Industrial dynamic tracking semantic analysis method and system based on artificial intelligence

Abstract

The invention provides an industry dynamic tracking semantic analysis method and system based on artificial intelligence, and the method comprises the steps: firstly obtaining an industry dynamic text set which comprises a plurality of industry dynamic text units with source identifiers, constructing source semantic hierarchical mapping based on the source identifiers, and obtaining a hierarchical association relationship between the text units; calling the pre-training model to generate a semantic association rule set, constructing a dynamic semantic tracking link according to the semantic association rule set, extracting a core semantic development direction to determine a core tracking dimension, and integrating semantic information according to the dimension to generate an industry dynamic analysis report, so that the industry dynamic semantics can be efficiently and accurately tracked; and a comprehensive and accurate industry analysis result is provided for a user.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to an industry dynamic tracking semantic analysis method and system based on artificial intelligence. Background Technology

[0002] In today's information-saturated era, industries are developing rapidly. Timely and accurate tracking and analysis of industry dynamics are crucial for companies to formulate strategic decisions and for investors to grasp market trends. However, existing methods for analyzing industry dynamics have many shortcomings.

[0003] On the one hand, some traditional methods rely heavily on manual collection and organization of industry dynamic information, which is not only inefficient but also prone to missing important information. Due to the limitations of human processing capabilities, it is difficult to conduct a comprehensive and detailed analysis of massive amounts of industry dynamic text, resulting in incomplete and inaccurate analysis results.

[0004] On the other hand, some analytical methods based on simple keyword matching or rule setting lack a deep understanding of the semantics of industry dynamic texts. They can only identify surface-level lexical information in the text and cannot capture the inherent semantic connections and dynamic changes between texts. For example, these methods often fail to accurately interpret texts with implicit semantics or industry-specific expressions, thus failing to uncover the core development trends and key information of industry dynamics and making it difficult to meet users' needs for precise analysis of industry dynamics. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide an industry dynamic tracking semantic analysis method based on artificial intelligence, the method comprising: Obtain a set of industry dynamic texts, which contains multiple industry dynamic text units with source identifiers; Based on the source identifier, a source semantic hierarchy mapping of industry dynamic text units is constructed to obtain the hierarchical association relationship between industry dynamic text units; The pre-trained dynamic semantic tracking model is invoked to perform dynamic association rule generation processing on the hierarchical relationship, generating a set of dynamic semantic association rules for the industry. Based on the set of semantic association rules, the association links of industry dynamic text units are constructed to obtain the dynamic semantic tracking links of industry dynamics; Based on the dynamic semantic tracking link, the core semantic development direction of industry dynamics is extracted, and the core tracking dimension of industry dynamics is determined. The semantic information in the dynamic semantic tracking link is integrated and processed according to the core tracking dimensions to generate an industry dynamic analysis report.

[0006] In another aspect, embodiments of the present invention also provide an industry dynamic tracking semantic analysis system based on artificial intelligence, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0007] Based on the above, this embodiment of the invention comprehensively collects diverse information about industry dynamics by acquiring a set of industry dynamic texts with source identifiers. It constructs a source semantic hierarchy mapping based on the source identifiers, which can present the hierarchical relationships between industry dynamic text units. It calls a pre-trained dynamic semantic tracking model to generate a set of semantic association rules, leveraging the powerful learning capabilities of artificial intelligence to deeply explore the dynamic semantic relationships between industry dynamic texts. Based on the semantic association rules, it constructs a dynamic semantic tracking link, realizing the dynamic tracking and connection of industry dynamic semantics. This enables the display of the development process and evolution path of industry dynamics, extracts the core semantic development direction and determines the core tracking dimensions, accurately locates the key information and development trends of industry dynamics, and finally integrates semantic information according to the core tracking dimensions to generate an industry dynamic analysis report. This helps users quickly understand industry dynamics and make scientific and reasonable decisions. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the execution flow of the industry dynamic tracking semantic analysis method based on artificial intelligence provided in the embodiments of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of the AI-based industry dynamic tracking semantic analysis system provided in this embodiment of the invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an AI-based industry dynamic tracking semantic analysis method according to an embodiment of the present invention. The following is a detailed description of this AI-based industry dynamic tracking semantic analysis method.

