An artificial intelligence-based supply chain public opinion risk early warning method and system

Abstract

The application discloses a supply chain public opinion risk early warning method and system based on artificial intelligence, relates to the technical field of natural language processing and computer semantic analysis, and comprises the following steps: collecting manufacturing industry supply chain public opinion text data, extracting trigger words, generating a supply chain event object and a text evidence segment mapping, extracting a supply chain semantic relationship based on the supply chain event object, generating an event relationship graph with events as nodes, generating a constraint expression based on the event relationship graph and performing consistency checking, and outputting supply chain public opinion risk early warning information. The application finally outputs early warning information containing conflict source nodes or edges and evidence segments, so that the supply chain business side can locate according to event subject identification, confirm conflict types according to relationship edges, and review according to evidence segments, thereby forming a traceable early warning closed loop.

Description

Technical Field

[0001] This invention relates to the fields of natural language processing and computer semantic analysis, specifically to an artificial intelligence-based method and system for early warning of supply chain public opinion risks. Background Technology

[0002] In recent years, public opinion monitoring and early warning for supply chain operational risks have gradually evolved from manual assessment to data-driven approaches. Research and industry widely employ natural language processing, information extraction, and text mining technologies to aggregate and analyze publicly available texts from multiple sources, combining knowledge graphs and graph computing frameworks to achieve semantic associations across subjects and nodes. Against the backdrop of a continuous increase in high-frequency information sources such as corporate announcements, supplier notifications, logistics node announcements, and industry media reports, AI-based supply chain public opinion analysis is beginning to emphasize structured expression, traceable evidence, and verifiable judgment processes to support a closed loop of supply chain risk identification, event tracking, and handling.

[0003] Current supply chain public opinion risk early warning technologies primarily rely on sentiment, topic clustering, or popularity indicators, typically outputting text as a single score or label. This makes it difficult to directly correlate with supply chain elements and business constraints, resulting in semantic gaps between early warning results and actionable rules such as contracts, delivery, quality, and compliance. Traditional methods rarely analyze public opinion as calculable event objects, lacking a unified extraction and encapsulation mechanism for fields such as event trigger words, event subjects or objects, temporal semantics, spatial semantics, and quantitative semantics. This makes it difficult to form a consistent event expression across texts and time windows. Even when constructing relationship graphs, there is often a lack of strong binding between relationship edges and evidence fragments, making it impossible to pinpoint which textual evidence the conclusion originates during audit tracing, thus hindering consistency verification. Furthermore, existing technologies often lack structured expressions and consistency verification for supply chain constraints. When the same constraint object presents contradictory descriptions of delivery dates, embargo scope, recall batches, etc., in different texts or relational contexts, traditional public opinion scoring struggles to identify the source of conflict and cannot output conflict nodes, conflict edges, and evidence fragment combinations that can be used for business verification. Consequently, it fails to achieve a closed-loop early warning system driven by traceable evidence. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing supply chain public opinion semantic analysis and risk warning methods only output sentiment scores or topic tags, have difficulty mapping supply chain event objects and business elements, lack strong binding of evidence fragments and events or relationships leading to untraceable and unverifiable conclusions, lack consistency verification and conflict tracing mechanisms oriented towards contract, delivery, quality and compliance constraints, and have problems with how to structure multi-source public opinion texts into verifiable event relationship diagrams and constraint expressions and output traceable warning information accordingly.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a supply chain public opinion risk early warning method based on artificial intelligence, comprising collecting textual data of public opinion in the manufacturing supply chain, extracting trigger words, and generating a mapping of supply chain event objects and textual evidence fragments.

[0007] Based on supply chain event objects, semantic relationships in the supply chain are extracted, and an event relationship graph with events as nodes is generated.

[0008] Constraint representations are generated based on event relationship graphs and consistency checks are performed to output supply chain public opinion risk warning information.

[0009] As a preferred embodiment of the AI-based supply chain public opinion risk early warning method described in this invention, the collection of manufacturing supply chain public opinion text data includes: accessing text streams from manufacturing supply chain information sources and performing noise reduction and normalization processing. These information sources include company announcements, supplier notifications, logistics node announcements, industry media reports, and publicly available texts on social media platforms. The noise reduction and normalization processing includes deduplication and timestamp alignment of the accessed text stream, merging repeated statements by the same subject within the same time window into a single candidate text. The candidate text is then matched using an industry entity dictionary, unifying company names, factory names, route names, component names, and delivery node names into unique identifiers.

