CN-121980085-A - Block chain news real-time pushing platform based on artificial intelligence

Abstract

The invention relates to the technical field of information retrieval, in particular to a real-time block chain news pushing platform based on artificial intelligence, which comprises an atomic fact extraction and generation structuring unit for breaking news, wherein the building unit is constructed into a Merkle tree and stored in a block chain intelligent contract, an original text is vectorized and stored in a vector database, a bidirectional mapping is established, breaking news votes are verified, when the number of votes exceeds a fault tolerance threshold, state change is carried out and a broadcast constraint specification message is triggered, a constrained news text is generated, event metadata is read and decoded into emotion key parameters, semantic analysis parameters are extracted for the news text, the emotion key parameters and the semantic analysis parameters are fused according to preset weights to generate a mixed shape weight sequence and a language modulation curve, the mixed shape weight sequence and the language modulation curve are ordered according to block height, when a new task priority is higher than that of a current broadcasting task, emergency video content is inserted after the new task priority is received, and the original broadcasting content is automatically restored according to breakpoint information recorded on a chain after broadcasting is completed.

Inventors

• XU PENGHUI
• LIU XIUQING
• CHEN ZIYU

Assignees

• 南京艾穆数字技术有限公司

Dates

Publication Date

20260505

Application Date

20260123

Claims (7)

