A method and system for intelligent rumor detection and accurate rumor refutation based on blockchain

Abstract

The present invention belongs to the field of information technology, and more specifically, relates to a method and system for intelligent detection and accurate refutation of rumors based on blockchain. The method includes the following steps: S1. The social platform encrypts and uploads speech messages to IPFS, and reads data from the corresponding IPFS nodes; S2. Use a rumor discrimination model to conduct preliminary identification, detection and classification on the read data; S3. Encrypt the speech messages identified in step S2 using an encryption algorithm and upload them to the blockchain, and synchronously store them in the graph database; S4. Use a rumor detection model to match opposite semantic message pairs in the database for intelligent rumor detection; S5. Push the refuted content and correct messages to users who have held or viewed wrong opinions and messages.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and more specifically, relates to a method and system for intelligently detecting and accurately refuting rumors based on blockchain. Background Art

[0002] With the widespread use of mobile Internet and social tools, cyberspace has gradually become a gathering place for rumors, and in disguise has provided a convenient channel for the spread of rumors. Internet rumors are confusing, latent and harmful, which will not only damage personal interests, but also disrupt social order and even affect the image and security of the country. It is urgent to build an accountability mechanism and a rumor-refuting platform to control the breeding and spread of rumors.

[0003] The difficulty in holding people accountable for online rumors lies in the complexity of online information, the difficulty in tracing the source of information, and the low efficiency of rumor-refuting platforms in reality, which still cannot break through the time limit of manual review. In addition, the opacity of the process weakens the credibility of the rumor-refuting system. In addition, the difficulty in tracing the source of online rumors also makes it difficult to accurately refute rumors. Summary of the invention

[0004] To this end, it is necessary to provide a method and system for intelligent rumor detection and accurate rumor refutation based on blockchain, and to monitor, discover, refute and deal with the entire process of speech messages published on social platforms. To achieve the above purpose, the present invention provides a method for intelligent rumor detection and accurate rumor refutation based on blockchain, which is applied to an intelligent rumor detection and accurate rumor refutation system based on blockchain.

[0005] The present invention suppresses the increase of rumors, cleans up the existing rumor stock, and prevents their repeated propagation; solves the problem of rumor tracing and improves the credibility of the evidence retrieved; solves the time limitation problem of manual review of rumors; solves the problem of information cocoons and avoids group polarization; alleviates the problem of information asymmetry in the process of refuting rumors; and solves the problem of the lack of a sound working mechanism for governing online rumors.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A blockchain-based rumor intelligent detection and accurate rumor refutation method, comprising the following steps:

[0008] S1. The social platform encrypts the speech message and uploads it to IPFS, and reads the data from the corresponding IPFS node;

[0009] S2. Use the rumor discrimination model to perform preliminary identification, detection and classification on the read data;

[0010] S3, encrypting the speech message identified in step S2 using an encryption algorithm and uploading it to the blockchain, and synchronously storing it in the graph database;

[0011] S4. Use the rumor detection model to match opposite semantic message pairs in the database for intelligent rumor detection;

[0012] S5. Push the content of rumor refutation and correct messages to users who hold or have viewed wrong opinions and messages.

[0013] For further optimization of this technical solution, it further includes step S6. Trace the source of the rumor and hold relevant users legally accountable according to the information in the graph database and the blockchain.

[0014] For further optimization of this technical solution, the intelligent detection in step S4 includes

[0015] S41. Text content understanding;

[0016] S42. Antonym analysis;

[0017] S43. Conduct rumor detection on opposite semantic speech pairs.

[0018] For further optimization of this technical solution, in S41, a text summarization model based on Transformer is used to extract the summary of the original text, and the TextRank algorithm is used to extract the keywords in the original text.

