Artificial intelligence information auditing system and method based on distributed micro-service architecture

Abstract

The invention relates to the technical field of intelligent auditing, in particular to an artificial intelligence information auditing system and method based on a distributed micro-service architecture, and the method comprises the following steps: analyzing to-be-audited information at a gateway layer, and extracting a lightweight decision context which comprises an information type, a semantic key frame and an emergency degree mark; binding a target micro-service instance according to the decision context, pre-loading the decision context to a memory working area of the target micro-service instance, and triggering a distributed auditing engine to execute multi-modal analysis and generate an atomic auditing decision; and aggregating a plurality of atomic auditing decisions of the same information flow, and generating a final global consistency decision through time window conflict resolution processing. According to the invention, the data volume compression ratio is improved, and the bandwidth and storage pressure of the auditing system are reduced. Meanwhile, in combination with an information urgency degree dynamic calculation mechanism, rapid priority response to sudden public opinion information is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent auditing technology, and in particular to an artificial intelligence information auditing system and method based on a distributed microservice architecture. Background Technology

[0002] Bidding platforms involve a large amount of information submitted by users, such as bidding company information, project performance certificates, qualification documents, responses to bid clarifications, and online comments. This information is complex in type and diverse in format, including both structured data and a large amount of unstructured information such as text, images, and attachments.

[0003] To ensure compliant operation and transaction security, bidding service platforms generally introduce information verification mechanisms to identify and block inappropriate content such as confidential information, false qualifications, malicious attacks, commercial defamation, and misleading links. However, existing information verification schemes face the following prominent problems in the actual operation of bidding platforms:

[0004] The text in bidding information often exhibits a mix of "semi-formal" and "business terminology," containing numerous industry abbreviations, policy references, document numbers, and qualification descriptions, making it prone to misjudgment or omission by standard language models. Furthermore, some illegal information (such as misleading external links and forged official seal images) is highly concealed, requiring a combination of semantic and contextual judgment, thus raising the bar for the content comprehension capabilities of the review system.

[0005] Before the centralized bidding or bid closing on the platform, user traffic surges. Traditional centralized review systems suffer from issues such as review queue congestion and model calculation delays, leading to delayed or even invalid review results. This disrupts the normal bidding process and poses a significant system bottleneck risk. Currently, most platform information review relies on static rule engines or fixed text classifiers, which cannot identify similar violations posted by the same bidding entity on multiple pages and at multiple nodes. This results in fragmented review results, frequent judgment conflicts, and a lack of globally consistent judgment capabilities.

[0006] As industry policies, compliance standards, and bidding strategies continue to evolve, so too do the ways in which violations are expressed. Existing audit systems often lack online optimization and deviation calibration mechanisms for models, making them prone to the phenomenon of "expired models continuing to run," leading to a gradual decline in audit accuracy. Summary of the Invention

[0007] This invention provides an artificial intelligence information review system and method based on a distributed microservice architecture, which has the ability to perform information review with lightweight context expression, dynamic semantic scheduling, global consistency judgment and adaptive calibration.

[0008] An AI-based information verification method based on a distributed microservice architecture includes the following steps:

[0009] S1: Parse the information to be reviewed at the gateway layer and extract a lightweight decision context, which includes information type, semantic keyframes and urgency markers;

[0010] S2: Bind the target microservice instance according to the decision context, preload the decision context into the memory work area of ​​the target microservice instance, trigger the distributed audit engine to perform multimodal analysis and generate atomic audit decisions;

[0011] S3: Aggregates multiple atomic review decisions from the same information flow and generates a final globally consistent decision through time window conflict resolution.

[0012] Optionally, S1 specifically includes execution at the gateway layer:

[0013] S11, parse the text data stream to obtain semantic keyframes;

[0014] S12, calculate the initial urgency level based on the credibility of the information source and the density of sensitive words in the content;

[0015] S13, Information type identifier: Based on the characteristics of the transmission protocol header and the content structure, it is determined to be text, image, video or cross-modal combination type.

[0016] Optionally, the parsing of the text stream includes extracting the core predicate-argument structure through a dependency parser and generating semantic keyframes, wherein the semantic keyframes include action subject, object, and modifier triples.