[0010] As a preferred embodiment of the AI-based supply chain public opinion risk early warning method described in this invention, the step of extracting trigger words and generating supply chain event objects and text evidence fragment mappings includes: performing syntactic dependency analysis and semantic role labeling on manufacturing supply chain public opinion text data; using event nouns and verb phrases as candidate trigger words; and identifying event type, event subject, event object, temporal semantics, spatial semantics, and quantitative semantics in the context surrounding the candidate trigger words. For causal, adversative, and conditional semantics, connectors and clause structures are identified and recorded as semantic logic tags. The set of fields corresponding to the same trigger word is encapsulated as a supply chain event object; for each field of the supply chain event object, the smallest evidence fragment in the original text is selected and its start and end positions are recorded to generate a text evidence fragment mapping.

[0011] As a preferred embodiment of the AI-based supply chain public opinion risk early warning method described in this invention, the extraction of supply chain semantic relationships based on supply chain event objects includes: using the event subject and event object in the supply chain event object as anchor points, performing relationship discrimination between the same candidate text and adjacent candidate texts within the same time window, including discrimination of supply and demand relationships, dependency relationships, substitution relationships, shared capacity relationships, and same-route relationships. For supply and demand relationships, directionality is identified based on predicate patterns, and supplier and demand identifiers are recorded. For dependency relationships, the dependency strength level is calculated based on constraint phrases and bound to relationship attributes. For substitution relationships, substitution subjects and substitution conditions are identified, and substitution conditions are written into relationship attributes.

[0012] As a preferred embodiment of the AI-based supply chain public opinion risk early warning method described in this invention, the generation of an event relationship graph with events as nodes includes: defining each supply chain event object as a node in the event relationship graph and assigning node attributes, including event type, event subject identifier, event object identifier, temporal semantics, spatial semantics, and quantitative semantics; defining supply chain semantic relationships as edges in the event relationship graph and assigning edge attributes, including relationship type, direction marker, intensity level, and corresponding text evidence fragment mapping; and performing aggregation on nodes within the same time window with the same event subject identifier, retaining all text evidence fragment mappings and recording aggregation rules during aggregation.

[0013] As a preferred embodiment of the AI-based supply chain public opinion risk early warning method described in this invention, the step of generating constraint expressions based on event relationship graphs and performing consistency verification includes: extracting semantic fragments related to contracts, delivery, quality, and compliance from the node and edge attributes of the event relationship graph, and mapping these semantic fragments to structured constraint items. Each structured constraint item includes a constraint object identifier, constraint type, constraint parameters, and constraint range. Delivery-related constraint types include delivery node names and delivery date descriptions; quality-related constraint types include recall descriptions and affected batch descriptions; and compliance-related constraint types include descriptions of prohibited regions and regulatory investigation status. When constraint parameters appear in the text as interval expressions, parameter standardization is performed based on temporal and quantitative semantics, and the standardization basis is bound to the text evidence fragment mapping. Structured constraint items are aggregated into constraint expressions according to event subject identifiers, and the reference relationship between the constraint expressions and the event relationship graph is recorded.

[0014] As a preferred embodiment of the AI-based supply chain public opinion risk early warning method of the present invention, the output of supply chain public opinion risk early warning information includes: performing an evidence consistency check on each structured constraint item in the constraint expression. The evidence consistency check is performed based on the text evidence fragment mapping bound to the structured constraint item, including field consistency and relation consistency. Field consistency verification checks whether the constraint object identifier, constraint parameters, and constraint range of the structured constraint item can be located and semantically matched in the corresponding text evidence fragment mapping. Relationship consistency verification checks whether the structured constraint item does not conflict with the supply and demand relationship, dependency relationship, and substitution relationship in the event relationship graph. When either field consistency or relation consistency fails, the structured constraint item is marked as inconsistent, and the conflict source node and conflict source edge are recorded. When the number of structured constraint items marked as inconsistent in the constraint expression reaches a preset proportion, supply chain public opinion risk early warning information is generated, and the corresponding supply chain event object identifier, structured constraint item identifier, and text evidence fragment mapping are written into the supply chain public opinion risk early warning information.

[0015] As a preferred embodiment of the AI-based supply chain public opinion risk early warning system of the present invention, it includes a public opinion event semantic extraction module, an event relationship graph construction module, and a constraint consistency risk early warning module.

[0016] The semantic extraction module for public opinion events is used to collect textual data of public opinion in the manufacturing supply chain, extract trigger words, and generate supply chain event objects and text evidence fragment mappings.

[0017] The event relationship graph construction module is used to extract semantic relationships in the supply chain based on supply chain event objects and generate an event relationship graph with events as nodes.