[0011] Step S110: Obtain an industry dynamic text set, which contains multiple industry dynamic text units with source identifiers.

[0012] In this embodiment, the text acquisition module in the computer system is first activated. This module is pre-configured with mainstream information source addresses for the target industry, including industry-specific media platforms, official corporate release channels, government and industry association platforms, financial information platforms, and other information source channels. The text acquisition module automatically accesses these channels at a preset time frequency, capturing text content related to the target industry published within a set time range. During the capture process, a keyword filtering mechanism (preset keywords include the target industry's core products, technologies, policies, company names, etc.) is used to filter valid text, excluding invalid content such as advertising text and entertainment news unrelated to the target industry. While capturing each piece of text, the source identifier is simultaneously extracted. The source identifier consists of three parts: channel type, publisher name, and publication time. The channel type is used to distinguish different types of information sources, and the publication time uses a standard time format. After continuous collection and filtering within a set time frame, a collection of industry dynamic text units is obtained. Each industry dynamic text unit has a unique source identifier. For example, the source identifier of a certain industry dynamic text unit includes its corresponding publishing channel type, publishing entity, and publishing time, and the corresponding text content is the specific dynamic information related to the target industry published by the publishing entity at that time. Similarly, the source identifier of another industry dynamic text unit also includes its corresponding publishing channel type, publishing entity, and publishing time, and the text content is the relevant industry dynamic information published by another publishing entity.

[0013] Step S120: Construct a source semantic hierarchy mapping of industry dynamic text units based on the source identifier to obtain the hierarchical association relationship between industry dynamic text units.

[0014] After acquiring the set of industry dynamic texts, the semantic hierarchy mapping construction module in the computer system will construct the source semantic hierarchy mapping based on the source identifier of each industry dynamic text unit, thereby obtaining the hierarchical relationship between industry dynamic text units.

[0015] Step S121: Parse the source identifier of each industry dynamic text unit and extract the source entity information, source field information and source release channel information from the source identifier.

[0016] In this embodiment, the semantic hierarchy mapping construction module first parses the source identifier of each industry dynamic text unit. For the source identifier, according to preset parsing rules, its various components are separated, and then the source entity information, source domain information, and source publishing channel information are extracted. The source entity information refers to the specific entity that publishes the industry dynamic text unit, such as various enterprises, institutions, and media; the source domain information refers to the specific domain to which the source entity belongs, for example, in the case of multiple industries, clarifying the main domain scope of the entity; and the source publishing channel information refers to the specific channel through which the text unit is published, such as an official website, social media platform, or industry journal. For example, for a source identifier, after parsing, the extracted source entity information is a technology company, the source domain information is the software development domain to which the technology company belongs, and the source publishing channel information is the company's official blog. Through the above parsing process, these three types of information are extracted for each industry dynamic text unit, serving as the basic data for subsequently constructing the source semantic hierarchy mapping.

[0017] Step S122: Construct a source subject hierarchy system based on the source subject information. The source subject hierarchy system contains multiple levels of subject nodes. Each subject node corresponds to a unique subject code, and different subject nodes are connected through subject association strength.

[0018] After extracting the source subject information of all industry dynamic text units, the subject hierarchy construction module in the computer system will construct a source subject hierarchy system based on this source subject information.

[0019] Step S1221: Deduplicate the source subject information of all industry dynamic text units to obtain a unique source subject set.

[0020] The subject-level construction module first performs a deduplication operation on all collected source subject information. By comparing the source subject information of dynamic text units from different industries, identical source subject information is merged, retaining only one instance, thus obtaining a unique source subject set. For example, if multiple dynamic text units have source subject information for the same company, then only that company's subject information is retained in the unique source subject set.

[0021] Step S1222: Extract the subject attribute information of each unique source subject, the subject attribute information including subject type, subject size and subject industry influence.