[0017] Optionally, S12 also introduces a time decay function: when information is associated with real-time hot events, the initial urgency is corrected by the time decay function to obtain the final urgency label.

[0018] Optionally, the binding of the target microservice instance in S2 includes calculating the semantic keyframe hash value in the decision context and performing Hamming distance matching with the context signature set of each microservice instance's memory workspace; if there is a microservice instance with a matching distance less than the dynamic matching threshold, then the current microservice instance is bound as the target microservice instance; otherwise, a new microservice instance is created and the signature set is initialized.

[0019] Optionally, S2 further includes encoding the decision context in the memory working area snapshot format and writing it directly to the memory working area snapshot area of ​​the target microservice instance, and triggering an atomic copy from the snapshot area to the execution area through a memory barrier instruction;

[0020] In the target microservice instance, multi-violation matching is performed based on preloaded semantic keyframes, and the output includes atomic audit decisions including decision type and confidence level. The violation matching preferentially uses the historical decision pattern cache in the memory work area to perform near real-time comparison.

[0021] Optionally, the dynamic matching threshold comprehensively considers the current system load and the urgency score of the information to be reviewed. Specifically, it includes setting a basic matching threshold to represent the 64-bit semantic hash signature within the maximum acceptable Hamming distance range, obtaining the current system CPU utilization and normalizing it into a load coefficient, using the urgency score corresponding to the information to be reviewed, setting a relaxation factor based on the urgency score, and calculating the dynamic matching threshold based on the basic matching threshold, the current system CPU utilization, and the relaxation factor.

[0022] Optionally, S3 includes attaching decision flow fingerprints to all atomic review decisions in the same information flow, and aggregating decision sequences within a sliding time window based on the decision flow fingerprints, wherein the length of the time window is dynamically adjusted according to the urgency.

[0023] Optionally, the conflict resolution process includes:

[0024] If there is a type conflict in the decision sequence within the time window, activate the hidden state backtracker and execute:

[0025] i. Input the atomic decision sequence as the observation value into the hidden Markov model;

[0026] ii. Decode the optimal hidden state sequence using the Viterbi algorithm and output a globally consistent decision.

[0027] An AI-powered information verification system based on a distributed microservice architecture is used to implement the aforementioned verification methods, including:

[0028] Gateway layer module: Used to receive information to be reviewed and perform semantic parsing to extract lightweight decision context, including information type, semantic keyframes and urgency markers;

[0029] Microservice instance cluster: used to bind the target microservice instance according to the decision context, and preload the context into the memory workspace of the target microservice instance;

[0030] The audit engine module is deployed in each target microservice instance and configured to perform multimodal analysis based on the decision context and generate atomic audit decisions.

[0031] The decision aggregation and conflict resolution module is used to aggregate and determine conflicts among multiple atomic review decisions in the same information flow within a time window, and output the final globally consistent review result.

[0032] The beneficial effects of this invention are:

[0033] 1. This invention introduces a dependency syntax-driven semantic keyframe extraction mechanism at the gateway layer of information review, compressing the information to be reviewed into predicate-argument triples and generating a 64-bit semantic hash signature. Compared with traditional BERT embedding or full-modal feature extraction, this structure improves the data volume compression ratio and reduces the bandwidth and storage pressure of the review system. Simultaneously, combined with a dynamic calculation mechanism for information urgency, it enables rapid and prioritized response to sudden public opinion events. By completing structural compression and urgency scoring at the gateway layer, this invention achieves edge-end content prediction, filtering, and distribution, effectively supporting millions of concurrent review tasks.

[0034] 2. Traditional microservice scheduling typically employs round-robin or static rules, which can easily lead to context fragmentation and high state transition costs. This invention addresses this by constructing a Hamming distance matching mechanism based on "semantic keyframe hash signature → in-memory signature set," enabling semantically driven dynamic instance binding. By dynamically adjusting the matching threshold based on system load factors and information urgency, adaptive optimization of the scheduling strategy is achieved. Furthermore, the target instance atomically preloads its context and performs decision generation, caching historical decision patterns for redundant judgment reuse and controlling overall response latency. This mechanism maintains high scalability while effectively improving the consistency and decision accuracy of similar information processing, making it particularly suitable for attack variant-intensive scenarios (such as advertising inducement and borderline content).

