A method and system for extracting techno-economic risk events based on monitored entities

Abstract

This application provides a method and system for extracting technical and economic risk events based on monitored entities, belonging to the field of entity extraction technology. The method includes: performing entity recognition on multi-source scientific and technological texts, comparing and filtering technical and economic monitoring entities, and outputting a candidate monitoring entity list; extracting semantic segments using the candidate monitoring entity list as anchors, and determining event types based on trigger word matching rules; performing dependency parsing and semantic role labeling on the trigger word set, constructing a standardized event structure, extracting semantic relationships, and outputting a risk event candidate structure set; identifying chapter-level main events, aggregating paragraph-level related events to form a master-slave event structure; assembling event chains based on the master-slave event structure, and performing standardized output. This application can solve the technical problem of poor reliability in the extraction of economic risk events in the prior art, achieving the technical effect of improving the reliability of extraction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of entity extraction, and particularly relates to a technical and economic risk event extraction method and system based on monitoring entities. BACKGROUND

[0002] With the continuous development of technology intelligence systems and intelligent text processing technologies, higher requirements are put forward for the automatic extraction and structured analysis of important risk events in the technical and economic field.

[0003] At present, the existing event extraction method mainly depends on a general natural language processing model and a general event type system, and although it has certain event recognition and structured ability in open fields and news public opinion scenes, it is difficult to cover high-risk event types when facing technical and economic security scenes, and lacks an extraction perspective centered on monitoring entities, and often cannot form a systematic tracking of entity risk dynamics.

[0004] To sum up, in the prior art, due to the lack of a professional event type system for the technical and economic security field and a monitoring entity-driven extraction mechanism, it is difficult to accurately extract event structures with industry characteristics and high-risk directionality, which further affects the technical effect and system reliability of intelligent identification, trend early warning and knowledge graph construction of key entity risk situation in the technical and economic field. SUMMARY

[0005] The purpose of the present application is to provide a technical and economic risk event extraction method and system based on monitoring entities, to solve the technical problem in the prior art that due to the lack of a professional event type system for the technical and economic security field and a monitoring entity-driven extraction mechanism, it is difficult to accurately extract event structures with industry characteristics and high-risk directionality, which further affects the technical effect and system reliability of intelligent identification, trend early warning and knowledge graph construction of key entity risk situation in the technical and economic field.

[0006] In view of the above problems, the present application provides a technical and economic risk event extraction method and system based on monitoring entities.

[0007] In a first aspect, the application provides a technical and economic risk event extraction method based on monitoring entities, which is implemented by a technical and economic risk event extraction system based on monitoring entities, and includes the following steps: performing entity recognition on multi-source technical texts, extracting technical and economic monitoring entities, performing entity comparison and screening on the technical and economic monitoring entities based on a preset monitoring entity library, and outputting a candidate monitoring entity list; performing semantic segment extraction with the candidate monitoring entity list as an anchor point, performing event trigger word recognition, and performing event type determination based on trigger word matching rules; performing dependency syntax analysis and semantic role labeling on the identified trigger word set, constructing a standardized event structure, extracting semantic relationships, and outputting a risk event candidate structure set; identifying a chapter-level main event based on the risk event candidate structure set, aggregating paragraph-level associated events, forming a master-slave event structure, assembling an event chain according to the master-slave event structure, and performing standardized output.

[0008] Preferably, the technical and economic risk event extraction method based on monitoring entities further includes the following steps: obtaining historical technical and economic risk events, determining event type identifiers, listing trigger words according to the historical technical and economic risk events, obtaining a trigger word mode set, defining participating elements through the historical technical and economic risk events, obtaining argument role templates, associating semantic relationships based on the event type identifiers, the trigger word mode set, and the argument role templates, obtaining semantic relationship rules, and constructing a technical and economic risk event knowledge system.

[0009] Preferably, the technical and economic risk event extraction method based on monitoring entities further includes the following steps: dividing the multi-source technical texts into sentence units according to semantic boundaries to obtain sentence-level segmentation, performing morphological analysis on the sentence-level segmentation to obtain entity clues, performing dependency tree structure analysis on the sentence-level segmentation to obtain sentence semantic relationships, and performing error correction and deduplication on the entity clues and sentence semantic relationships, and combining to obtain multi-dimensional text information.

[0010] Preferably, the technical and economic risk event extraction method based on monitoring entities further includes the following steps: performing entity literal matching on the multi-dimensional text information according to the preset monitoring entity library to obtain fuzzy matching entities, performing entity expansion recognition on the multi-dimensional text information based on the fuzzy matching entities to obtain expanded matching entities, identifying the expanded matching entities in the multi-dimensional text information, and extracting context windows before and after the text with the entity identifier as the center to output a candidate monitoring entity list.

[0011] Preferably, the technical and economic risk event extraction method based on monitoring entities further comprises: performing dependency syntax analysis on the trigger word set, and outputting a dependency structure tree; performing semantic role labeling according to the dependency structure tree, and obtaining semantic roles; classifying and normalizing the semantic roles based on the argument role template, and forming the standardized event structure; and constructing logical, causal, and temporal semantic relationships between the arguments of the trigger word set according to the standardized event structure, and outputting the risk event candidate structure set.

[0012] Preferably, the technical and economic risk event extraction method based on monitoring entities further comprises: performing semantic segment extraction on the candidate monitoring entity list, and obtaining display markers; introducing natural language prompts to guide trigger word prediction, obtaining the starting position and ending position of the trigger word, and obtaining the trigger word corresponding to the event; and performing bidirectional mapping determination on the trigger word event based on the technical and economic risk event knowledge system, and outputting the trigger word set and the event type determination result.