[0019] A blockchain-based intelligent rumor detection and accurate rumor refutation system includes

[0020] Data collection module: used to collect speech messages published on social platforms and relevant information such as likes, forwards, and comments between entities;

[0021] Message recognition module: Use the rumor discrimination model to conduct preliminary identification detection and classification on the read data;

[0022] Data storage module: The data collected by the data collection module is transmitted to the data storage module for storage;

[0023] Rumor detection module: Use the rumor detection model to match opposite semantic message pairs in the database for intelligent rumor detection;

[0024] Rumor refutation push module: used to accurately push the content of rumor refutation and correct messages to users.

[0025] For further optimization of this technical solution, it further includes an accountability module: Trace the source of the rumor and hold relevant users legally accountable according to the information in the database.

[0026] For further optimization of this technical solution, the data collection module encrypts and uploads the collected information to IPFS.

[0027] In a further optimization of the technical solution, the data storage module includes a graph database and a blockchain.

[0028] This technical solution is further optimized. The blockchain uses Ethereum blockchain technology to build an Ethereum alliance chain for storing speech information.

[0029] This technical solution is further optimized, and the rumor detection module adopts deep learning technology to achieve intelligent detection.

[0030] Different from the prior art, the above technical solution has the following advantages:

[0031] 1. It provides a full-process supervision platform for monitoring, discovering, refuting rumors and handling speech and messages on social platforms.

[0032] 2. Rumors and information about relevant responsible persons are not easy to tamper with or forge.

[0033] 3. Classify speech messages according to semantic content themes and process speech messages more specifically.

[0034] 4. Use deep learning technology to achieve intelligent rumor detection in order to break through the time limitations of manual review.

[0035] 5. The technical means used to accurately refute rumors are conducive to breaking the information cocoon and avoiding group polarization. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is the basic framework diagram of the system;

[0037] Figure 2 It is a system hierarchy diagram;

[0038] Figure 3 This is a schematic diagram of the interaction between data and Ethereum;

[0039] Figure 4 Flowchart for matching opposite semantic message pairs and rumor identification. DETAILED DESCRIPTION

[0040] In order to explain the technical content, structural features, achieved objectives and effects of the technical solution in detail, the following is a detailed description in conjunction with specific embodiments and accompanying drawings.

[0041] A blockchain-based intelligent rumor detection and accurate rumor refutation method, which realizes uploading and reading data through IPFS, preliminary identification and classification of speech messages, chaining and database storage of speech messages, rumor detection based on matching opposite semantic pairs, accurate push of rumor refutation information, and accountability for rumors. The method is as follows:

[0042] S1. Uploading and Reading Data via IPFS: In this solution, the speech message data comes from a social platform. The social platform submits a data file on a front-end page of the system, and the file will be directly uploaded to IPFS, and a hash value of the file will be returned. When the system needs to process the data subsequently, the data file can be searched and retrieved locally in IPFS only through the hash value obtained during upload, and the speech messages in the data file can be parsed for subsequent use by the system.

[0043] S2. Preliminary Identification and Classification of Speech Messages: In this solution, a binary classification model for rumor identification is trained based on LSTM. The dataset for model training comes from an open-source Weibo rumor dataset on the Internet, which contains 1,538 rumors and 1,849 non-rumors. After training, the final recognition accuracy of the model can reach 87.4%. For the obtained speech messages, first, use the trained binary classification model for rumor identification based on LSTM to identify. If the probability that the model believes the speech message belongs to the rumor label is higher than 90%, it is directly determined as a rumor. Otherwise, it is temporarily considered not a rumor, and the trained content classification model based on BERT is used to classify the content of the speech message. In this solution, the speech messages can be classified into three categories by the content classification model based on BERT, namely current affairs news, scientific common sense, and opinion output. The dataset used to train the content classification model based on BERT is collected and produced by ourselves. Among them, the data of current affairs news mainly comes from the news content on Sina Weibo, the data of scientific common sense mainly comes from the Q&A data on Baidu Q&A, and the data of opinion output mainly comes from the emotional daily content published by ordinary users on Sina Weibo.