[0035] 3. To address the potential asynchronicity, errors, and conflicts in atomic review decisions within a distributed environment, this invention proposes a hidden state backtracking resolution mechanism based on the Viterbi algorithm. This mechanism maps the atomic decision sequence within a sliding time window to an observation sequence, derives the optimal hidden state path using a Hidden Markov Model, and outputs a consistent and fault-tolerant global review result. It also possesses stronger noise resistance and semantic consistency recognition capabilities. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the review method flow according to an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the audit system composition according to an embodiment of the present invention. Detailed Implementation

[0039] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. For some well-known technologies, those skilled in the art may also use other alternative methods to implement the invention. Moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0040] like Figure 1 As shown, the AI-based information verification method based on a distributed microservice architecture includes the following steps:

[0041] S1: Parse the information to be reviewed at the gateway layer and extract a lightweight decision context, which includes information type, semantic keyframes and urgency markers.

[0042] S1 performs the following steps at the gateway layer:

[0043] S11, Text Stream Processing: Extracting the core predicate-argument structure using a dependency parser to generate semantic keyframes. These semantic keyframes consist of triples in the format: <subject, object, modifier>; for example, <user, report, fraudulent link>, used to capture core semantic relationships in the text and significantly compress the feature representation dimension.

[0044] The dependency syntax-driven semantic keyframe extraction process is as follows:

[0045] The process involves performing basic cleaning operations on the text to be reviewed, including unifying full-width and half-width characters, removing emoticons, and normalizing URLs and numbers.

[0046] b segments the text into independent sentences and tags each sentence with part-of-speech tags, providing the dependency parser with prior information on word classes, such as verbs, nouns, and adverbs.

[0047] c calls the built-in lightweight dependency parser to build a dependency tree for each sentence, extracting the following three types of key dependency relations:

[0048] Subject-verb relationship (SBV): Identifies the agent of the action (i.e., the subject of the action);

[0049] Verb-Object Relationship (VOB): Identifies the object that performs the action (i.e., the action object);

[0050] Adverbial-verb relations (ADV, ATT): Identify modal words, time words, place words, etc., used to modify actions.

[0051] d Traverse the dependency tree, select all main verbs as semantic center predicate nodes, exclude function words and conjunction structures, and ensure that the semantic expression is clear and independent.

[0052] The 'e' module is centered around a verb and combines it with its connected subject, object, and modifiers to construct a triple structure. The unified format is: <action subject, object, modifier>. If a certain item is missing, it can be set to empty or a default value to ensure structural consistency. The final triple structure is packaged into a structured object, which, together with the information type and urgency score, constitutes a lightweight decision context for downstream microservices to call.

[0053] S12, Dynamic Urgency Tag: The urgency level is dynamically generated based on content sensitivity and public opinion hotspots.

[0054] Where, base urgency The basic urgency level is represented by a weighted average of sensitive word density (e.g., "pornography" or "fraud") and the credibility of the information source. α represents the relevance coefficient of the hot topic, scored by online public opinion platforms (e.g., Weibo trending topics) to assess the correlation between information content and the hot topic. β is a time decay factor, controlling the rate at which the heat of the sudden event decays over time; initially β = 0.1, decreasing to β = 0.01 later. now t represents the current time. event This indicates the time when the associated event was first monitored. Real-time computation is performed in the FPGA acceleration module deployed on the gateway node, with a single operation latency of less than 2ms, meeting the requirements for large-scale concurrency.

[0055] base urgency=λ1·S content +λ2·(1-C source ); where S content This represents the content sensitivity score, indicating the density of sensitive words, high-risk phrases, and offensive semantics in the information to be reviewed. The value range is [0,1], with higher values ​​indicating greater sensitivity. (C) source The credibility score for the information source represents the information publisher's historical reputation level or platform certification weight, with a value range of [0,1]. The higher the value, the more credible the information. λ1 and λ2 are weight coefficients that satisfy λ1+λ2=1. λ1=0.7 and λ2=0.3 are set to emphasize the dominant role of content factors.