[0013] Preferably, the technical and economic risk event extraction method based on monitoring entities further comprises: archiving the risk event candidate structure set according to the monitoring entity dimension and the paragraph position dimension, and constructing a chapter-level event candidate index table; performing fusion scoring and sorting based on three types of strategies according to the chapter-level event candidate index table, and selecting a risk theme main event; performing paragraph-level associated event aggregation based on three types of association strategies according to the risk theme main event, and obtaining an associated event set; and performing master-slave event structure output construction based on the risk theme main event and the associated event set, and obtaining the master-slave event structure.

[0014] Preferably, the technical and economic risk event extraction method based on monitoring entities further comprises: the three types of strategies include a title semantic guidance strategy, an event influence evaluation strategy, and a position priority strategy; and the three types of association strategies include an entity coincidence type association, an event causal chain association, and a semantic similarity type association.

[0015] Preferably, the technical and economic risk event extraction method based on monitoring entities further comprises: constructing the event chain according to the master-slave event structure based on a chain construction dimension, wherein the chain construction dimension includes a time sequence chain, an entity dominant chain, and a causal logic chain; and performing standardized representation based on the event chain.

[0016] In a second aspect, the application also provides a technical and economic risk event extraction system based on monitoring entities, which is used to execute the technical and economic risk event extraction method based on monitoring entities as described in the first aspect, and comprises: an entity identification module, which is used to perform entity identification on multi-source technical texts, extract technical and economic monitoring entities, and perform entity comparison and screening on the technical and economic monitoring entities based on a preset monitoring entity library to output a candidate monitoring entity list; a semantic segment extraction module, which is used to perform semantic segment extraction with the candidate monitoring entity list as an anchor point, perform event trigger word identification, and perform event type determination based on trigger word matching rules; a semantic role labeling module, which is used to perform dependency syntax analysis and semantic role labeling on the identified trigger word set, construct a standardized event structure, and extract semantic relationships to output a risk event candidate structure set; a master-slave event structure forming module, which is used to identify a chapter-level main event based on the risk event candidate structure set, aggregate a paragraph-level associated event, and form a master-slave event structure; and a standardized output module, which is used to assemble an event chain according to the master-slave event structure and perform standardized output.

[0017] The technical solutions provided in the application have at least the following technical effects or advantages: by achieving the technical target of constructing a structured risk event set and establishing an intelligent extraction process of a multi-dimensional event chain with monitoring entities as the center, the technical effects of improving the identification accuracy of technical and economic security events, enhancing the depth of event understanding, and supporting risk decision assistance capabilities in multiple scenarios are achieved.

[0018] The above description is only a summary of the technical solutions of the application. In order to more clearly understand the technical means of the application, the specific embodiments of the application can be implemented in accordance with the content of the description, and in order to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following specific embodiments of the application are described. It should be understood that the content described in this section is not intended to identify key or important features of embodiments of the application, nor is it intended to limit the scope of the application. Other features of the application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are only exemplary, and those skilled in the art can obtain other drawings without creative labor on the basis of the provided drawings.

[0020] Figure 1 The flowchart of the technical and economic risk event extraction method based on monitoring entities of the application.

[0021] Figure 2A structural schematic diagram of a technical and economic risk event extraction system based on a monitoring entity is provided in the present application.

[0022] Reference signs: entity recognition module 11, semantic fragment extraction module 12, semantic role labeling module 13, master-slave event structure forming module 14, standardized output module 15. DETAILED DESCRIPTION

[0023] The present application provides a technical and economic risk event extraction method and system based on a monitoring entity, which solves the technical problems in the prior art that due to the lack of a professional event type system for the technical and economic security field and a monitoring entity driven extraction mechanism, the event structure with industry characteristics and high risk direction cannot be accurately extracted, which further affects the technical effect and system reliability of the intelligent identification of key entity risk situation, trend warning and knowledge graph construction and other applications in the technical and economic field. The technical target of constructing a structured risk event set and establishing an intelligent extraction process of a multi-dimensional event chain with the monitoring entity as the center is achieved, and the technical effects of improving the identification accuracy of technical and economic security events, enhancing the depth of event understanding and supporting the risk decision assistance capability in multiple scenarios are achieved.

[0024] Hereinafter, the technical solutions in the present application will be described clearly and completely with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application. In addition, it should be noted that, for convenience of description, only parts related to the present application are shown in the drawings, not all.

[0025] Embodiment one, please refer to the accompanying Figure 1 The present application provides a technical and economic risk event extraction method based on a monitoring entity, which is applied to a technical and economic risk event extraction system based on a monitoring entity, and specifically includes the following steps:

[0026] Perform entity recognition on multi-source technical texts, extract technical and economic monitoring entities, and perform entity comparison and screening on the technical and economic monitoring entities based on a preset monitoring entity library to output a candidate monitoring entity list.