[0018] The constraint consistency risk warning module is used to generate constraint expressions based on the event relationship diagram and perform consistency verification, and output supply chain public opinion risk warning information.

[0019] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program as steps to implement an artificial intelligence-based supply chain public opinion risk early warning method.

[0020] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of an artificial intelligence-based supply chain public opinion risk early warning method.

[0021] The beneficial effects of this invention are as follows: By accessing, denoising, and standardizing public opinion texts related to the manufacturing supply chain, and performing trigger word recognition and event field extraction on candidate texts, unstructured public opinion texts are encapsulated into supply chain event objects containing event type, event subject identifier, event object identifier, and time, space, and quantity semantics. Simultaneously, a locationable text evidence fragment mapping is established for each field of the event object. This event object and evidence mapping are then used as a unified input and traceability basis for subsequent relationship discrimination and graph construction, supporting the consistency of event expression across information sources and expression methods. Finally, this invention achieves the beneficial effect of transforming public opinion descriptions into computable event units, enabling verification and traceability based on evidence fragments in subsequent verification stages.

[0022] By using the event subject and object within a supply chain event object as anchor points, semantic relationships such as supply and demand, dependency, and substitution are determined between candidate texts within the same time window and adjacent candidate texts. Relationship attributes such as direction markers, intensity levels, and substitution conditions are mapped and bound to corresponding textual evidence fragments, achieving integrated recording from relationship conclusions to evidence sources. Event objects are then mapped to nodes in the event relationship graph, and semantic relationships are mapped to edges. Within the same time window, events of the same subject and type are aggregated according to rules, and conflicting relationships are marked to form a structured, computable, and traceable event relationship graph input. Ultimately, this achieves the beneficial effect of making supply chain connections in scattered texts explicit and structured while preserving conflicts and sources.

[0023] By extracting semantic fragments related to contracts, delivery, quality, and compliance from the node and edge attributes of the event relationship graph, and mapping them into structured constraint items containing constraint object identifiers, constraint types, constraint parameters, and constraint scopes, and then aggregating them by event subject identifiers to form referable constraint expressions, the supply chain business constraints are transformed from natural language descriptions into verifiable structured expressions. Then, based on the mapping of textual evidence fragments, field consistency verification is performed, and combined with the supply and demand, dependency, and substitution relationships in the event relationship graph, relationship consistency verification is performed. This results in a calculable consistency or inconsistency determination for each structured constraint item, and when the triggering condition is met, the conflict source node and conflict source edge are written to support subsequent handling. Finally, this achieves the beneficial effect of upgrading the early warning output from generalized risk alerts to early warning results that can pinpoint the source of conflict and the evidence, enabling the supply chain business side to locate the conflict by event subject identifier, confirm the conflict type by relationship edge, and review the evidence fragments, forming a traceable early warning closed loop. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 The above is an overall flowchart of an artificial intelligence-based supply chain public opinion risk early warning method provided in Embodiment 1 of the present invention. Detailed Implementation

[0026] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0027] Example 1, referring to Figure 1 As an embodiment of the present invention, a supply chain public opinion risk early warning method based on artificial intelligence is provided, including: S1: Collect textual data on public opinion in the manufacturing supply chain, extract trigger words, and generate mappings of supply chain event objects and textual evidence fragments.

[0028] The text stream is accessed from manufacturing supply chain information sources, including company announcements, supplier notifications, logistics milestone announcements, industry media reports, and publicly available text on social media platforms. The noise reduction and normalization process involves deduplication and timestamp alignment of the accessed text stream, merging repeated statements from the same subject within the same time window into a single candidate text. The candidate text is then matched using an industry entity dictionary, unifying company names, factory names, route names, component names, and delivery node names into unique identifiers.

[0029] Furthermore, accessing text streams from manufacturing supply chain information sources includes configuring a unified access interface for different information sources and extracting the text body, title, publication time, and publication subject fields, with the publication time serving as the timestamp source. When the information source provides a webpage or announcement layout, the text area is extracted first, and navigation, disclaimers, and advertising fields are removed. When the information source provides structured fields, the original fields are retained as candidate metadata and associated with the text for later completion or validation when event fields are missing.

[0030] Furthermore, timestamp alignment includes using the publication time as the primary timestamp and extracting the event time as a secondary timestamp when a clear event occurrence time exists in the text. The time window is a time interval divided by a preset duration based on the primary timestamp. Text from the same subject within the same time window is sorted by primary timestamp and then merged as repeated paraphrases. During the merging of repeated paraphrases, the earliest appearing text is retained as the primary text, and the remaining text is recorded as a set of cited sources so that the original source can be located during subsequent evidence tracing.