1. 1. Block chain news real-time push platform based on artificial intelligence, its characterized in that includes: The data storage module is used for extracting atomic facts of burst news, generating a structuring unit, constructing a hash value into a Merkle tree, storing the Merkle tree in a blockchain intelligent contract, vectorizing an original text, storing the vectorized original text into a vector database, and establishing bidirectional mapping between leaf nodes on a chain and vector records under the chain; The content generation module is used for verifying the authenticity vote of the breaking news, and when the number of the endorsements exceeds a fault tolerance threshold, the intelligent contract state is changed and triggers the broadcast constraint specification message to be converted into a system prompt word so as to generate a constrained news text; The audio-visual rendering module reads event metadata from leaf nodes on the chain, decodes the event metadata into emotion key parameters, and extracts semantic analysis parameters for semantic analysis of news text; The priority scheduling module is used for sequencing the broadcasting task queues according to the block heights, issuing a switching instruction when the priority of the new task is higher than that of the current broadcasting task, inserting emergency video content after receiving the instruction, and automatically recovering the original broadcasting content according to breakpoint information recorded on a chain after broadcasting is completed.
2. 2. The real-time push platform of the block chain news based on the artificial intelligence of claim 1, wherein the Merkle tree construction process comprises the steps of carrying out entity identification and relation extraction on a breaking news original text, identifying and extracting time information, geographic position information, event type information, data information and information source information, carrying out standardized processing on the extracted information according to a predefined field format, and generating a structuring unit comprising a timestamp field, a geographic coordinate field, an event type enumeration field, a numerical field and a confidence label field; The method comprises the steps of splicing fields of a structuring unit according to a predefined sequence to form a character string to be hashed, carrying out hash operation on the character string to be hashed to obtain leaf node hash values, pairing leaf node hash values in pairs, carrying out hash operation to obtain parent node hash values, carrying out iterative calculation layer by layer until obtaining unique Merkle root hash values, writing the unique Merkle root hash values into a fact registry of a blockchain intelligent contract, executing increment verification after block confirmation, updating a visibility bitmap, and marking retrieval visible states of data corresponding to each leaf node.
3. 3. The real-time push platform of blockchain news based on artificial intelligence of claim 1, wherein the establishment of the bidirectional mapping comprises obtaining a transaction hash returned by the blockchain as a unique identifier recorded on the chain after the storage on the chain is completed; inputting an original text of breaking news into a vector coding model, outputting vector representation with fixed dimension through text word segmentation, feature extraction and dimension mapping treatment, storing the vector representation as main data into a vector database, and writing the transaction hash and the corresponding Merkle tree leaf node position index into a metadata field; Based on the transaction hash, supporting forward retrieval of inquiring the under-chain vector record from the position of the leaf node on the chain and reverse retrieval of inquiring the verification state of the leaf node on the chain from the metadata of the under-chain vector record.
4. 4. The real-time push platform of the blockchain news based on the artificial intelligence is characterized in that the conversion process of the system prompt word comprises the steps of continuously monitoring an event log of the intelligent contract through long connection, capturing constraint specification information triggered by state change, analyzing data load of the constraint specification information, extracting constraint elements of four types including a fact list, a forbidden word list, numerical precision requirements and a mood parameter, converting the fact list into a positive guide instruction to indicate that a large language model contains specified fact elements in generated contents, converting the forbidden word list into a negative constraint instruction to indicate that the large language model is forbidden to use specified ambiguous words and speculative expressions in the generated contents, converting the numerical precision requirements into format constraint instructions to specify the accuracy degree and expression format of reference values in the generated contents, converting the mood parameter into a style instruction to specify the expression style of the generated contents according to the event type and the severity, and combining the positive guide instruction, the negative constraint instructions, the format constraint instructions and the instruction according to a template to complete predefined system prompt word.
5. 5. The blockchain news real-time pushing platform based on artificial intelligence according to claim 4 is characterized in that the generation process of the news text comprises the steps of taking a system prompt word as a model behavior constraint, combining the system prompt word with a user prompt word containing a generation task description to form prompt word input, calling a large language model interface to input the prompt word into a model for reasoning calculation, carrying out controlled text generation by the large language model based on constraint conditions in the prompt word input, outputting a news text draft, extracting key fact elements in the news text draft by a post verification module, carrying out consistency comparison with atomic fact records stored on a chain item by item, determining the news text draft as a final news text and outputting the final news text to an audio-visual rendering module if all the key fact elements are consistent, marking inconsistent content items if the fact elements inconsistent in comparison result exist, adding marked information to the prompt word input, and then recalling the large language model for generation until the comparison passes and/or reaches a preset retry number upper limit.
6. 6. The real-time push platform of blockchain news based on artificial intelligence of claim 5, wherein the process of generating the mixed shape weight sequence and the language modulation curve comprises the steps of reading metadata of a current news event from a fact registry of a blockchain, and extracting three emotion related fields of event severity level, event type classification and emotion polarity scoring; the method comprises the steps of inquiring a preset emotion base tone mapping table, taking a field combination of event severity level, event type classification and emotion polarity score as an inquiry key, obtaining a corresponding facial expression parameter reference interval, giving a first weight coefficient to emotion base tone parameters, giving a second weight coefficient to emotion analysis parameters, carrying out sentence segmentation on a news text according to punctuation marks, carrying out emotion polarity and emotion strength analysis on each sentence, arranging emotion strength values of each sentence according to time sequence to form a sentence-level emotion curve, scanning news text recognition emphasized words, turning words and pause marks, marking recognition results as rhythm nodes and recording position information in the text, determining the facial expression parameter reference interval as emotion base tone parameters, determining sentence-level emotion curves and rhythm nodes as semantic analysis parameters, giving a first weight coefficient to emotion base tone parameters, giving a second weight coefficient to emotion analysis parameters, carrying out weighted fusion calculation on the two types of parameters according to the weight coefficients, and generating a mixed shape weight sequence which changes along with time, and generating a voice output rhythm matching curve according to the position distribution of the rhythm nodes and the fluctuation characteristics of the sentence-level emotion curves.
7. 7. The real-time push platform for the block chain news based on the artificial intelligence is characterized in that the process of inserting the emergency video content comprises the steps of receiving a switching instruction issued by an intelligent contract through long connection by a broadcasting node, analyzing the instruction to obtain a content hash index of a target emergency video and a block height range of a recommended switch, marking a switch to-be-executed state in the broadcasting node, continuing to normally output the current video stream, positioning and pulling emergency video stream data from a content distribution network according to the content hash index, decoding and initializing the pulled emergency video stream, caching decoded frame data into a to-be-broadcast buffer area, continuously monitoring frame type information of the current output video stream, identifying the arrival time of a key frame, suspending the frame output of the current video stream when the current video stream is output to a key frame boundary, switching an output source to the emergency video stream of the to-be-broadcast buffer area, realizing smooth switching of the video content, recording the current playing position of the interrupted video stream as breakpoint information, writing the breakpoint information together with the block height of the interrupt time into a block chain, completing the broadcasting of the emergency video stream and/or effectively reading the video stream from the breakpoint position, and recovering the breakpoint position from the original video stream position after the playing of the breakpoint position.