[0027] Specifically, from multiple sources of technology-related text data, key entities with independent semantics are identified through natural language processing technology. The sources of multi-source technology text may include news reports, technology papers or enterprise announcements, etc., which have characteristics such as large structural differences and information distribution. Technical and economic monitoring entities refer to specific objects in the intersection of technology and economy that have risk attention value, such as enterprise names, university research institutions, industrial projects, key technical equipment, etc., which may be the core carrier or source of potential risk events. The identified text entities are matched and corrected with the existing standardized entity set to remove ambiguity or redundant information, such as merging “Company A” and “Company A Technology Co., Ltd.” into the same standard entity “Company A”. The screening process further filters the matching results by combining risk labels, historical event frequencies, and appearance frequencies of entities, etc., to retain the most representative monitoring objects. The final output of the candidate monitoring entity list is a set of entities with strong certainty and high risk relevance in the current text, and each entity is also labeled with its position and surrounding semantics in the text, providing accurate anchors for subsequent event recognition. For example, from a technology news about chip supply chain, the system may identify multiple entities such as “Entity A”, “Entity B”, “Entity C”, etc. After comparison and screening, only “Entity A” and “Entity B” are retained as candidate monitoring entities, focusing on objects with real technical and economic risk value.

[0028] Semantic segment extraction is performed based on the candidate monitoring entity list as the anchor, event trigger word identification is performed, and event type determination is performed based on trigger word matching rules.

[0029] Specifically, the occurrence position of the candidate monitoring entity in the text is taken as the core reference point, and the context before and after it is expanded by several words, and the sentence or phrase fragment with complete semantic expression is intercepted to capture the occurrence clues of potential events. The length of the semantic fragment is generally set according to the syntactic structure or the fixed window size, for example, the context content is extracted with a window of thirty Chinese characters before and after each entity. Then, event trigger word recognition is performed, that is, the key words representing the occurrence or state change of the event, such as "release", "terminate cooperation", etc., which can reflect the important nodes of the monitoring entity participating in the technical and economic activities, are identified in the semantic fragment. Then, the identified trigger words are matched with the pre-constructed trigger word pattern set, each trigger word pattern corresponds to a specific event type, such as "event D" and "event E" matching R-class events, and "event F" and "event G" matching S-class events, and the rule comparison mechanism can determine the event type represented by the trigger word. The overall process forms a closed loop chain from candidate entity positioning semantic range, event recognition triggered by semantic range, and event classification determined by trigger word rules, so that event extraction has both pertinence and clear structure. For example, when "a company" is identified as a monitoring entity and the semantic fragment "due to a measure" is extracted, "a measure" can be identified as a trigger word, and according to the matching rule, it is determined as a "a measure type" event, thereby effectively realizing the typing of risk events.

[0030] The identified trigger word set is subjected to dependency syntax analysis and semantic role labeling, a standardized event structure is constructed, and semantic relationships are extracted, and a risk event candidate structure set is output.

[0031] Specifically, a dependency relationship graph is established for each sentence containing a trigger word, to clearly determine the modification, domination and connection relationship between the words in the sentence, such as the dependency path between the verb and the subject, object, time adverbial, so as to determine the subject and object associated with the trigger word. Semantic role labeling is to assign semantic functions to the words in the sentence, such as identifying which word represents the initiator, performer, affected object, time of occurrence, location of occurrence, etc., so as to realize the semantic structuring of event elements. Constructing a standardized event structure is to encode and organize the trigger word and its related semantic roles according to the pre-set event template, forming a unified event triple or quintuple expression method, to ensure that the event structure has comparability and can be used in subsequent computing and processing format. Then, the semantic relationship is extracted, which refers to further identifying the logical relationship, causal relationship and time sequence relationship between different arguments in the standardized structure. Finally, the risk event candidate structure set is output, which is a set of preliminary complete risk event units extracted from the original text, each event unit contains core entities, trigger behaviors and argument information, which can be used for subsequent main event recognition, chain construction and graph reasoning tasks.

[0032] Based on the set of risk event candidate structures, a chapter-level main event is identified, and paragraph-level associated events are aggregated to form a master-slave event structure.

[0033] Specifically, a plurality of standardized event structures extracted from the text are taken as an input set, and each event structure includes a monitoring entity, a trigger word, an event type, and corresponding argument information, which can reflect risk information in the text from different perspectives. Identifying a chapter-level main event means that in a complete text, the semantic weight, position distribution, and importance score between all candidate events are analyzed, and the key event that best represents the core theme of the entire text is selected. The main event has the highest sentence vector aggregation value or closely matches the semantic intent implied by the article title. Aggregating paragraph-level associated events means that after determining the main event, other events that are consistent in entities, similar in event types, or connected in causal relationships with the main event are identified from each paragraph of the article and aggregated as subordinate events of the main event, such as event M, and associated events can be event N and event L. Forming a master-slave event structure means organizing the main event and multiple associated events according to a hierarchical structure, constructing a structured risk expression framework with the main event as the core and the subordinate events as supplements, thereby enhancing the logical readability and risk analysis capability between events.

[0034] According to the master-slave event structure, an event chain is formed and standardized output is performed.

[0035] Specifically, the structure containing the main event and multiple associated events is taken as the input basis, which clearly defines the core subordinate relationship between events and their trigger words, entities, and argument information. Assembling an event chain means that these events are connected in a certain logical order to form a sequence structure with time continuity, causal association, or entity dominance. The event chain construction methods include a time sequence chain, which arranges events according to the time sequence of event occurrence; an entity dominant chain, which connects all related events with a key entity as the core; and a causal logic chain, which emphasizes the cause-and-effect relationship between events, such as event O leading to event P, which in turn triggers event Q. Standardized output means converting the constructed event chain into a unified structured expression format, commonly presented in the form of triples, JSON objects, or edges-nodes in knowledge graphs, thereby facilitating subsequent storage, query, visualization, and reasoning tasks by computer systems. For example, a group of events can be encoded as "Enterprise A - Event H - Time B" and "Event H - Cause - Project Suspension", and the output is a double-layer graph structure of time chain + causal chain, supporting multi-language labeling and risk label mapping. Ultimately, the standardized event chain not only improves the semantic organization capability of the event extraction system.