[0022] For each unique source entity in the set of unique source entities, the entity hierarchy construction module further extracts its entity attribute information. Entity type is categorized based on the nature of the source entity, such as enterprise, research institution, industry association, media, etc.; entity size can be determined based on comprehensive factors such as the number of personnel, business scope, and asset size, such as large, medium, small, etc.; entity industry influence is assessed through factors such as the entity's reputation within the industry, the authority of its published information, and the degree of its participation in industry activities. For example, for a unique source entity that is a large technology company, its entity type is enterprise, its entity size is large, and its industry influence is high.

[0023] Step S1223: Based on the subject type, perform first-level classification on the unique source subjects to obtain multiple first-level subject categories. Each first-level subject category contains multiple unique source subjects with the same subject type.

[0024] The subject hierarchy construction module categorizes unique source subjects at the primary level according to subject type. Unique source subjects with the same subject type are grouped into the same category, forming multiple primary subject categories. For example, all unique source subjects of the subject type "enterprise" are grouped into the "enterprise" primary subject category, all unique source subjects of the subject type "research institution" are grouped into the "research institution" primary subject category, and so on, resulting in multiple different primary subject categories.

[0025] Step S1224: Based on the subject size, perform secondary classification on the unique source subjects in each primary subject category to obtain multiple secondary subject subcategories. Each secondary subject subcategory contains multiple unique source subjects with the same subject size.

[0026] After obtaining the primary subject categories, the subject hierarchy construction module performs secondary classification on the unique source subjects within each primary subject category based on subject size. Unique source subjects with the same subject size within the same primary subject category are grouped into a subcategory, forming multiple secondary subject subcategories. For example, in the primary subject category of enterprises, large enterprises are classified as the large enterprise secondary subject subcategory, medium-sized enterprises as the medium-sized enterprise secondary subject subcategory, and small enterprises as the small enterprise secondary subject subcategory.

[0027] Step S1225: Based on the industry influence of the main body, perform a three-level ranking of the unique source entities in each secondary subject subcategory to determine the hierarchical position of each unique source entity in the secondary subject subcategory.

[0028] For each second-level entity subcategory, the entity hierarchy construction module sorts entities in three levels based on their industry influence. Entities are arranged in descending order of industry influence to determine their hierarchical position within that second-level entity subcategory. For example, in the second-level entity subcategory of large enterprises, enterprises are sorted according to their industry influence, with the most influential enterprise occupying the highest level position, the next most influential enterprise in the next level, and so on.

[0029] Step S1226: Assign a corresponding subject code to each unique source subject. The subject code includes a first-level subject category identifier, a second-level subject sub-category identifier, and a hierarchical position identifier.

[0030] The entity hierarchy construction module assigns a unique entity code to each unique source entity. This entity code consists of a first-level entity category identifier, a second-level entity subcategory identifier, and a hierarchy position identifier. Each identifier has its specific meaning and coding rules. Through this coding method, the position of the unique source entity in the source entity hierarchy can be clearly reflected. For example, for a unique source entity that belongs to the first-level entity category of "Enterprise," the second-level entity subcategory of "Large Enterprise," and is located at the third level, its entity code can be composed of a first-level identifier representing the "Enterprise" category, a second-level identifier representing the "Large Enterprise" category, and an identifier representing the third-level position.

[0031] Step S1227: Calculate the subject association strength between subjects with different unique sources, wherein the subject association strength is determined based on subject cooperation history information and subject information interaction frequency.

[0032] The subject-level construction module calculates the subject association strength between subjects with different unique sources. First, it collects historical information on subject cooperation between these subjects, including the number of cooperative projects, the depth and breadth of cooperation, and the time span of cooperation. Simultaneously, it collects the frequency of subject information interaction, such as the number of times information is mutually cited, the frequency of mutual forwarding of updates, and the number of times each subject is mentioned in public. Then, according to preset calculation rules, the subject cooperation history information and subject information interaction frequency are transformed into quantifiable indicators, and the subject association strength is obtained through weighted calculation. The weights of subject cooperation history information and subject information interaction frequency are determined based on their influence on the subject association strength. For example, if two companies have multiple in-depth cooperative projects and frequent mutual information interaction, their subject association strength is high; conversely, their subject association strength is low.