[0031] It should be noted that a preferred scheme for industry entity dictionary matching specifically includes establishing a manufacturing supply chain entity dictionary that includes aliases for company names, factory names, flight routes, component names, and delivery node names. After performing entity recognition on candidate text, the matched entities are normalized to the standard names in the entity dictionary, and the normalization result is output as a unique identifier. When the same text matches multiple possible standard names, disambiguation is performed based on regional terms, product terms, or organizational hierarchy terms in the text context, and the disambiguation basis is recorded as traceable matching rule information.

[0032] Syntactic dependency analysis and semantic role labeling are performed on the text data of public opinion concerning the manufacturing supply chain. Event nouns and verb phrases are used as candidate trigger words. Event type, event subject, event object, temporal semantics, spatial semantics, and quantitative semantics are identified within the context surrounding the candidate trigger words. Specifically, temporal, spatial, and quantitative semantics are obtained through time expression identification, place name or facility name identification, and numerical value and unit identification, respectively. For causal, adversative, and conditional semantics, conjunctions and clause structures are identified and recorded as semantic logic tags. The set of fields corresponding to the same trigger word is encapsulated as a supply chain event object. For each field of the supply chain event object, the smallest piece of evidence in the original text is selected, and its start and end positions are recorded to generate a text evidence fragment mapping.

[0033] Furthermore, the determination of candidate trigger words involves screening predicate headwords and their governing verb phrases from the syntactic dependency analysis results as candidate trigger words, and setting a minimum context window for each candidate trigger word to extract a candidate event fragment composed of several words before and after the trigger word. When the event subject and event object cannot be identified simultaneously within a candidate event fragment, the candidate trigger word is marked as a trigger word to be confirmed and enters the subsequent cross-sentence completion process. By searching for the same referent entity in adjacent sentences of the same paragraph to complete the subject or object, the supply chain event object is then constructed.

[0034] Furthermore, the selection of the minimum evidence fragment involves targeting each field of the supply chain event object, prioritizing the shortest continuous text fragment containing the key entity or key value of that field as the evidence fragment, and recording the start and end positions of this evidence fragment in the candidate text. When field information is inferred from across sentences, the evidence fragments corresponding to each sentence are recorded separately and associated with the same field identifier, forming a field-level evidence set. The text evidence fragment mapping includes at least the field name, the start and end positions of the evidence fragment, the text content of the evidence fragment, and the source candidate text identifier to ensure that the event object can be verified and traced.

[0035] It should be noted that this step focuses on public opinion in the manufacturing supply chain. First, candidate texts are generated through information source access and noise reduction alignment. Then, entity dictionary normalization is used to ensure consistent identification of supply chain elements. Subsequently, event trigger words are extracted based on syntactic dependency and semantic role labeling, and supply chain event objects are constructed. Finally, the minimum evidence fragment mapping is used to achieve field traceability and verifiability, providing reliable input for subsequent relationship modeling and constraint verification.

[0036] S2: Extract semantic relationships from supply chain event objects to generate an event relationship graph with events as nodes.

[0037] Using the event subject and object within the supply chain event object as anchor points, relationship discrimination is performed between the same candidate text and adjacent candidate texts within the same time window, including supply and demand relationships, dependency relationships, substitution relationships, shared capacity relationships, and same-route relationships. For supply and demand relationships, directionality is identified based on predicate patterns, and supplier and demand identifiers are recorded. For dependency relationships, dependency strength levels are calculated based on constraint phrases and bound to relationship attributes. For substitution relationships, substitution subjects and substitution conditions are identified, and substitution conditions are written into relationship attributes.

[0038] Furthermore, adjacent candidate texts are those that are adjacent in timestamp within the same time window. Relationship determination is based on the co-occurrence of the event subject identifier and the event object identifier as the trigger condition for candidate relationships. When the event subject identifier or event object identifier is missing in the current candidate text, the text evidence fragment mapping generated by S1 is invoked to retrieve the corresponding referent entity in the adjacent candidate text to complete the subject or object before performing relationship determination. The text evidence fragment mapping on which the completion is based is then bound to the corresponding supply chain semantic relationship. Relationship determination for shared production capacity is based on the identification of expressions related to shared production lines, shared equipment, contract manufacturing capacity, or capacity allocation in the candidate text, and the production capacity subject identifier and the shared scope field are written into the relationship attribute. Relationship determination for the same flight route is based on the matching of referents in the flight route name or segment description in the candidate text, and the text evidence fragment mapping corresponding to the matched flight route name is bound to the relationship attribute.