[0056] S13, Information Type Identifier: Employs a fusion determination mechanism, combining dual characteristics of the network protocol layer and content layer.

[0057] Based on the transmission protocol header, the information is initially determined to be text, image, or video.

[0058] If a discrepancy is detected between the protocol header and content characteristics—for example, if the header is labeled as text / html but a video frame signature is extracted from the content—an anomaly detector is activated to determine if it is a spoofing attack. If it is not an attack, it is marked as a cross-modal composite information type. This discrepancy resolution mechanism can effectively prevent malicious information type spoofing, improving the accuracy and security of auditing.

[0059] Semantic keyframe extraction, urgency calculation, and information type determination are all performed at the gateway node to avoid bottlenecks to backend services caused by large-scale information forwarding. Predicate-argument triples and hash signatures are used to replace high-dimensional semantic embedding to achieve lightweight audit context encapsulation. Type confirmation is performed by combining protocol layer and content layer information to avoid the risk of spoofing and to provide a more stable type foundation for subsequent dynamic binding of microservices.

[0060] S2: Bind the target microservice instance according to the decision context, preload the decision context into the memory workspace of the target microservice instance, trigger the distributed audit engine to perform multimodal analysis and generate atomic audit decisions.

[0061] A target microservice instance refers to a specific microservice runtime unit in a distributed microservice architecture, bound to a decision context for a piece of information to be reviewed, based on semantic feature matching or scheduling strategies, and used to execute the review task for that information. Each microservice instance is a relatively independent review processing unit, integrating a review engine. It runs on a node in the microservice cluster and has the following capabilities:

[0062] Receive and preload the decision context;

[0063] Maintain a context signature cache (set) in memory;

[0064] Execute multimodal auditing logic (pattern matching, confidence level judgment);

[0065] Output atomic review decisions.

[0066] The target microservice instance is the one most suitable for the current information flow, selected from all available instances using a semantic hash signature matching algorithm.

[0067] Either similar content has already been processed (semantically similar);

[0068] It is either created specifically for the current information (first appearance, cold start);

[0069] For example: A comment text: "The explosion scene video is so realistic!" has a semantic keyframe hash of 0x8F21...E31A; there are 20 review instances in the current cluster; among them, the signature set of instance #7 contains 0x8F21...E31B (Hamming distance 1); then this instance is selected as the target microservice instance, that is, the binding processor of this message.

[0070] S2 specifically includes the following steps:

[0071] S21, Dynamic Instance Binding: For the semantic keyframe of the information to be reviewed, calculate its hash signature value and compare it with the context signature set maintained in the memory working area of ​​each active target microservice instance, using Hamming distance as the matching metric. If the Hamming distance between the current hash signature value and the context signature set of an instance is less than the dynamic matching threshold d, the binding is performed. th If a match is found, the instance is immediately bound; otherwise, a new instance is dynamically created and its context signature set is initialized.

[0072] The dynamic matching threshold is calculated as follows:

[0073] d th =base threshold ·(1-load factor )+urgency·adaptive slack ;

[0074] Where, base threshold Indicates the maximum number of differences allowed in a 64-bit hash signature, base threshold =8, load factor Indicates the current system load level. adaptive slack The dynamic relaxation factor is defined as:

[0075]

[0076] The hash signature value of a semantic keyframe is calculated as follows:

[0077] S211. Input semantic keyframes from a text stream, in the format of normalized triples: <action subject, action, object>; for example: <user, report, fraudulent link>.

[0078] S212. Encoding Preparation: Process the three elements of the triplet as follows:

[0079] Standardize using Unicode;

[0080] Remove stop words and spaces;

[0081] If the elements are phrases, they are concatenated into a single string in word order.

[0082] For example: <User, report, fraudulent link> → "User reports fraudulent link".

[0083] S213. String Normalization and Bit String Generation: Using the above concatenation result as the input string, the lightweight hash encoding algorithm xxHash is used to generate a 64-bit binary signature value, which serves as the hash signature of the current semantic keyframe. Example output format: Semantic keyframe: <User, Report, Fraudulent Link> → Hash Signature: 0x9A7F4D33B8021A7C; This signature is the contextual semantic fingerprint that can be compared.

