unit, which calculates the text characteristics of the news text based on the number of occurrences fi of each rumor keyword in the news text, the sentiment intensity value Dj of each word in the news text, and the number of typos B in the news text to obtain the characteristics WT of the news text. It is set that: ; The text feature WT is comprehensively calculated based on multi-dimensional features such as the frequency of keyword occurrences, emotional intensity, and the number of typos, improving the system's judgment ability and accuracy for news texts, and thus enhancing the accuracy of online rumor recognition. Among them, w1 is the keyword weight, w2 is the emotion weight, w3 is the typo weight, w1 + w2 + w3 = 1, I is the number of rumor keywords in the news text, N is the total number of words in the news text, and NZ is the total number of characters in the news text; The user analysis module is used to analyze the user behavior characteristics based on the number of comments Ct, likes R, shares L, and replies S of the news text publisher within the first monitoring period to obtain the behavior characteristics HT of the user. It is set that HT = α × Ct / T1 + β × (R + L + S) / T1, where T1 is the duration of the first monitoring period, α is the comment weight, β is the interaction weight, and α + β = 1; The identification module also includes a source analysis unit, which constructs the source anomaly coefficient of the news text based on the number ys of historical rumor texts disseminated by the news text publisher, and adjusts the identification process of the news text according to the source anomaly coefficient, and sets the adjusted preset anomaly coefficient as u1; The recognition module further includes a consistency analysis unit, which analyzes the consistency parameter YZ of the news text based on the release days t0 of the news text and the preset release days t1, and compares the consistency parameter YZ of the news text with the preset consistency threshold yz0. If the consistency parameter of the news text does not meet the threshold, the preset adjustment ratio is set to η1, and η1 = η × {1 + exp[3 × (YZ - yz0) / (YZ + yz0) - 3]}.

2. The network rumor recognition system based on artificial intelligence according to claim 1, characterized in that, The propagation analysis module includes a speed analysis unit, which is used to analyze the propagation speed of the news text based on the total number of shares fx of the news text in the second monitoring period to obtain the propagation speed Ve of the news text.

3. The network rumor recognition system based on artificial intelligence according to claim 2, wherein, The propagation analysis module includes a construction unit, which is used to construct the propagator coefficient according to the number of propagators of the news text in the second monitoring period. The construction results of the propagator coefficient include P1, P2, and P3.

4. The network rumor recognition system based on artificial intelligence according to claim 3, characterized in that, The propagation analysis module further includes a propagation analysis unit, which is used to analyze the propagation characteristics of the news text based on the propagation speed Ve of the news text, the number of propagators zc of the news text in the second monitoring period, and the construction results of the propagator coefficient to obtain the propagation characteristics CT of the news text. It is set that: ; where VT is the preset propagation speed, CB is the preset propagation quantity threshold, and P(k) is the propagator coefficient of the kth propagator of the news text.

5. The network rumor recognition system based on artificial intelligence according to claim 4, wherein The recognition module includes a recognition unit, which is used to recognize the news text based on the analysis results of the characteristics of the news text, the analysis results of the user behavior characteristics, and the analysis results of the propagation characteristics of the news text, where: If γ1 × WT + γ2 × HT + γ3 × CT ≤ u0, the recognition unit determines that the news text is a normal news text; If γ1 × WT + γ2 × HT + γ3 × CT > u0, the recognition unit determines that the news text is a rumor risk news text; where γ1 is the text weight, γ2 is the behavior weight, γ3 is the propagation weight, and u0 is the preset abnormal coefficient.

Images