[0039] The predicate pattern should include at least a set of supply-related predicates and a set of procurement-related predicates related to supply and demand semantics. During directional identification, the assignment rules for supplier and demand identifiers are determined based on the semantic role labeling results of the sentences containing the predicates. When negative or conditional structures appear, negative or conditional tags are written into the relation attributes. When the same entity is identified as having opposite supply and demand relationships within the same time window, two supply and demand relationship edges are retained and their corresponding text evidence fragments are mapped to them respectively, so that the source of conflict can be located in the subsequent consistency verification stage.

[0040] Furthermore, the calculation of dependency strength levels includes hierarchical mapping of constraint phrases and outputting discrete levels. Constraint phrases at least include expressions related to unique source, critical component, irreplaceable, certification cycle, and limited capacity. When multiple constraint phrases appear simultaneously, they are merged into a single dependency strength level according to a preset priority and written into the relation attribute. Simultaneously, the text evidence fragment mapping corresponding to the constraint phrase triggering that level is recorded. When the dependency strength level cannot be determined, the dependency is marked as a pending dependency, and its candidate phrase set is retained for subsequent updates.

[0041] It should be noted that the identification of substitute subjects and substitute conditions includes extracting entity pairs corresponding to the transfer predicates or replacement predicates related to the substitution from the candidate text, and writing the identifier of the substituted subject and the identifier of the substitute subject into the relation attribute. Substitute conditions include at least time conditions, spatial conditions, or delivery node conditions, and are written into the relation attribute in the form of structured fields. When a substitute condition appears only in a vague expression, the original vague fragment is retained as conditional evidence and bound to the text evidence fragment mapping to ensure the traceability of the substitution relationship.

[0042] Each supply chain event object is defined as a node in the event relationship graph and assigned node attributes, including event type, event subject identifier, event object identifier, temporal semantics, spatial semantics, and quantitative semantics. Supply chain semantic relationships are defined as edges in the event relationship graph and assigned edge attributes, including relationship type, direction marker, strength level, and corresponding text evidence fragment mapping. Aggregation is performed on nodes within the same time window with the same event subject identifier. During aggregation, all text evidence fragment mappings are preserved, and the aggregation rules are recorded.

[0043] Furthermore, node aggregation is based on the premise that the event subject identifiers are the same and the event types are the same, and temporal semantic consistency is used as an aggregation constraint. The aggregation rules at least include the set of node identifiers participating in the aggregation, the aggregation time window identifier, and the method for merging the aggregated node attributes. When merging node attributes, multiple values ​​are retained for quantitative semantics, and the corresponding text evidence fragment mappings are recorded. For edge attributes with conflicting relation types or direction markers, multiple edges are retained, and the conflict markers and their text evidence fragment mappings are recorded in the edge attributes.

[0044] It should be noted that this step, based on the supply chain event objects output by S1, establishes a verifiable semantic relationship layer around the event subject and the event object. By performing relationship discrimination on adjacent candidate texts within the same time window, and combining semantic role labeling and predicate patterns to determine supply and demand direction, dependency strength, and substitution conditions, the relationships acquire directional, strength attributes, and conditional fields. Simultaneously, the relationships are mapped and bound to corresponding textual evidence fragments to ensure subsequent traceability and verifiability. Furthermore, the event objects and semantic relationships are mapped to nodes and edges in an event relationship graph. Aggregation rules are used to merge events of the same subject and type, and conflicting relationships are retained and marked, forming a structured and computable graph input, providing a stable basis for subsequent constraint expression generation and consistency verification.

[0045] S3: Generate constraint expressions based on event relationship graphs and perform consistency checks, outputting supply chain public opinion risk warning information.

[0046] Semantic fragments related to contracts, delivery, quality, and compliance are extracted from the node and edge attributes of the event relationship graph. These semantic fragments are then mapped to structured constraint items, which include the constraint object identifier, constraint type, constraint parameters, and constraint scope. Delivery-related constraint types include delivery node names and delivery date descriptions; quality-related constraint types include recall descriptions and batch impact descriptions; and compliance-related constraint types include descriptions of restricted areas and regulatory investigation status. When constraint parameters appear as intervals in the text, parameter standardization is performed based on temporal and quantitative semantics, and the standardization basis is bound to the textual evidence fragment mapping. Structured constraint items are aggregated into constraint expressions based on event subject identifiers, and the reference relationships between constraint expressions and the event relationship graph are recorded.