[0084] The Hamming distance calculation method (used for microservice instance binding matching) is as follows:

[0085] S214. Comparison object: Semantic keyframe hash signature (binary string A) of the information to be reviewed;

[0086] A set of historical signatures (binary string B) pre-stored in the memory workspace of a microservice instance.

[0087] S215. Hamming Distance Definition: Hamming distance represents the number of corresponding bits that differ between two binary strings of equal length. Let two 64-bit signatures be:

[0088] A = a1a2...a 64 ;

[0089] B = b1b2...b 64 ;

[0090] The Hamming distance is calculated as follows: in, This is an indicator function; it takes the value 1 if the condition is true, and 0 otherwise.

[0091] S216. Actual Implementation: At the implementation level, bitwise operations can be used to accelerate the process.

[0092] Perform an XOR operation on two 64-bit integers:

[0093] The number of binary bits that are 1 in the statistical results (i.e., the number of bit differences) can be obtained using the CPU's built-in instruction popcount(diff). For example:

[0094] A: 1001101001111111...;

[0095] B: 1001101001111011...;

[0096] XOR: 0000000000000100...;

[0097] Hamming distance: 1.

[0098] S217. Matching Strategy: Combine the above calculation results with the dynamic matching threshold d. th Compare;

[0099] If there exists a Hamming(A,B) i ) <d th If the condition is met, the current instance is identified as a "semantically compatible" instance and bound to the target microservice instance; otherwise, the cold start logic is triggered to create a new instance and initialize its context signature set.

[0100] S22, Context Preloading: After the target microservice instance is determined, the current decision context is encoded in a structured snapshot format and written to the instance's memory working area snapshot area via direct memory write operations, achieving uninterrupted context transfer. Subsequently, an atomic copy from the snapshot area to the execution area is completed through the memory barrier instruction (MOVDIR64B under x86 architecture), ensuring execution context consistency and supporting high concurrency.

[0101] Snapshot area: Read-only data area, storing a copy of the structured context;

[0102] Execution Zone: Real-time access area for online decision-making and execution units;

[0103] Atomic replication characteristics: 512 bits wide per replication round, with a delay of less than 10 ns.

[0104] S23, Atomic Decision Generation: After completing context preloading, the target microservice instance immediately performs multimodal violation pattern matching analysis based on the semantic keyframes in the context to generate an atomic audit decision.

[0105] Includes fields:

[0106] Review types include: violation, normal, and suspected.

[0107] Decision confidence level (floating-point value 0 to 1);

[0108] Instance memory workspace maintenance decision mode triangular matrix cache:

[0109] Pattern 1: <Action Subject, Action, Object> → Decision Type;

[0110] Pattern 2: <Visual signature, text triple> → Confidence correction coefficient;

[0111] If the Hamming distance between the current semantic keyframe and a certain pattern in the cache is ≤2, the decision result is directly reused and the confidence is attenuated (default attenuation is 20%). When a new instance is created and initialized, the Top 100 high-frequency patterns can be pulled from neighboring nodes as a pre-warming cache to shorten the initial processing latency.

[0112] Application example scenario: Rumor verification in breaking news events.

[0113] Input information: Semantic keyframe: <Unknown account, posted, chemical plant explosion in a certain area>

[0114] Urgency score = 0.95;

[0115] Signature calculation: Current frame hash: 0x8A3D...C7F2;

[0116] Signature set of instance A: [0x8A3D...C7F1,0x91B2...E4A3] (Hamming distance from the first signature = 1);

[0117] Dynamic matching threshold calculation (current CPU utilization is 70%): Matching result: 1 < 6 → Successfully bound to instance A;

[0118] Decision generation:

[0119] Preload the context into the snapshot area of ​​instance A, and atomically copy it to the execution area;

[0120] Matching cache pattern <*, release, chemical plant accident> → preset to "false information";

[0121] Original confidence level: 0.92 → Confidence level adjusted after cache reuse:

[0122] Confidence final=0.92·0.8=0.736;

[0123] Output atomic audit decision: {"Type":"Violation","Confidence":0.736}.