[0044] S3. Chain Upload and Database Storage of Speech Messages: For the speech messages preliminarily identified, their message content, publisher, publication time, etc. are packaged and encrypted and uploaded to the blockchain using a self-written smart contract. To improve the efficiency of the blockchain and prevent the chain from being too large and causing slow chain upload, only the above necessary content is recorded on the blockchain. At the same time, these necessary contents and other relevant non-necessary contents are also stored in the Neo4j graph database, and then the return information obtained from the chain upload, the ID of the data in the graph database, and all the necessary and non-necessary information of the speech messages are stored in the MySQL database together.

[0045] S4. Rumor Detection Based on Matching Opposite Semantic Pairs: In a social platform, if there is a pair of speech messages with opposite semantics for a certain thing, then one of this pair of messages is very likely to be a rumor. Based on this idea, this solution detects rumors by continuously searching for opposite semantic speech pairs describing the same thing in the database and identifying this pair of speeches.

[0046] For the specific process of judging semantic opposition and rumor recognition, refer to Figure 4 as shown below, which includes the following steps:

[0047] S41. Content understanding module: It is mainly responsible for understanding the content of the original speech message, including using a Transformer-based text summarization model to extract the summary of the original text and using the TextRank algorithm to extract the keywords in the original text. The purpose of extracting keywords is to determine whether two speeches describe the same thing. In this solution, the five keywords with the highest weights in the original speech are extracted using the TextRank algorithm, and it is considered that if more than three of the keywords have similar meanings, it can prove that the speeches describe the same thing. The summary extraction of the text is to improve the accuracy of subsequent semantic opposition judgment. In this solution, the semantics of the summary is regarded as the semantics of the original text to improve the accuracy of judgment.

[0048] S42. Antonym analysis module: It is mainly responsible for analyzing the semantics of the text summary and giving a judgment on whether the semantics are opposite. The specific judgment method is as follows: First, calculate the syntactic-based text similarity of the two summaries to ensure that the two texts describe the same aspect of the same thing. In this solution, it is considered that a similarity greater than 0.6 means that the two texts describe the same aspect of the same thing; then calculate the semantic-based text similarity of the two summaries. Only when the similarity is very low is it possible that the semantics of the two texts are opposite. In this solution, it is considered that a similarity less than 0.2 means that the semantics of the two texts may be opposite. Finally, perform semantic role annotation on the two summaries, marking the respective subject, predicate, object, etc. roles of the two summaries. Only when these roles meet the rules set in this solution (such as the subject or object must be similar), the two texts can be considered to have opposite semantics.

[0049] S43. Rumor detection for pairs of speeches with opposite semantics: According to the different classifications of speech content in S1, there are different rumor detection methods here. For current affairs news, because current affairs news is often sudden, only using a deep learning model may not get good discrimination results. Therefore, in this solution, the two texts are submitted to an authoritative institution for manual review; for scientific common sense, the two texts are respectively input into the LSTM-based rumor recognition binary classification model mentioned in S1 for identification, obtaining the probability of each being a rumor, and the speech with a higher probability is identified as a rumor, and the other speech is used as the reason for refuting the rumor to provide interpretability for the model to refute the rumor; for opinion output, no rumor detection is performed here, and only the record of semantic opposition is made.

[0050] S5. Precise Push of Rumor Refutation Information: To minimize the consequences of rumors as much as possible, this solution designs a function for precise push of rumor refutation information. It records the reading history of all users on the social platform and precisely pushes rumor refutation information to all users who have ever read a certain rumor. For opinion-output messages, since rumor detection is not performed, only messages with the opposite semantics are precisely pushed to users who have ever read a certain statement, thus playing a role in breaking the "information cocoon."