[0033] Step S1228: Organize the unique source subjects with subject association strength according to the hierarchical classification results to construct a source subject hierarchical system. Each subject node in the source subject hierarchical system contains subject code, subject attribute information and subject association strength of associated subject nodes.

[0034] The subject hierarchy construction module organizes unique source subjects with subject association strength according to the previous hierarchical classification results, constructing a source subject hierarchy system. In this system, each unique source subject is a subject node, and each subject node contains its corresponding subject code, subject attribute information, and subject association strength with other related subject nodes. Through this organization method, a well-defined source subject hierarchy system with strong associations between nodes is formed, clearly demonstrating the hierarchical relationships and degree of association between different source subjects.

[0035] Step S123: Classify the industry dynamic text units according to the source domain information to obtain domain text groups divided by domain. Each domain text group corresponds to at least one subject node in the source subject hierarchy system.

[0036] The semantic hierarchy mapping construction module classifies industry dynamic text units based on the extracted source domain information. Industry dynamic text units with the same or similar source domain information are grouped together, forming domain-specific text groups. Each domain text group corresponds to at least one subject node in the source subject hierarchy, because the source subject of these industry dynamic text units belongs to the source subject represented by that subject node. For example, industry dynamic text units whose source domain information is all in the field of artificial intelligence algorithms are grouped into one domain text group, and this domain text group corresponds to all subject nodes in the source subject hierarchy that belong to the field of artificial intelligence algorithms.

[0037] Step S124: Extract the domain semantic features of the industry dynamic text units in each domain text group. The domain semantic features include domain core vocabulary and domain expression style features.

[0038] For each domain text group, the semantic hierarchy mapping construction module extracts the domain semantic features of the industry dynamic text units within it. First, using text analysis techniques, it extracts frequently occurring words that represent the core content of the domain from the industry dynamic text units within the domain text group, designating them as domain core vocabulary. Simultaneously, it analyzes the expression methods of the industry dynamic text units within the domain text group, including sentence structure (e.g., the prevalence of declarative, interrogative, or exclamatory sentences) and vocabulary usage preferences (e.g., the frequency of use of formal vocabulary and professional terminology), thus forming domain expression style characteristics. For example, in a financial text group, domain core vocabulary might include "stocks," "bonds," and "interest rates," and the domain expression style characteristics might manifest as the prevalence of declarative sentences and professional financial terminology.

[0039] Step S125: Calculate the semantic similarity between different domain text groups based on the domain semantic features, and generate a domain semantic similarity matrix. The elements in the domain semantic similarity matrix represent the degree of semantic similarity between two domain text groups.

[0040] The semantic hierarchy mapping construction module calculates the semantic similarity between different domain text groups based on the extracted domain semantic features and generates a domain semantic similarity matrix.

[0041] For example, step S1251: perform word vector transformation on the domain core words in the domain semantic features of each domain text group to obtain the word vector representation corresponding to each domain core word.

[0042] First, a pre-trained word vector model is invoked to transform the core vocabulary of each domain text group into word vectors. The word vector model maps each core vocabulary to a multi-dimensional vector space, obtaining a word vector representation corresponding to each core vocabulary. This word vector reflects the semantic information of the vocabulary. For example, for the core vocabulary of "artificial intelligence" and "machine learning," the word vectors obtained after word vector transformation have a certain similarity in the vector space.

[0043] Step S1252: Calculate the average value of the core vocabulary vectors of all domain text groups in each domain text group to obtain the core vocabulary vector of that domain text group.

[0044] For each domain-specific text group, the word vectors corresponding to all the core domain terms contained within it are averaged to obtain the core vocabulary vector for that text group. This core vocabulary vector integrates the semantic information of all the core terms in the text group and can represent the semantic features of the text group in terms of core terms to a certain extent. For example, if a text group has multiple core domain terms, the word vectors of these terms are added together and then divided by the number of terms to obtain the core vocabulary vector for that text group.

[0045] Step S1253: Extract the sentence structure features and vocabulary usage frequency features from the domain expression style features of each domain text group, and convert the sentence structure features and vocabulary usage frequency features into style feature vectors.