[0124] The aforementioned atomic decision generation is as follows:

[0125] S231. After receiving the context data written from the gateway layer, the target microservice instance first extracts the semantic keyframes of the current audit task from the execution area memory, including:

[0126] Predicate-argument triples, hash signatures, and urgency scores;

[0127] S232. Start the violation pattern matching engine (pattern library call): The microservice instance maintains a violation pattern knowledge base internally, with a multimodal mapping table structure, mainly including the following two types of matching rules:

[0128] Pattern Type 1: Text Pattern (Structured Semantic Matching)

[0129] Format: <Main keywords, action keywords, object keywords □ → Decision type>

[0130] Matching methods: exact match or fuzzy match (supports partial match and semantic synonym match);

[0131] Each rule has a priority weight value (such as a preset weight of 0.9 for "<report, explosion, video>").

[0132] Mode Type 2: Multimodal Fusion Mode (Visual + Text)

[0133] Format: <Visual signature, semantic triple → confidence correction factor;

[0134] Matching method: The hash signature is compared with the existing visual signature by Hamming distance. If the distance is less than 2 bits, it is considered an "approximate match". The semantic keyframe information is combined for judgment, and a confidence correction coefficient of 0.8 is provided.

[0135] S233. Matching Process and Rule Selection:

[0136] a) Preliminary matching: Traverse the current semantic keyframe and the known violation patterns in the cache. If there is a complete match or a partial match, record the matching item of the pattern and the corresponding decision suggestion type (such as "false information" or "advertising inducement").

[0137] b) When multiple rules conflict: sort them according to priority weight. If there is a decision conflict (such as one rule matching "violation" and another matching "suspected"), the confidence fusion module is triggered to calculate the comprehensive judgment value.

[0138] S234. Confidence scoring and decay mechanism:

[0139] Original confidence score calculation: Confidence score is calculated based on matching pattern weights and urgency.

[0140] confidence raw =w match ·(1+γ·urgency); where w match The priority weight for rule matching is γ, which is an urgency correction coefficient ranging from 0.2 to 0.5. A cache reuse decay mechanism is used to avoid over-reliance on history; if a historical decision pattern in the cache is hit (hash distance ≤ 2), the confidence level is reduced by 20% from the original level.final =confidence raw 0.8;

[0141] S235. The final structured audit results are output by the decision generation unit within the instance:

[0142] Review type (enumeration values: violation / normal / suspected);

[0143] Confidence score (floating-point value 0 to 1).

[0144] S3: Aggregates multiple atomic review decisions from the same information flow and generates a final globally consistent decision through time window conflict resolution.

[0145] S3 specifically includes the following:

[0146] S31, Streaming Decision Aggregation: Assign a unique decision flow fingerprint to each information flow to be reviewed. This fingerprint is obtained by hashing the "information flow identifier and the timestamp of the first atomic decision" and is used to identify the aggregation affiliation of the information within the entire system.

[0147] fingerprint = SHA3(FlowID||T0); where FlowID represents the unique identifier of the information flow, T0 represents the timestamp of the first atomic review decision, || represents the string concatenation operation, and the SHA3 output is a 256-bit hash value.

[0148] All atomic decisions with the same fingerprint are uniformly aggregated within a sliding time window for subsequent consistency judgment. The length of the time window is dynamically controlled by the urgency of the information.

[0149]

[0150] S32, Conflict Resolution: If type conflicts occur in the atomic review decisions collected within the time window (e.g., some are "violations" and some are "compliances"), the system will activate the hidden state backtracker to resolve the conflicts.

[0151] The atomic decision sequence is treated as an observation sequence and input into a pre-defined Hidden Markov Model (HMM):

[0152] State space:

[0153] S1: Violation status; S2: Compliance status;

[0154] Initial probabilities: P(S1) = 0.6, P(S2) = 0.4;

[0155] State transition probability (example):

[0156] P(S1→S1)=0.7, P(S1→S2)=0.3;

[0157] P(S2→S2)=0.6, P(S2→S1)=0.4;

[0158] The Viterbi algorithm is used to decode the atomic decision sequence and output the optimal hidden state path as the final globally consistent decision, as follows:

[0159] S321: Problem Modeling and Input:

[0160] Problem modeling: Treat the aggregated atomic review decision sequence within the time window as an observation sequence O = [O1, o2, ..., o T ], each o t The "decision type" (e.g., "violation" or "compliance") of the t-th atomic decision is represented, and T represents the total length (number of time steps) of the atomic review decision sequence.