[0051] S6. Rumor Accountability: After identifying a rumor, use the Neo4j graph database to hold those responsible for the rumor accountable. The graph database records the entire process of the release, forwarding, commenting, and liking of a statement. Based on this data, a liability graph can be constructed. In this solution, the following algorithm is used to find the key nodes in the graph: Assign an initial influence to each node in the graph that has no in-degree, and set the initial influence of other nodes to 0. Transfer the influence of these nodes to the nodes they point to, and then delete these nodes. Repeatedly find nodes with no in-degree and perform the above operations of deletion and influence transfer until only the last node remains in the graph, thus completing the calculation of the influence of all nodes. Set an influence threshold of 1000. In this solution, nodes with an influence higher than this threshold are key nodes, and the persons responsible for these key nodes shall bear the responsibility for spreading rumors. At the same time, provide the corresponding statement data stored in the blockchain and its corresponding Token as evidence for accountability. Because the blockchain is immutable, it can be used as the basis for the authenticity and reliability of the data in the graph database.

[0052] Refer to Figure 1 As shown, for the basic framework diagram of the system, a blockchain-based intelligent rumor detection and precise rumor refutation system includes:

[0053] Data Acquisition Module: Using technical means such as IPFS and encryption algorithms, the system conducts different processing and review for messages from three different sources: incremental messages, stock messages, and user-reported messages to ensure the security of the messages.

[0054] Message Identification Module: For the messages that pass the review, use the message classification model to classify them into three categories according to the semantic content theme: current affairs news, scientific knowledge, and opinion output.

[0055] Data Storage Module: Store the original text of the messages on the social platform, the publishers, the release time, and other necessary information in the blockchain, and store the remaining data such as likes and views in the graph database.

[0056] Rumor Detection Module: Use the trained model to separately find message pairs with opposite semantics in various databases, and combine them with the reported or officially refuted rumors to intelligently identify and determine the truth or falsehood in the opposite semantics.

[0057] Rumor-busting push module: For current affairs and scientific knowledge rumors, the correct information will be automatically sent to users who have expressed relevant opinions and browsed the rumors; for opinion-outputting rumors, the opposing statements will be pushed to each other.

[0058] Accountability module: Conduct legal accountability based on the genealogy constructed using the graphic database.

[0059] In the data collection module, for incremental messages, a rumor discrimination model is used to perform preliminary identification and detection. Considering that the accuracy of model recognition needs to be improved, a specified threshold is set. If the value returned by the model is greater than the specified threshold, the statement is considered to be a rumor, refused to be published, and relevant information is fed back to the platform.

[0060] In the rumor-debunking push module, for current affairs news and scientific knowledge rumors, the system automatically sends the correct information to users who have expressed relevant opinions and browsed the rumors; for opinion-output rumors, the opposing parties' statements are pushed to each other, which helps break the information cocoon.

[0061] See also Figure 2 The following is a system hierarchy diagram. It includes:

[0062] Data collection layer: Data on speech information is encrypted and uploaded to the distributed file IPFS system, and the regulatory platform obtains data from IPFS by calling smart contracts.

[0063] Data storage layer: A combination of graph database and blockchain is used for storage. Taking advantage of the blockchain's tamper-proof feature, the original message, publisher, release time and other necessary information are stored in the blockchain. Other data is stored in the graph database, which uses the characteristics of the graph database as a point and edge-based storage unit to more intuitively display the relationship between entities and ensure that it cannot be tampered with.

[0064] Functional module layer: includes the administrator's registration and login module, existing and new message entry module, rumor detection module, topic classification module, message matching module, rumor tracing and accountability module, and targeted rumor push refuting module.

[0065] The innovative features of the present invention are as follows:

[0066] 1. Information is open and transparent: Information cannot be tampered with or forged, is highly reliable, and the system is decentralized and collectively maintained;

[0067] 2. The code is open and cannot be tampered with: Smart contracts are code contracts that reflect the contract, with open source code, high credibility, and auditability;

[0068] 3. The traceability platform is efficient and secure: Based on the advantages of blockchain, all relevant information of the rumor and the data of all people involved in the spread will be recorded on the chain and cannot be tampered with.