[0046] Extract sentence structure features and vocabulary frequency features from the domain expression style features of each domain text group. For sentence structure features, count the proportion of different sentence types (such as declarative sentences, passive sentences, etc.) in the text group; for vocabulary frequency features, count the frequency of specific types of words (such as adjectives, adverbs, etc.). Then, arrange these statistically obtained proportions and frequency values ​​in a preset order to form a style feature vector. For example, in the sentence structure features of a certain domain text group, declarative sentences account for 0.7%, passive sentences account for 0.2%, and other sentence types account for 0.1%; in the vocabulary frequency features, adjectives account for 0.3% and adverbs account for 0.15%. Arrange these values ​​in order to form a style feature vector.

[0047] Step S1254: Perform weighted fusion on the core vocabulary vector and style feature vector to obtain the domain semantic feature vector of each domain text group. The weight of the weighted fusion is determined based on the contribution of the domain semantic features to the semantic similarity.

[0048] For each domain text group, the core vocabulary vector and style feature vector are weighted and fused. The weights of the core vocabulary and style features are determined based on their importance in determining semantic similarity, i.e., their contribution to semantic similarity. Then, the core vocabulary vector is multiplied by its corresponding weight, and the style feature vector is multiplied by its corresponding weight. The two results are then concatenated to obtain the domain semantic feature vector for that text group. For example, if the weight of the core vocabulary vector is 0.6 and the weight of the style feature vector is 0.4, then the core vocabulary vector is multiplied by 0.6, and the style feature vector is multiplied by 0.4. These two weighted vectors are then concatenated to form the domain semantic feature vector.

[0049] Step S1255: Calculate the similarity value between the domain semantic feature vectors of any two domain text groups using the cosine similarity algorithm, arrange the similarity values ​​between all two domain text groups in matrix form to generate a domain semantic similarity matrix. The rows and columns of the domain semantic similarity matrix correspond to different domain text groups, and the matrix elements are the similarity values ​​between the corresponding two domain text groups.

[0050] The cosine similarity algorithm is used to calculate the similarity value between the domain semantic feature vectors of any two domain text pairs. The cosine similarity algorithm measures the similarity between two vectors by calculating the cosine of the angle between them; the closer the cosine value is to 1, the more similar the two vectors are. For each pair of domain text pairs, the cosine similarity value between their domain semantic feature vectors is calculated. Then, all these similarity values ​​are arranged in a matrix, where the rows and columns of the matrix correspond to different domain text pairs, and each element in the matrix is ​​the similarity value between the two domain text pairs represented by its corresponding row and column, thus generating a domain semantic similarity matrix.

[0051] Step S126: Construct a source semantic hierarchy mapping based on the source subject hierarchy system and the domain semantic similarity matrix. The source semantic hierarchy mapping includes a bidirectional association between the subject hierarchy dimension and the domain semantic dimension.

[0052] The semantic hierarchy mapping construction module constructs a source semantic hierarchy mapping based on the pre-built source subject hierarchy system and domain semantic similarity matrix. In this mapping, the subject hierarchy dimension represented by the source subject hierarchy system is associated with the domain semantic dimension reflected in the domain semantic similarity matrix, forming a bidirectional association. That is, not only can information about the associated domain semantic dimensions be found from the subject hierarchy dimension, such as the domain text group corresponding to a certain subject node and its semantic similarity with other domain text groups, but also information about the associated subject hierarchy dimension can be found from the domain semantic dimension, such as the subject node corresponding to a certain domain text group and its position and association strength in the hierarchy system. Through this bidirectional association, a complete source semantic hierarchy mapping is constructed.

[0053] Step S127: Extract the hierarchical relationship between industry dynamic text units from the source semantic hierarchy mapping, wherein the hierarchical relationship includes the relationship between text units at the same level and the relationship between text units across levels.

[0054] The semantic hierarchy mapping construction module extracts the hierarchical relationships between industry dynamic text units from the source semantic hierarchy mapping. Same-level text unit relationships refer to the relationships between industry dynamic text units corresponding to subject nodes at the same level, such as the relationships between industry dynamic text units corresponding to different subject nodes within the same second-level subject subcategory. Cross-level text unit relationships refer to the relationships between industry dynamic text units corresponding to subject nodes at different levels, such as the relationships between industry dynamic text units corresponding to subject nodes in different first-level subject categories, or the relationships between industry dynamic text units corresponding to subject nodes in different second-level subject subcategories within the same first-level subject category. Through the above extraction, the hierarchical relationships between industry dynamic text units are obtained.