[0036] Further, the application further comprises: obtaining historical technical and economic risk events, determining event type identifiers; listing trigger words according to the historical technical and economic risk events, obtaining a trigger word mode set; defining participating factors through the historical technical and economic risk events, obtaining argument role templates; performing semantic relationship association based on the event type identifiers, the trigger word mode set and the argument role template, obtaining semantic relationship rules, and constructing a technical and economic risk event knowledge system.

[0037] Specifically, obtaining historical technical and economic risk events and determining event type identifiers means extracting typical cases from past events related to technical and economic security risks that have occurred and been recorded, for summarizing and concluding the type labels of various risk events.

[0038] After the event type is determined, trigger words are listed according to the historical technical and economic risk events, and a trigger word mode set is obtained. Trigger words are core verbs, nouns or phrases in multi-source scientific texts that indicate the occurrence of events and serve as event trigger signals in the text. The trigger word mode set is an induction of the common combination forms of trigger words in context.

[0039] Then, participating factors are defined through historical technical and economic risk events, and argument role templates are obtained. Participating factors are various key entities or information units, such as the time, place, subjects involved, behavior patterns and event results of event occurrence. After classifying the participating factors, argument role templates can be established.

[0040] Finally, based on the event type identifiers, trigger word mode sets and argument role templates, semantic relationship association is performed to obtain semantic relationship rules and construct a technical and economic risk event knowledge system. Among them, semantic relationship association refers to identifying the logical relationship between trigger words and arguments in the event structure, such as "a certain behavior is performed by a certain subject" and "a certain cause leads to a certain result", forming a stable event expression structure. By continuously inducing semantic rules, a complete knowledge system with event type, trigger word, argument template and semantic logic is constructed, thereby providing basic support for subsequent risk event identification, classification and early warning.

[0041] Further, the application further comprises: dividing the multi-source scientific texts into sentence units according to semantic boundaries to obtain sentence-level segmentation; performing morphological analysis on the sentence-level segmentation to obtain entity clues; performing dependency tree structure analysis on the sentence-level segmentation to obtain sentence semantic relationships; correcting and removing duplicates of the entity clues and sentence semantic relationships, and combining to obtain multi-dimensional text information.

[0042] Specifically, the technology text data from different sources (such as technology news, scientific research reports, etc.) are preliminarily processed, and long texts are divided into multiple sentences with complete meanings according to the language interruption sentence rules and semantic completeness, obtaining sentence-level segmentation. The semantic boundary not only refers to the physical separation at the punctuation mark, but also includes the unit expressing a complete event or fact in semantics. For example, a three-hundred-word notice may be divided into ten independent sentence-level units, each of which constitutes a sentence-level segmentation unit, providing a basic unit for subsequent analysis.

[0043] The words in each sentence are processed by part-of-speech tagging, word segmentation, and stem reduction, etc. to identify the words that may constitute important entities, obtaining entity clues. Morphological analysis helps to determine nouns, verbs, adjectives, etc., and to identify continuous word terms such as "A company" and "key materials". In technical texts, there are often proper nouns composed of multiple words, so morphological analysis is crucial for preliminary entity extraction. The entity clues obtained thereby are the basis for subsequent entity recognition and matching.

[0044] The syntactic dependency structure between words within the sentence-level segmented sentence is constructed to represent the subject-verb, verb-object, and other relationships between words in a tree form, obtaining the semantic relationship of the sentence. Dependency structure can reveal the deep semantic relationship of the sentence, for example, the dependency analysis of "the enterprise closes the factory" will show that "close" is the verb core, "enterprise" is the subject, and "factory" is the object, thus clearly defining the action, the executor, and the affected object. Through structure analysis, it can provide grammatical support for subsequent trigger word positioning and event role labeling.

[0045] After completing the morphological and dependency structure analysis, the identified entity candidates and their corresponding syntactic relationships are cross-checked to eliminate duplicate entities or recognition errors, while unifying the entity expression format to improve the overall recognition quality. The error correction operation includes removing spelling errors, syntax analysis errors, etc., and the deduplication refers to unifying the same entity under different expressions, for example, "a certain company technology Co., Ltd." and "a certain company" are merged into one entity, and then the entity, syntactic relationship, context location, etc. Information is integrated to obtain a structured text representation containing multi-dimensional language features, which is used for subsequent entity matching and event extraction.

[0046] Further, the application also includes: performing entity literal matching on the multi-dimensional text information according to the preset monitoring entity library to obtain fuzzy matching entities; performing entity expansion recognition based on the fuzzy matching entities on the multi-dimensional text information to obtain expanded matching entities; identifying the expanded matching entities in the multi-dimensional text information, and extracting context windows before and after the entity identification as the center, outputting a candidate monitoring entity list.

[0047] Specifically, the entity names in the previously constructed monitoring entity library are compared with the current multi-dimensional text information to be processed, and the possible corresponding entities are identified based on the literal similarity of the words in the text. The preset monitoring entity library contains entries such as enterprises, institutions and projects in the technical and economic field that have risk monitoring value, and the literal matching is to compare the expressions in the text with the entity library using string similarity, edit distance, spelling variants and other technologies, so as to identify similar but not necessarily identical names.

[0048] Based on the fuzzy matching entities, combined with the syntactic structure, part-of-speech information, and context semantic features, further entities that are not directly matched in literal but have actual reference relationship are identified, and through the naming entity recognition model, context co-reference resolution algorithm and other technologies, supplementary objects such as pronouns, abbreviations and implied entities can be identified.