[0161] Modeling assumptions: The actual review status of the information flow exhibits Markov property in the short term, that is, its hidden review state sequence Q = [q1, q2, ..., q T Satisfying the first-order Markov condition, each q t Let represent the hidden audit state at the t-th time step. The transition is controlled by the inter-state probability, and the observation is controlled by the state emission probability.

[0162] State space definition: q t ∈{S1: violation, S2: compliance}, that is:

[0163] S1 is a hidden state 1, indicating a "violation" state;

[0164] S2 hidden state 2 indicates "compliance" status;

[0165] S322, Algorithm Input (HMM Parameters): The Viterbi algorithm requires a Hidden Markov Model (HMM) containing the following parameters:

[0166] 1. Initial state probability vector π: π = [π1, π2] = [P(q1 = S1), P(q1 = S2)]; π1 = 0.6, π2 = 0.4;

[0167] 2. State transition probability matrix A:

[0168]

[0169] 3. Emission probability matrix B (i.e., the probability of a state generating an observation): The emission probability of each hidden state to an observation is modeled as follows:

[0170]

[0171] S323, Viterbi algorithm decoding steps:

[0172] Given an observation sequence O = [o1, o2, ..., o T Perform the following steps:

[0173] 1: Initialization (t=1): For each state s∈{S1,S2}, initialize the path probability matrix δ1(s) and the path pointer ψ1(s):

[0174]

[0175] ψ1(s)=0;

[0176] π s Let δs represent the initial probability of the hidden state s, i.e., the prior probability that the system believes the first time step is in that state. Let δ1(s) represent the probability value of the path with the highest probability among all paths starting from the starting point and ending in state s at time t=1. Let ψ1(s) represent the state before the optimal path to state s at time t=1 (used for backtracking paths). The emission probability represents the probability that the hidden state s generates the observation value o1.

[0177] 2: Recursion (t = 2 to T): For each time step t ∈ [2, T] and each current state s j ,implement:

[0178]

[0179] Represents the state transition probability, from state s i Transition to state s j The probability, Indicates the emission probability: hidden state s j Generate observations o t The probability of;

[0180] 3: Termination: Find the state with the highest termination probability:

[0181] Record the probability of the maximum path: This indicates the state where the optimal path terminates at time T;

[0182] 4: Path backtracking: From Begin by using the pointer matrix ψ t Backtrack the hidden state path:

[0183]

[0184] The most likely hidden state path is finally obtained:

[0185] S324, Global Consistency Decision Generation: Optimal Hidden State Path Q * Statistical analysis was conducted, and the state that appeared most frequently was selected as the global consistency audit result within that time window.

[0186] If S1 (violation) occurs the most times → output "violation";

[0187] If S2 (compliance) is the primary parameter, then the output will be "compliance".

[0188] like Figure 2 As shown, an AI-powered information review system based on a distributed microservice architecture is used to implement the aforementioned review methods, including:

[0189] Gateway layer module: Used to receive information to be reviewed and perform semantic parsing to extract lightweight decision context, including information type, semantic keyframes and urgency markers;

[0190] Microservice instance cluster: used to bind the target microservice instance according to the decision context, and preload the context into the memory workspace of the target microservice instance;

[0191] The audit engine module is deployed in each target microservice instance and configured to perform multimodal analysis based on the decision context and generate atomic audit decisions.

[0192] The decision aggregation and conflict resolution module is used to aggregate and determine conflicts among multiple atomic review decisions in the same information flow within a time window, and output the final globally consistent review result.

[0193] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0194] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An AI-based information verification method based on a distributed microservice architecture, characterized in that, Includes the following steps: S1: Parse the information to be reviewed at the gateway layer and extract a lightweight decision context, which includes information type, semantic keyframes and urgency markers; S2: Bind the target microservice instance according to the decision context, preload the decision context into the memory work area of ​​the target microservice instance, trigger the distributed audit engine to perform multimodal analysis and generate atomic audit decisions; S3: Aggregates multiple atomic review decisions from the same information flow and generates a final globally consistent decision through time window conflict resolution.