[0055] Step S130: Call the pre-trained dynamic semantic tracking model to perform dynamic association rule generation processing on the hierarchical relationship, and generate a set of dynamic semantic association rules for the industry.

[0056] After obtaining the hierarchical relationships between industry dynamic text units, the rule generation module in the computer system will call the pre-trained dynamic semantic tracking model to perform dynamic association rule generation processing on the hierarchical relationships, thereby generating a set of industry dynamic semantic association rules.

[0057] Step S131: Input the hierarchical relationship into the rule initialization module of the dynamic semantic tracking model, and extract the association attribute information in the hierarchical relationship. The association attribute information includes the association type, association frequency and association semantic direction.

[0058] The rule generation module first inputs the hierarchical relationship into the rule initialization module of the dynamic semantic tracking model. The rule initialization module analyzes the hierarchical relationship and extracts the association attribute information. Association type refers to the nature of the association between industry dynamic text units, such as causal association, parallel association, and subordinate association; association frequency refers to the number of times industry dynamic text unit pairs with the same association type appear; and association semantic direction refers to the semantic direction presented by the association, such as the semantic flow from one industry dynamic text unit to another. For example, in a hierarchical relationship, if two industry dynamic text units are causally related, and this association appears multiple times in the hierarchical relationship with a semantic direction from the cause text unit to the result text unit, this information will be extracted as association attribute information.

[0059] Step S132: Construct an initial association rule pool based on the association attribute information. Each initial association rule in the initial association rule pool includes an association trigger condition and an association result description. The association trigger condition is set based on the association type and association frequency.

[0060] The rule initialization module constructs an initial association rule pool based on the extracted association attribute information. For each association type, a corresponding association trigger condition is set based on its association frequency; that is, the rule is triggered when a certain association type and frequency are met. Simultaneously, the association result is described according to the semantic direction and specific content of the association, forming an association result description. Each initial association rule consists of the above association trigger condition and association result description. All generated initial association rules are aggregated to form the initial association rule pool. For example, for causal association types, when the association frequency reaches a certain number, it is set as an association trigger condition, and the association result description is that when this trigger condition is met, there is a causal semantic association between two industry dynamic text units.

[0061] Step S133: The rule optimization module of the dynamic semantic tracking model performs a fit evaluation on each initial association rule in the initial association rule pool, and selects the initial association rules whose fit evaluation results exceed the preset fit threshold to form a candidate association rule set.

[0062] The rule optimization module of the dynamic semantic tracking model will evaluate the fit of each initial association rule in the initial association rule pool.

[0063] Step S1331: Obtain a historical industry dynamic data set, which includes multiple historical industry dynamic text units and their corresponding historical hierarchical relationships.

[0064] The rule optimization module first acquires a historical industry dynamic data set, which contains multiple historical industry dynamic text units from a past period, as well as the corresponding historical hierarchical relationships between these text units. This historical data will serve as the basis for evaluating the suitability of the initial association rules.

[0065] Step S1332: Apply each initial association rule in the initial association rule pool to the historical industry dynamic data set, and count the number of times each initial association rule is triggered in the historical industry dynamic data set.

[0066] For each initial association rule in the initial association rule pool, the rule optimization module applies it to the historical industry dynamic data set. It checks whether the historical hierarchical relationships in the historical industry dynamic data set meet the association triggering conditions of the initial association rule. If they do, the rule is triggered, and the number of times each initial association rule is triggered in the historical industry dynamic data set is counted.

[0067] Step S1333: For each initial association rule, extract the actual association result corresponding to the historical industry dynamic data that triggered the initial association rule, and compare it with the association result description of the initial association rule.

[0068] For each initial association rule, the rule optimization module extracts the actual association results corresponding to those historical industry dynamic data that triggered the rule in the historical industry dynamic data set, that is, the actual association between these historical industry dynamic text units, and then compares them with the association result description of the initial association rule to determine whether the two are consistent.