[0049] All the identified expanded matching entities are positioned in the original text, and their specific occurrence positions are recorded, and then the text content of a fixed length is respectively extended forward and backward from each entity as the center, which is used for subsequent context analysis and event identification. The context window contains twenty to fifty words, which is used to retain the semantic environment around the entity and ensure that complete contextual information can be used when judging the event. Finally, all entities with position information and context windows are unified and organized into a candidate monitoring entity list, which provides input for the next trigger word recognition and event type judgment.

[0050] Further, the application also includes: performing dependency syntax analysis on the trigger word set, and outputting a dependency structure tree; performing semantic role labeling according to the dependency structure tree, and obtaining semantic roles; classifying and normalizing the semantic roles based on the argument role template, forming the standardized event structure; constructing the logical, causal and temporal semantic relationships between the multiple arguments of the trigger word set according to the standardized event structure, and outputting the risk event candidate structure set.

[0051] Specifically, based on the identified multiple trigger words, the natural language processing technology is used to analyze the syntactic structure of the sentence where the trigger word is located, to determine the grammatical dependency relationship between the trigger words, and to express it in a tree structure. Dependency syntax analysis is a method used to represent the dependency relationship between words in a sentence, which can identify subject-predicate relationship, verb-object relationship, and determinative-complement relationship.

[0052] After the syntactic structure of a sentence is determined, the semantic roles of each word in the sentence in an event are further identified, such as who is the performer of the action, who is the receiver of the action, when it happens, where it happens, what is the cause, and the like. Semantic role labeling is a method of parsing a sentence from a semantic level, and common roles include agent, patient, time, place, cause, and manner. Taking “the enterprise suspends production due to shortage of raw materials” as an example, “the enterprise” is the agent, “shortage of raw materials” is the cause, and “suspension of production” is the event action.

[0053] The identified semantic roles are classified and standardized by using a pre-set event argument role template. The role template is a structural pattern summarized according to different event types, and is used to standardize the naming and scope of various semantic roles. After aligning the semantic roles with the template, various expressions can be standardized into a unified structure, such as “the institution withdraws investment in April” can be standardized as “subject = the institution; action = withdrawal of investment; time = April”.

[0054] After obtaining the standardized event structure, the semantic connections between the roles in the event are further analyzed to construct a logical chain. Logical relationships include parallel relationship, conditional relationship, reverse relationship, and the like. The cause-and-effect relationship reveals the motivation and result of the event, and the time relationship describes the sequence of the event.

[0055] Further, the application also includes: performing semantic segment extraction on the candidate monitoring entity list and obtaining display markers; introducing a natural language prompt guide, inputting the display markers to trigger word prediction, obtaining trigger words, and obtaining the start position and end position of the trigger words; obtaining the event corresponding to the trigger words, performing bidirectional mapping determination on the trigger word event based on the technical and economic risk event knowledge system, and outputting the trigger word set and the event type determination result.

[0056] Specifically, on the basis of the identified candidate monitoring entities, the context semantic content of the text position of each entity is extracted to form a semantic segment, and a marker with a prompting effect is generated for the segment. The semantic segment is centered on the monitoring entity and extends a certain number of words or sentences forward and backward to express the semantic environment in which the entity is located, and the display marker is a label for the entity in visual or model processing, such as a special symbol, color, or prompt word to represent the existence of the entity, which facilitates the subsequent model to identify the key information in the context and focus on the semantic area related to the entity.

[0057] By adding a natural language form of prompt in the pre-trained language model, the pre-trained language model is guided to identify the event-related verb or expression, i.e. trigger word. The prompt-guided way can be to add a question template such as "what happened to the enterprise?" "does the following text contain a risk event?" to enhance the perception of event triggers. When the pre-trained language model identifies the trigger word, it will mark its specific position in the original semantic segment, including the start character or word position and the end character or word position, to ensure that the trigger word can be used for positioning and further event structure construction.

[0058] The identified trigger word is compared with the pre-constructed event type identification to determine the specific event type to which it belongs, and the mapping is verified for logical consistency. Bidirectional mapping means that it not only maps from the trigger word to the existing event type in the technical and economic risk event knowledge system (such as "change of control" and the like), but also checks back from the event type to see if there is a typical use of the trigger word to ensure the rationality and accuracy of the matching. The final output result includes a complete set of trigger words and their associated risk event type labels related to the monitoring entity, providing semantic anchor points for subsequent argument extraction and event structure generation.

[0059] Further, the application also includes: archiving the set of risk event candidate structures according to the monitoring entity dimension and the paragraph position dimension, constructing a chapter-level event candidate index table; performing fusion scoring and sorting based on three types of strategies according to the chapter-level event candidate index table, selecting a risk theme main event; performing paragraph-level associated event aggregation based on three types of association strategies according to the risk theme main event, obtaining an associated event set; performing master-slave event structure output construction based on the risk theme main event and the associated event set, obtaining the master-slave event structure.

[0060] Specifically, the plurality of risk event candidate structures identified are sorted and classified according to the monitoring entity and the paragraph position in the text, and an index table is established. The monitoring entity dimension represents the merging of events related to the same entity such as a certain enterprise, technology, institution, etc., while the paragraph position dimension refers to the clear labeling of the distribution position of the event in the full text, such as appearing in the 2nd paragraph or the 5th paragraph. The two-dimensional archiving method helps to grasp the context and entity relevance of the event, providing an organizational basis for subsequent event focusing and screening.