[0069] 4. Breaking the information cocoon: Different targeted push methods are adopted for speech information of different types and with different degrees of harm to help users break free from the shackles of viewpoints, improve the current platform recommendation mechanism, and break the information cocoon.

[0070] 5. Innovating the rumor detection method: In a social network, if there are two speech messages with opposite semantics regarding the same event, then one of these two speech messages is very likely to be a rumor. Therefore, this solution adds a module for matching speech pairs with opposite semantics before traditional rumor detection, and only conducts rumor detection on speech pairs with opposite semantics, reducing the workload of the rumor detection system and providing interpretability for the rumor detection model.

[0071] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or terminal device. Without further limitations, elements defined by the statement "including..." or "comprising..." do not exclude the existence of additional elements in the process, method, article or terminal device including the said elements. In addition, in this article, "greater than", "less than", "more than" are understood not to include the present number; "above", "below", "within" are understood to include the present number.

[0072] Although the above embodiments have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the above description is only the embodiment of the present invention, and does not limit the patent protection scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be included in the patent protection scope of the present invention by the same token.

Claims

1. A method for intelligent detection and accurate refutation of rumors based on blockchain, characterized in that, It includes the following steps: S1. The social platform encrypts and uploads the speech messages to IPFS and reads the data from the corresponding IPFS nodes; S2. Use the rumor discrimination model to conduct preliminary identification detection and classification on the read data; S3. Encrypt the speech messages identified in step S2 using an encryption algorithm and upload them to the blockchain, and synchronously store them in the graph database; S4. Use the rumor detection model to match opposite semantic message pairs in the database for intelligent rumor detection; S5. Push the content of rumor refutation and correct messages to users who have held or browsed wrong views and messages.

2. The method for intelligent detection and accurate refutation of rumors based on blockchain according to claim 1, characterized in that, It also includes step S6. Trace the origin of the rumor and hold relevant users legally accountable according to the information in the graph database and the blockchain.

3. The method for intelligent detection and accurate refutation of rumors based on blockchain according to claim 1, wherein, The intelligent detection in step S4 includes S41. Text content understanding; S42. Antonym analysis; S43. Conduct rumor detection on opposite semantic speech pairs.

4. The method for intelligent rumor detection and accurate rumor refutation based on blockchain according to claim 3, wherein, In S41, a text summarization model based on Transformer is used to extract the summary of the original text, and the TextRank algorithm is used to extract the keywords in the original text.

5. A blockchain-based intelligent rumor detection and accurate rumor refutation system, characterized in that, It includes a data collection module: used to collect speech messages published on the social platform and relevant information such as likes, forwards, and comments between entities; A message identification module: uses the rumor discrimination model to conduct preliminary identification detection and classification on the read data; A data storage module: the data collected by the data collection module is transmitted to the data storage module for storage; A rumor detection module: uses the rumor detection model to match opposite semantic message pairs in the database for intelligent rumor detection; A rumor refutation push module: used to accurately push the content of rumor refutation and correct messages to users.

6. The method for intelligent rumor detection and accurate rumor refutation based on blockchain according to claim 5, wherein, It also includes a responsibility pursuit module: trace the origin of the rumor and hold relevant users legally accountable according to the information in the database.

7. The method for intelligent rumor detection and accurate rumor refutation based on blockchain according to claim 5, characterized in that, The data collection module encrypts and uploads the collected information to IPFS.

8. The method for intelligent rumor detection and accurate rumor refutation based on blockchain according to claim 5, wherein The data storage module includes a graph database and a blockchain.

9. The method for intelligent rumor detection and accurate rumor refutation based on blockchain according to claim 8, characterized in that, The blockchain uses Ethereum blockchain technology to build an Ethereum consortium chain for storing speech information.

10. The method for intelligent rumor detection and accurate rumor refutation based on blockchain according to claim 5, characterized in that, The rumor detection module uses deep learning technology to achieve intelligent detection.

Images