[0047] Furthermore, parameter standardization is used to unify interval parameters into comparable parameters under the same measurement benchmark, so that subsequent calculations of field consistency and relational consistency can be verified using the same scale. The standardization results are also mapped one-to-one with the textual evidence fragments that triggered the standardization for traceability. Interval parameter standardization is expressed as: in, Indicates the first The standardized constraint parameters of each structured constraint term take real numbers. , Indicates the first The lower and upper bounds of the intervals extracted from the text by each structured constraint term.

[0048] Indicates the first The unit identifier of the parameters of each structured constraint item. This represents a unit conversion function, which will... The corresponding units are converted to the system's unified base units, and the output is a positive real number conversion factor. Indicates the structured constraint index identifier, indicating the first... One constraint item.

[0049] For each structured constraint in the constraint representation, an evidence consistency check is performed. This check is based on the text evidence fragment mapping bound to the structured constraint and includes field consistency and relation consistency. Field consistency verifies whether the constraint object identifier, constraint parameters, and constraint scope of the structured constraint can be located and semantically matched in the corresponding text evidence fragment mapping. Relationship consistency verifies whether the structured constraint does not conflict with the supply and demand relationships, dependency relationships, and substitution relationships in the event relationship graph.

[0050] Furthermore, field consistency is determined by the alignment of structured constraint fields with evidence fragments to form a field matching quantity, while relationship consistency is determined by the relationship edges associated with the constraint object identifier in the event relationship graph to form a conflict quantity. The combined consistency score is used to determine whether the structured constraint can be judged as consistent or inconsistent, and will be used for subsequent alert triggering. To unify the verification results into comparable judgment quantities, a consistency score is constructed. The combined score of field consistency and relationship consistency is expressed as follows: in, Indicates the first The consistency score of each structured constraint item, with a value range of (0,1). This indicates the matching indicator value of the constraint object identifier field. It is 1 when the constraint object identifier can be located in the bound text evidence fragment mapping and semantically matches, and 0 otherwise. This indicates that the constraint parameter field matches the indicator value, when the standardized parameter... If the value and unit can be derived from the bound evidence fragment and are not contradictory, take 1; otherwise, take 0. This indicates the constraint range field matching indicator. It is set to 1 when the constraint range can be located in the bound evidence fragment mapping and is consistent with the range description semantics, otherwise it is set to 0. This represents the relationship conflict count, and its value is a non-negative integer. It indicates the number of edges in the event relationship graph that conflict with the semantic constraint of this structured constraint among the supply-demand, dependency, and substitution relationships associated with and included in the verification process.

[0051] Furthermore, Conflict determination is performed according to the constraint type of the structured constraint item. When the constraint type is a delivery date description, if there is a supply-demand relationship edge in the event relationship graph pointing to the same constraint object identifier, and its direction marker indicates that the constraint object identifier is the demand side identifier, and the delay duration obtained by standardization of the constraint parameter of the structured constraint item contradicts the on-time delivery or delivered statement in the text evidence fragment mapping bound to the supply-demand relationship edge, then it is counted as a conflict edge. When the constraint type is a restricted area description, if there is a same route relationship edge or supply-demand relationship edge in the event relationship graph, and the text evidence fragment mapping bound to the edge indicates that the route or delivery node involves a restricted area, and the constraint scope of the structured constraint item indicates that it does not involve the area, then it is counted as a conflict edge. When the constraint type is a recall description or an affected batch description, if there is a substitution relationship edge in the event relationship graph and its substitution condition indicates that the substitution effective time is earlier than the recall occurrence time, and the text evidence fragment mapping still associates the substituted subject identifier as an affected batch object, then it is counted as a conflict edge.

[0052] When either field consistency or relation consistency fails, the structured constraint item is marked as inconsistent, and the conflict source node and conflict source edge are recorded. When the number of structured constraint items marked as inconsistent in the constraint expression reaches a preset proportion, a supply chain public opinion risk warning message is generated, and the corresponding supply chain event object identifier, structured constraint item identifier, and text evidence fragment mapping are written into the supply chain public opinion risk warning message.

[0053] Furthermore, the preset ratio trigger uses a ratio calculation method based on the total number of structured constraint items. When issuing an alert, in addition to writing the identifier, it also writes the identifiers of the conflict source node and conflict source edge that caused the inconsistency, to guide subsequent manual review, supply chain element verification, or strategy handling. The inconsistency ratio and alert trigger criteria are expressed as follows: in, This represents the proportion of inconsistencies in the constraint expression, and its value ranges from [0,1]. This represents the set of structured constraint terms in a constraint expression. This represents the total number of structured constraint terms and takes a positive integer value. This indicates an indicator function; the value is 1 if the condition within the parentheses is true, and 0 otherwise. This represents the consistency threshold for constraint terms, with a value range of (0, 1). When... The time will be the first Several structured constraint terms were determined to be inconsistent. This invention, without historical calibration, will... Set to 0.80.