[0069] Step S1334: Calculate the application accuracy of each initial association rule, where the application accuracy is the proportion of the number of times the actual association result matches the association result description to the total number of triggers.

[0070] Based on the comparison results, the rule optimization module calculates the application accuracy of each initial association rule. The application accuracy is calculated by dividing the number of times the actual association result matches the association result description by the total number of times the initial association rule is triggered. The resulting ratio reflects the applicability of the initial association rule in historical data.

[0071] Step S1335: Extract the application accuracy of each initial association rule in historical industry dynamic data in different domains, and generate the rule domain adaptation distribution.

[0072] The rule optimization module further extracts the application accuracy of each initial association rule in historical industry dynamic data across different domains. For example, it separately calculates the application accuracy of the rule in historical data from different domains such as finance, technology, and healthcare. Then, it compiles these accuracy data from different domains to generate a rule domain adaptation distribution, which reflects the adaptation of the initial association rule across different domains.

[0073] Step S1336: Construct an adaptation evaluation index system based on the application accuracy and rule domain adaptation distribution. The adaptation evaluation index system includes an overall accuracy index and a domain adaptation index.

[0074] Based on the calculated application accuracy and the generated rule domain adaptation distribution, the rule optimization module constructs an adaptation evaluation index system. Among them, the overall accuracy index is mainly determined based on the application accuracy, reflecting the accuracy of the rule in the overall historical data; the domain adaptation index is determined based on the rule domain adaptation distribution, reflecting the degree of adaptation balance of the rule in different domains, such as the fluctuation of application accuracy in different domains.

[0075] Step S1337: Calculate the fit score of each initial association rule according to the fit evaluation index system. The fit score is obtained by weighted summation of the overall accuracy index score and the domain fit index score.

[0076] The rule optimization module calculates the fit score for each initial association rule based on the fit evaluation index system. First, it assigns weights to both the overall accuracy index and the domain fit index, with the weights determined by their importance in the fit evaluation. Then, it multiplies the overall accuracy score by its weight and the domain fit index score by its weight, and finally adds the two products to obtain the fit score for each initial association rule.

[0077] Step S1338: Compare the fit score of each initial association rule with the preset fit threshold, and record the initial association rules whose fit scores exceed the preset fit threshold.

[0078] The rule optimization module compares the fit score of each initial association rule with a preset fit threshold. If the fit score of an initial association rule exceeds the preset fit threshold, the rule is considered to have good fit and is recorded. These recorded initial association rules form a candidate association rule set.

[0079] Step S134: Perform rule conflict detection on the candidate association rules in the candidate association rule set to identify candidate association rule pairs with semantic conflicts.

[0080] The rule generation module performs rule conflict detection on the candidate association rules in the candidate association rule set. By analyzing the association triggering conditions and association result descriptions of different candidate association rules, it determines whether there are semantic contradictions or conflicts. For example, if two candidate association rules address the same association triggering condition but provide completely opposite association result descriptions, this constitutes a pair of candidate association rules with semantic conflicts. Through the above detection, all candidate association rule pairs with semantic conflicts are identified.

[0081] Step S135: Perform rule fusion processing on candidate association rule pairs with semantic conflicts, readjust the association triggering conditions and association result descriptions based on the association semantic direction and association frequency, and generate fused association rules.

[0082] For candidate association rule pairs identified as having semantic conflicts, the rule generation module performs rule fusion processing. Combining the association semantic direction and association frequency of this pair of candidate association rules, the association triggering conditions are re-examined and adjusted to make them more accurate. At the same time, based on the association semantic direction and association frequency, the description of the association results is corrected and integrated to eliminate semantic conflicts and generate a new fused association rule that can comprehensively reflect the reasonable parts of the original two conflicting rules.

[0083] Step S136: Integrate the fused association rules with the non-conflicting candidate association rules to obtain a preliminary association rule set. Call the rule iteration module of the dynamic semantic tracking model to dynamically iterate and optimize the preliminary association rule set. Adjust the rule parameters according to the hierarchical association relationship of the latest industry dynamic text units to generate an industry dynamic semantic association rule set.