[0061] Three different strategies are applied to the candidate events archived in the index table to comprehensively score and sort the most representative main event according to the total score. The three strategies include a title semantic guidance strategy, that is, whether the event is consistent with the title meaning; an event influence evaluation strategy, that is, to measure the frequency of the event in the full text, entity weight, reasoning depth and other indicators; and a position priority strategy, that is, whether the event is located near the beginning or end of the paragraph and other key positions. The event with the highest score after fusion scoring is determined as the main event of the article.

[0062] Around the main event, other events with semantic, logical or entity association are found in the whole text and aggregated into a set. The three association strategies include entity coincidence type association, that is, the main event and other events share the same monitoring entity; event causal chain association, that is, there is an obvious cause and effect or aftereffect relationship; and semantic similarity type association, that is, different events are similar or mutually supporting in semantic level, so that the relevant events scattered in different paragraphs can be organized, and the context behind the main event can be enriched and supplemented.

[0063] The main event is taken as the core, and other associated events are taken as auxiliary content, and a hierarchical structure is formed according to the main and subordinate levels. The main and subordinate structure not only reveals the whole picture of the main event, but also reflects its extension in a larger semantic context.

[0064] Further, the application also includes: the three strategies include a title semantic guidance strategy, an event influence evaluation strategy and a position priority strategy; and the three association strategies include an entity coincidence type association, an event causal chain association and a semantic similarity type association.

[0065] Specifically, the three strategies include a title semantic guidance strategy, an event influence evaluation strategy and a position priority strategy. The title semantic guidance strategy refers to using the central meaning contained in the chapter title to guide the determination of the main event. Specifically, the semantic similarity of the candidate event and the title is matched in terms of sentence vector or keyword, and if an event is highly related to the title, it can be determined that the event is more dominant.

[0066] The event influence evaluation strategy refers to evaluating the candidate event based on its performance strength and risk weight in the article. The factors considered by the event influence evaluation strategy include whether the entity involved in the event is a key monitoring object, whether the event is mentioned multiple times, whether the event has a potential chain propagation effect, etc. For example, an event about "core material supply disruption" is mentioned more than three times in the text and involves multiple industry links, so its influence is greater.

[0067] The position priority strategy refers to incorporating the event position information in the text structure into the sorting basis. It is generally believed that the events located in the first and last segments are more important because they often bear the functions of article guidance and summary. Taking a copy containing six paragraphs as an example, if an event is emphasized in both the first and last segments, its priority is significantly higher than that of an event mentioned only in a middle paragraph.

[0068] On the other hand, the three types of association strategies include entity coincidence type association, event causal chain association and semantic similarity type association. The entity coincidence type association refers to the same or highly relevant monitoring entities between the candidate event and the main event, for example, a certain enterprise, scientific research institution or technical field in the main event appears again in other events, which can be regarded as having entity association with the main event.

[0069] The event causal chain association refers to the existence of obvious causal relationship or logical sequence between two events, for example, the main event is “event W”, and a certain associated event is “high-end equipment project forced to terminate”, then the latter is the result of the former, and the two constitute a causal relationship, which can establish a chain connection.

[0070] The semantic similarity type association refers to the existence of certain degree of semantic similarity between events in terms of expression content, field attribution or potential meaning, for example, the main event is “industrial chain interruption”, and the event described in a certain paragraph is “unstable supply of key raw materials”, although they are not completely consistent, but there is a close relationship in the semantic of industrial risk, which can be regarded as a semantic similarity type of associated event.

[0071] Further, the application also includes: constructing the event chain according to the chain construction dimension of the main and subordinate event structure, wherein the chain construction dimension includes time sequence chain, entity dominant chain and causal logic chain; and performing standardized representation based on the event chain.

[0072] Specifically, based on the various internal relationships between the main event and its multiple subordinate events, the events are connected into an ordered event chain through preset dimension standards. The chain construction dimension refers to three logical ways of organizing the event chain, which are time sequence chain, entity dominant chain and causal logic chain. The time sequence chain is sorted according to the time points of the occurrence of events in the original text, for example, the event “event X” occurring first and the event “event Y” occurring later form a time sequence relationship. The entity dominant chain is connected according to the same or highly relevant monitoring entities involved in the events, for example, two events “event I” and “event J” involving the same scientific research institution can be connected into an event development chain dominated by the entity. The causal logic chain refers to the connection based on the causal reasoning between events, and there is an explicit causal logic between events, which can be connected into a complete causal chain.

[0073] After the event chain is constructed, the event chain needs to be standardized. The standardized representation refers to encoding the event chain in the original text in a unified format, structure and semantic label, so that it is convenient for storage, retrieval and reasoning analysis. The standardized form can be expressed in the form of event triplets, graph structure or hierarchical JSON format, etc., wherein each event node contains trigger words, time, argument information and its relationship type with the previous and next nodes. For example, in a certain standardized chain, event A is connected to event B through a causal relationship, and event B is connected to event C through a time sequence relationship, forming an event chain expression with clear structure and available for algorithm processing.

[0074] In summary, the technical and economic risk event extraction method based on monitoring entities provided in the present application has the following technical effects: by achieving the technical goal of centering on monitoring entities, constructing a structured risk event set and establishing an intelligent extraction process of multi-dimensional event chain, the technical effects of improving the accuracy of technical and economic security event identification, enhancing the depth of event understanding and supporting the risk decision assistance capability in multiple scenarios are achieved.