Description

Block chain news real-time pushing platform based on artificial intelligence Technical Field The invention relates to the technical field of information retrieval, in particular to a block chain news real-time pushing platform based on artificial intelligence. Background In the existing digital news production and broadcasting technology, news manuscripts are automatically generated by using a deep learning large model, and video rendering and streaming media live broadcasting are gradually popularized by combining a virtual digital man engine. The existing system generally adopts a serial processing architecture of text input-model reasoning-video synthesis, directly inputs a model to generate a script after capturing Internet information through a crawler, and carries out linear pushing according to a preset time table through a streaming media tool. However, the prior art has significant technical defects in practical application, namely firstly, a large generation model is essentially based on probability prediction, a rigid constraint and authenticity verification mechanism for input source data are lacked, phantom content which is inconsistent with objective facts is extremely easy to generate, the false information is difficult to trace once generated, secondly, a traditional information retrieval and database structure cannot provide untampered evidence storage support for news data of high-frequency burst, so that the confidence of retrieval results cannot be quantified, thirdly, the traditional digital person driving only relies on semantic analysis of generated texts, direct parameter mapping based on objective event attributes (such as disaster grades) is lacked, emotion expressive force and event authenticity urgency are disjointed when broadcasting is caused, and finally, a conventional streaming media pushing mechanism lacks a highest priority interrupt channel based on a consensus mechanism, so that seamless switching of millisecond-level live streams cannot be realized under the condition of no manual intervention when an emergency is faced. Therefore, there is a need for a solution that can verify rights from the underlying data structure, logically constrain during generation, and respond immediately at the output. For this reason, a blockchain news real-time pushing platform based on artificial intelligence is proposed. Disclosure of Invention The invention aims to provide an artificial intelligence-based blockchain news real-time pushing platform which generates news contents through atomic fact chain evidence and intelligent contract constraint and combines intelligent audio-visual rendering and priority scheduling to realize high-credibility, controllable and high-timeliness broadcasting of news pushing. In order to achieve the above purpose, the present invention provides the following technical solutions: Block chain news real-time push platform based on artificial intelligence includes: The data storage module is used for extracting atomic facts of burst news, generating a structuring unit, constructing a hash value into a Merkle tree, storing the Merkle tree in a blockchain intelligent contract, vectorizing an original text, storing the vectorized original text into a vector database, and establishing bidirectional mapping between leaf nodes on a chain and vector records under the chain; The content generation module is used for verifying the authenticity vote of the breaking news, and when the number of the endorsements exceeds a fault tolerance threshold, the intelligent contract state is changed and triggers the broadcast constraint specification message to be converted into a system prompt word so as to generate a constrained news text; The audio-visual rendering module reads event metadata from leaf nodes on the chain, decodes the event metadata into emotion key parameters, and extracts semantic analysis parameters for semantic analysis of news text; The priority scheduling module is used for sequencing the broadcasting task queues according to the block heights, issuing a switching instruction when the priority of the new task is higher than that of the current broadcasting task, inserting emergency video content after receiving the instruction, and automatically recovering the original broadcasting content according to breakpoint information recorded on a chain after broadcasting is completed. Preferably, the Merkle tree construction process comprises the steps of carrying out entity identification and relation extraction on an original text of breaking news, identifying and extracting time information, geographic position information, event type information, data information and information source information, carrying out standardization processing on the extracted information according to a predefined field format, and generating a structuring unit comprising a timestamp field, a geographic coordinate field, an event type enumeration field, a numerical value field and a