[0054] when Timely generation of supply chain public opinion risk warning information, This represents a preset ratio threshold, with a value range of (0,1]. This invention will, when historical calibration is not performed, [implement this function]. Set to 0.25, and allow configuration within the range of [0.15, 0.35] based on the noise level of the information source. Specifically, when the proportion of publicly available text on social media platforms in the manufacturing supply chain information source exceeds the preset proportion of 0.6, [the following will occur]. The ratio will be increased to 0.30. When the ratio of enterprise announcements and supplier notifications exceeds the preset ratio of 0.6, [the ratio will be adjusted accordingly]. The value has been reduced to 0.20 to ensure that the constraint expression under the same event subject identifier has a consistent triggering caliber regardless of the source composition.

[0055] Furthermore, when triggered When generating supply chain public opinion risk warning information, the system must include at least the corresponding supply chain event object identifier, the identifier of the structured constraint item that is determined to be inconsistent, the text evidence fragment mapping bound to each inconsistent item, and the conflict source node and conflict source edge identifier. This allows for the execution of a handling process on the supply chain business side, which involves locating the risk source by event subject identifier, confirming the conflict type by relation edge, and reviewing the text basis by evidence fragment, forming a traceable warning closed loop.

[0056] It should be noted that this step uses an event relationship diagram as an intermediate representation, mapping the semantics related to contracts, delivery, quality, and compliance into structured constraints, and achieving cross-text and cross-unit comparable verification through parameter standardization. Subsequently, field consistency is constructed based on the mapping of textual evidence fragments, and relationship consistency is constructed by combining supply and demand, dependency, and substitution relationships in the event relationship diagram, forming a calculable consistency score and conflict count. Finally, an alert is triggered based on the inconsistency ratio and written into the conflict source node, edge, and evidence fragment. Compared with public opinion methods that only output sentiment scores or topic tags, this achieves an alert closed loop where evidence is locatable, relationships are verifiable, and conflicts are traceable, enabling the alert results to directly support the review and handling on the supply chain business side.

[0057] Example 2, an embodiment of the present invention, provides an artificial intelligence-based supply chain public opinion risk early warning system, including a public opinion event semantic extraction module, an event relationship graph construction module, and a constraint consistency risk early warning module.

[0058] The semantic extraction module for public opinion events is used to collect textual data of public opinion in the manufacturing supply chain, extract trigger words, and generate supply chain event objects and text evidence fragment mappings.

[0059] The event relationship graph construction module is used to extract semantic relationships in the supply chain based on supply chain event objects and generate an event relationship graph with events as nodes.

[0060] The constraint consistency risk warning module is used to generate constraint expressions based on the event relationship diagram and perform consistency verification, and output supply chain public opinion risk warning information.

Claims

1. A supply chain public opinion risk early warning method based on artificial intelligence, characterized in that, include: Collect textual data on public opinion in the manufacturing supply chain, extract trigger words, and generate mappings of supply chain event objects and text evidence fragments; Based on supply chain event objects, semantic relationships in the supply chain are extracted to generate an event relationship graph with events as nodes; Constraint representations are generated based on event relationship graphs and consistency checks are performed to output supply chain public opinion risk warning information.

2. The supply chain public opinion risk early warning method based on artificial intelligence as described in claim 1, characterized in that: The collected textual data on public opinion in the manufacturing supply chain includes... The text stream is accessed from manufacturing supply chain information sources and then processed for noise reduction and normalization. These sources include company announcements, supplier notifications, logistics node announcements, industry media reports, and public texts on social media platforms. The noise reduction and normalization process includes deduplication and timestamp alignment of the incoming text stream, merging repeated paraphrases of the same subject within the same time window into a single candidate text; and matching the candidate text with an industry entity dictionary to unify company names, factory names, route names, component names, and delivery node names into a unique identifier.

3. The supply chain public opinion risk early warning method based on artificial intelligence as described in claim 2, characterized in that: The process of extracting trigger words and generating supply chain event objects and text evidence fragment mappings includes... Syntactic dependency analysis and semantic role labeling are performed on the text data of public opinion in the manufacturing supply chain. Event nouns and verb phrases are used as candidate trigger words. Event type, event subject, event object, time semantics, spatial semantics and quantitative semantics are identified in the context around the candidate trigger words. For causal, adversative, and conditional semantics, identify conjunctions and clause structures and record them as semantic logic tags; The set of fields corresponding to the same trigger word is encapsulated into a supply chain event object. For each field of the supply chain event object, the smallest piece of evidence in the original text is selected and the start and end positions are recorded to generate a text evidence fragment mapping.