[0075] Embodiment two, based on the same inventive concept as the technical and economic risk event extraction method based on monitoring entities in the foregoing embodiments, the present application also provides a technical and economic risk event extraction system based on monitoring entities, please refer to the attached Figure 2 , including: an entity recognition module 11 for performing entity recognition on multi-source technical text, extracting technical and economic monitoring entities, and performing entity comparison and screening on the technical and economic monitoring entities based on a preset monitoring entity library, and outputting a candidate monitoring entity list; a semantic fragment extraction module 12 for performing semantic fragment extraction with the candidate monitoring entity list as an anchor point, identifying event trigger words, and determining event types based on trigger word matching rules; a semantic role labeling module 13 for performing dependency syntax analysis and semantic role labeling on the identified trigger word set, constructing a standardized event structure, and extracting semantic relationships, and outputting a risk event candidate structure set; a master-slave event structure forming module 14 for identifying chapter-level main events, aggregating paragraph-level related events, and forming a master-slave event structure based on the risk event candidate structure set; a standardized output module 15 for assembling event chains according to the master-slave event structure and performing standardized output.

[0076] Further, the technical and economic risk event extraction system based on monitoring entities is also used for: obtaining historical technical and economic risk events, determining event type identifiers; listing trigger words according to the historical technical and economic risk events, obtaining a trigger word mode set; defining participating elements through the historical technical and economic risk events, obtaining argument role templates; performing semantic relationship association based on the event type identifiers, the trigger word mode set and the argument role templates, obtaining semantic relationship rules, and constructing a technical and economic risk event knowledge system.

[0077] Further, the technical and economic risk event extraction system based on monitoring entities is also used for: dividing the multi-source technical text into sentence units according to semantic boundaries to obtain sentence-level segmentation; performing morphological analysis on the sentence-level segmentation to obtain entity clues; performing dependency tree structure analysis on the sentence-level segmentation to obtain sentence semantic relationships; correcting and deduplicating the entity clues and the sentence semantic relationships, and combining to obtain multi-dimensional text information.

[0078] Further, the technical and economic risk event extraction system based on monitoring entities is also used for: performing entity literal matching on the multi-dimensional text information according to the preset monitoring entity library to obtain fuzzy matching entities; performing entity expansion recognition on the multi-dimensional text information based on the fuzzy matching entities to obtain expanded matching entities; identifying the expanded matching entities in the multi-dimensional text information, and extracting context windows before and after the text respectively with the entity identifier as the center, and outputting a candidate monitoring entity list.

[0079] Further, the technical and economic risk event extraction system based on monitoring entities is also used for: performing dependency syntax analysis on the trigger word set to output a dependency structure tree; performing semantic role labeling according to the dependency structure tree to obtain semantic roles; performing role classification and normalization on the semantic roles based on the argument role templates to form the standardized event structure; constructing the semantic relationships of logic, causality and time between the multiple arguments of the trigger word set according to the standardized event structure, and outputting the risk event candidate structure set.

[0080] Further, the technical and economic risk event extraction system based on monitoring entities is also used for: performing semantic fragment extraction on the candidate monitoring entity list and obtaining display markers; introducing natural language prompts to guide, inputting the display markers to predict trigger words, obtaining trigger words, and obtaining the start position and end position of the trigger words; obtaining events corresponding to the trigger words, performing bidirectional mapping determination on the trigger word events based on the technical and economic risk event knowledge system, and outputting trigger word sets and event type determination results.

[0081] Further, the technical and economic risk event extraction system based on the monitoring entity is further used for: archiving the risk event candidate structure set according to a monitoring entity dimension and a paragraph position dimension, and constructing a passage-level event candidate index table; performing fusion scoring and sorting based on three types of strategies according to the passage-level event candidate index table, and selecting a risk theme main event; performing paragraph-level associated event aggregation based on three types of association strategies according to the risk theme main event, and obtaining an associated event set; and performing main-slave event structure output construction based on the risk theme main event and the associated event set, and obtaining the main-slave event structure.

[0082] Further, the technical and economic risk event extraction system based on the monitoring entity is further used for: the three types of strategies include a title semantic guidance strategy, an event influence evaluation strategy, and a position priority strategy; and the three types of association strategies include an entity coincidence type association, an event causal chain association, and a semantic similarity type association.

[0083] Further, the technical and economic risk event extraction system based on the monitoring entity is further used for: constructing the event chain according to a chain construction dimension, wherein the chain construction dimension includes a time sequence chain, an entity dominant chain, and a causal logic chain; and performing standardized representation based on the event chain.

[0084] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The technical and economic risk event extraction method and specific example in the foregoing embodiment one are also applicable to the technical and economic risk event extraction system based on the monitoring entity in the present embodiment. Those skilled in the art can clearly know the technical and economic risk event extraction system based on the monitoring entity in the present embodiment from the foregoing detailed description of the technical and economic risk event extraction method based on the monitoring entity. Therefore, for the sake of brevity of the specification, no longer detailed description is given here.

[0085] The above description of disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0086] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalents, the present application also intends to include these modifications and variations.