2. The artificial intelligence information review method based on a distributed microservice architecture according to claim 1, characterized in that, S1 specifically includes execution at the gateway layer: S11, parse the text data stream to obtain semantic keyframes; S12, calculate the initial urgency level based on the credibility of the information source and the density of sensitive words in the content; S13, Information type identifier: Based on the characteristics of the transmission protocol header and the content structure, it is determined to be text, image, video or cross-modal combination type.

3. The artificial intelligence information review method based on a distributed microservice architecture according to claim 2, characterized in that, The parsing of the text stream includes extracting the core predicate-argument structure through a dependency parser and generating semantic keyframes, which include action subject, object, and modifier triples.

4. The artificial intelligence information review method based on a distributed microservice architecture according to claim 1, characterized in that, S12 also introduces a time decay function: when information is associated with real-time hot events, the initial urgency is corrected by the time decay function to obtain the final urgency label.

5. The artificial intelligence information review method based on a distributed microservice architecture according to claim 1, characterized in that, The binding of the target microservice instance in S2 includes calculating the semantic keyframe hash value in the decision context and performing Hamming distance matching with the context signature set of each microservice instance's memory work area; if there is a microservice instance with a matching distance less than the dynamic matching threshold, then the current microservice instance is bound as the target microservice instance; otherwise, a new microservice instance is created and the signature set is initialized.

6. The artificial intelligence information review method based on a distributed microservice architecture according to claim 5, characterized in that, The S2 further includes encoding the decision context in the memory working area snapshot format and writing it directly to the memory working area snapshot area of ​​the target microservice instance, and triggering an atomic copy from the snapshot area to the execution area through a memory barrier instruction; In the target microservice instance, multi-violation matching is performed based on preloaded semantic keyframes, and the output includes atomic audit decisions including decision type and confidence level. The violation matching preferentially uses the historical decision pattern cache in the memory work area to perform near real-time comparison.

7. The artificial intelligence information review method based on a distributed microservice architecture according to claim 5, characterized in that, The dynamic matching threshold comprehensively considers the current system load and the urgency score of the information to be reviewed. Specifically, it includes setting a basic matching threshold to represent the 64-bit semantic hash signature within the maximum acceptable Hamming distance range, obtaining the current system CPU utilization and normalizing it into a load coefficient, using the urgency score corresponding to the information to be reviewed, setting a relaxation factor based on the urgency score, and calculating the dynamic matching threshold based on the basic matching threshold, the current system CPU utilization, and the relaxation factor.

8. The artificial intelligence information review method based on a distributed microservice architecture according to claim 1, characterized in that, S3 includes attaching decision flow fingerprints to all atomic review decisions in the same information flow, and aggregating decision sequences within a sliding time window based on the decision flow fingerprints, wherein the length of the time window is dynamically adjusted according to the urgency.

9. The artificial intelligence information review method based on a distributed microservice architecture according to claim 8, characterized in that, The conflict resolution process includes: If there is a type conflict in the decision sequence within the time window, activate the hidden state backtracker and execute: i. Input the atomic decision sequence as the observation value into the hidden Markov model; ii. Decode the optimal hidden state sequence using the Viterbi algorithm and output a globally consistent decision.

10. An AI-based information review system based on a distributed microservice architecture, used to implement the AI-based information review method based on a distributed microservice architecture as described in any one of claims 1-9, characterized in that, include: Gateway layer module: Used to receive information to be reviewed and perform semantic parsing to extract lightweight decision context, including information type, semantic keyframes and urgency markers; Microservice instance cluster: used to bind the target microservice instance according to the decision context, and preload the context into the memory workspace of the target microservice instance; The audit engine module is deployed in each target microservice instance and configured to perform multimodal analysis based on the decision context and generate atomic audit decisions. The decision aggregation and conflict resolution module is used to aggregate and determine conflicts among multiple atomic review decisions in the same information flow within a time window, and output the final globally consistent review result.

Images