4. The supply chain public opinion risk early warning method based on artificial intelligence as described in claim 3, characterized in that: The extraction of supply chain semantic relationships based on supply chain event objects includes... Using the event subject and event object in the supply chain event as anchor points, relationship discrimination is performed between the same candidate text and adjacent candidate texts within the same time window, including the discrimination of supply and demand relationship, dependency relationship, substitution relationship, shared capacity relationship and same route relationship; For supply and demand relationships, directionality is identified based on predicate patterns, and supplier and demand identifiers are recorded. For dependencies, the dependency strength level is calculated based on the constraint phrase and bound to the relation attribute; For substitution relationships, identify the substitution subject and substitution conditions, and write the substitution conditions into the relationship attributes.

5. The supply chain public opinion risk early warning method based on artificial intelligence as described in claim 4, characterized in that: The generation of the event relationship graph with events as nodes includes, Each supply chain event object is defined as a node in the event relationship diagram and assigned node attributes. The node attributes include event type, event subject identifier, event object identifier, time semantics, spatial semantics, and quantity semantics. The semantic relationships of the supply chain are defined as edges in an event relationship graph and assigned edge attributes, which include relationship type, direction label, strength level and corresponding text evidence fragment mapping. Aggregation is performed on nodes that are in the same time window and have the same event subject identifier. During aggregation, all text evidence fragment mappings are preserved and aggregation rules are recorded.

6. The supply chain public opinion risk early warning method based on artificial intelligence as described in claim 5, characterized in that: The process of generating constraint representations based on event relationship graphs and performing consistency checks includes, Semantic fragments related to contracts, delivery, quality, and compliance are extracted from the node and edge attributes of the event relationship graph, and the relevant semantic fragments are mapped to structured constraint items. The structured constraint items include constraint object identifier, constraint type, constraint parameters, and constraint scope. Delivery-related constraints include delivery node names and delivery date descriptions; quality-related constraints include recall descriptions and affected batch descriptions; and compliance-related constraints include descriptions of restricted areas and regulatory investigation status. When constraint parameters appear in the text as intervals, parameter standardization is performed based on temporal and quantitative semantics, and the standardization basis is bound to the text evidence fragment mapping. Structured constraint items are aggregated into constraint expressions based on event subject identifiers, and the reference relationships between constraint expressions and event relationship diagrams are recorded.

7. The supply chain public opinion risk early warning method based on artificial intelligence as described in claim 6, characterized in that: The output supply chain public opinion risk warning information includes For each structured constraint item in the constraint expression, an evidence consistency check is performed. The evidence consistency check is performed based on the text evidence fragment mapping bound to the structured constraint item, including field consistency and relation consistency. Field consistency verification checks whether the constraint object identifier, constraint parameters and constraint range of the structured constraint items can be located and semantically matched in the corresponding text evidence fragment mapping; relation consistency verification checks whether the structured constraint items do not conflict with the supply and demand relationship, dependency relationship and substitution relationship in the event relationship diagram. When either field consistency or relation consistency fails, the structured constraint item is marked as inconsistent and the conflict source node and conflict source edge are recorded. When the number of structured constraint items marked as inconsistent in the constraint expression reaches a preset proportion, supply chain public opinion risk warning information is generated and the corresponding supply chain event object identifier, structured constraint item identifier and text evidence fragment mapping are written into the supply chain public opinion risk warning information.

8. An artificial intelligence-based supply chain public opinion risk early warning system, employing the artificial intelligence-based supply chain public opinion risk early warning method as described in any one of claims 1 to 7, characterized in that: This includes a semantic extraction module for public opinion events, an event relationship graph construction module, and a constraint consistency risk warning module; The semantic extraction module for public opinion events is used to collect textual data of public opinion in the manufacturing supply chain, extract trigger words, and generate supply chain event objects and text evidence fragment mappings. The event relationship graph construction module is used to extract semantic relationships in the supply chain based on supply chain event objects and generate an event relationship graph with events as nodes. The constraint consistency risk warning module is used to generate constraint expressions based on the event relationship diagram and perform consistency verification, and output supply chain public opinion risk warning information.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the supply chain public opinion risk early warning method based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the supply chain public opinion risk early warning method based on any one of claims 1 to 7.

Images