Claims

1. A technical and economic risk event extraction method based on monitoring entities, characterized in that, The method comprises the following steps: performing entity recognition on the multi-source scientific and technological texts, extracting technical and economic monitoring entities, performing entity comparison and screening on the technical and economic monitoring entities based on a preset monitoring entity library, and outputting a candidate monitoring entity list; performing semantic segment extraction based on the candidate monitoring entity list as an anchor point, performing event trigger word recognition, and determining the event type based on trigger word matching rules; performing dependency syntax analysis and semantic role labeling on the identified trigger word set, constructing a standardized event structure, and extracting semantic relationships to output a risk event candidate structure set; based on the risk event candidate structure set, identifying a main event at the chapter level, aggregating related events at the paragraph level, and forming a master-slave event structure; assembling an event chain according to the master-slave event structure and performing standardized output; wherein the semantic segment extraction based on the candidate monitoring entity list as an anchor point, the event trigger word recognition, and the event type determination based on trigger word matching rules comprise: performing semantic segment extraction on the candidate monitoring entity list and obtaining display markers; introducing natural language prompts to guide the input of the display markers to predict trigger words, obtain the start and end positions of the trigger words, and obtain the trigger word set and event type determination results; the bidirectional mapping refers to mapping from the trigger word to the existing event type in the technical and economic risk event knowledge system, and checking whether there is a typical use of the trigger word from the event type. The method further comprises the following steps before the entity recognition on the multi-source scientific and technological texts, the extraction of technical and economic monitoring entities, and the entity comparison and screening on the technical and economic monitoring entities based on the preset monitoring entity library to output the candidate monitoring entity list:

2. The method of claim 1, wherein the method is based on a monitoring entity. determining event type identifiers based on historical technical and economic risk events; listing trigger words based on the historical technical and economic risk events to obtain a trigger word pattern set; defining participating elements based on the historical technical and economic risk events to obtain argument role templates; performing semantic relationship association based on the event type identifiers, the trigger word pattern set, and the argument role templates to obtain semantic relationship rules and construct a technical and economic risk event knowledge system. The method further comprises the following steps before the entity recognition on the multi-source scientific and technological texts, the extraction of technical and economic monitoring entities, and the entity comparison and screening on the technical and economic monitoring entities based on the preset monitoring entity library to output the candidate monitoring entity list:

3. The method of claim 2, wherein the method is based on a monitoring entity. dividing the multi-source scientific and technological texts into sentence units according to semantic boundaries to obtain sentence-level segmentation; performing morphological analysis on the sentence-level segmentation to obtain entity clues; performing dependency tree structure analysis on the sentence-level segmentation to obtain sentence semantic relationships; performing error correction and deduplication on the entity clues and sentence semantic relationships, and combining to obtain multi-dimensional text information. The method comprises the following steps:

4. The method of claim 3, wherein the method is based on a monitoring entity. ​ According to the preset monitoring entity library, the multi-dimensional text information is subjected to entity literal matching, to obtain fuzzy matching entities; The multi-dimensional text information is subjected to entity expansion recognition based on the fuzzy matching entities, to obtain expanded matching entities; The expanded matching entities are marked in the multi-dimensional text information, and a context window is extracted before and after the entity mark respectively, to output a candidate monitoring entity list.

5. The method of claim 2, wherein the method is based on a monitoring entity. The identified trigger word set is subjected to dependency syntax analysis and semantic role labeling, to construct a standardized event structure and extract semantic relationships, to output a risk event candidate structure set, including: The trigger word set is subjected to dependency syntax analysis, to output a dependency structure tree; According to the dependency structure tree, semantic roles are labeled, to obtain semantic roles; Based on the argument role template, the semantic roles are classified and normalized, to form the standardized event structure; According to the standardized event structure, the semantic relationships among the multiple arguments of the trigger word set are constructed, to output the risk event candidate structure set.

6. The method of claim 1, wherein the method is based on a monitoring entity. Based on the risk event candidate structure set, a chapter-level main event is identified, and paragraph-level associated events are aggregated, to form a master-slave event structure, including: The risk event candidate structure set is archived according to the monitoring entity dimension and the paragraph position dimension, to construct a chapter-level event candidate index table; According to the chapter-level event candidate index table, a fusion scoring and sorting based on three types of strategies is performed, to select a risk theme main event; According to the risk theme main event, a paragraph-level associated event aggregation based on three types of association strategies is performed, to obtain an associated event set; Based on the risk theme main event and the associated event set, a master-slave event structure output construction is performed, to obtain the master-slave event structure; The three types of strategies include a title semantic guidance strategy, an event influence evaluation strategy, and a position priority strategy; the three types of association strategies include an entity overlap type association, an event causal chain association, and a semantic similarity type association.

7. The method of claim 1, wherein the method is based on a monitoring entity. According to the master-slave event structure, an event chain is constructed and standardized output is performed, including: According to the chain construction dimension, the master-slave event structure is constructed into the event chain, wherein the chain construction dimension includes a time sequence chain, an entity dominant chain, and a causal logic chain; Based on the event chain, a standardized representation is performed.

8. A system for extracting technical and economic risk events based on monitoring entities, characterized in that, Steps for implementing the technical and economic risk event extraction method based on monitoring entities in any one of claims 1 to 7, including: An entity recognition module is configured to recognize entities in multi-source technical texts, extract technical and economic monitoring entities, and perform entity comparison and screening on the technical and economic monitoring entities based on a preset monitoring entity library, to output a candidate monitoring entity list; A semantic segment extraction module is configured to extract semantic segments based on the candidate monitoring entity list as an anchor point, recognize event trigger words, and determine event types based on trigger word matching rules; A semantic role labeling module is configured to perform dependency syntax analysis and semantic role labeling on the identified trigger word set, construct a standardized event structure, and extract semantic relationships, to output a risk event candidate structure set; The master-slave event structure forming module is configured to identify a chapter-level master event, aggregate paragraph-level associated events, and form a master-slave event structure based on the risk event candidate structure set. The standardization output module is configured to establish an event chain according to the master-slave event structure and perform standardization output.

Images