[0084] The rule generation module integrates the fused association rules with the candidate association rules that do not conflict, forming a preliminary set of association rules. Then, it calls the rule iteration module of the dynamic semantic tracking model to dynamically iterate and optimize the preliminary set of association rules.

[0085] Step S1361: Obtain the latest industry dynamic text unit set, perform hierarchical relationship analysis on the latest industry dynamic text unit set, and obtain the latest hierarchical relationship.

[0086] The rule iteration module first obtains the latest set of industry dynamic text units, which are recently released content related to the target industry. Then, following the method for constructing hierarchical relationships in step S120, it analyzes this latest set of industry dynamic text units to obtain the latest hierarchical relationships between them.

[0087] Step S1362: Apply each preliminary association rule in the preliminary association rule set to the latest hierarchical association relationship, and count the number of hits and misses of each preliminary association rule in the latest hierarchical association relationship.

[0088] The rule iteration module applies each preliminary association rule from the initial association rule set to the latest hierarchical association relationship. It checks whether the latest hierarchical association relationship meets the association triggering conditions of each preliminary association rule. If it does, it records the number of hits; otherwise, it records the number of misses.

[0089] Step S1363: Calculate the latest fit rate for each preliminary association rule, where the latest fit rate is the ratio of the number of hits to the total number of applications.

[0090] Based on the statistically obtained hit and miss counts, the rule iteration module calculates the latest fit rate for each preliminary association rule. The total number of applications is the sum of the hit and miss counts, and the latest fit rate equals the number of hits divided by the total number of applications. This metric reflects the fit of the preliminary association rule in the latest data.

[0091] Step S1364: For the preliminary association rules whose latest fit rate is lower than the preset iteration threshold, analyze the reasons for the miss. The reasons for the miss include overly strict association trigger conditions and deviations in the description of association results.

[0092] The rule iteration module compares the latest fit rate of each preliminary association rule with a preset iteration threshold. For preliminary association rules with a latest fit rate lower than the preset iteration threshold, it analyzes the reasons for their failure to match. Reasons for failure may include overly stringent association trigger conditions (i.e., the trigger conditions are set too strictly, making them rarely met in practice); or deviations in the description of the association results (i.e., the description of the association results deviates from the actual association situation, resulting in a mismatch even if the trigger conditions are met).

[0093] Step S1365: Adjust the parameters of the preliminary association rule according to the reason for the miss. If the reason for the miss is that the association trigger condition is too strict, relax the frequency requirement in the association trigger condition; if the reason for the miss is that the description of the association result is biased, correct the semantic expression in the description of the association result.

[0094] Based on the reasons for the missed associations obtained from the analysis, the rule iteration module adjusts the parameters of the initial association rules. If the reason for the missed associations is that the association triggering conditions are too strict, the frequency requirements in the association triggering conditions are appropriately relaxed, for example, by reducing the numerical requirements for the association frequency. If the reason for the missed associations is that the description of the association results is biased, the semantic expression in the description of the association results is corrected to make it more consistent with the actual association situation.

[0095] Step S1366: Reapply the adjusted preliminary association rules to the latest hierarchical association relationship and calculate the adjusted latest fit rate.

[0096] The rule iteration module reapplies the adjusted initial association rules to the latest hierarchical association relationships, recounts the number of hits and misses, and calculates the adjusted latest fit rate.

[0097] Step S1367: Repeat the parameter adjustment and adaptation rate calculation steps until the latest adaptation rate of the initial association rule exceeds the preset iteration threshold to obtain the optimized association rule. Integrate all the optimized association rules to generate an industry-dynamic semantic association rule set. Each semantic association rule in the semantic association rule set includes the optimized association trigger condition, association result description and latest adaptation rate.

[0098] The rule iteration module repeatedly performs the steps of parameter adjustment and fit rate calculation. That is, it continuously adjusts the parameters of the initial association rule and recalculates its latest fit rate in the latest hierarchical association relationship until the latest fit rate exceeds the preset iteration threshold. At this point, the optimized association rule is obtained. All optimized association rules are integrated to generate an industry-dynamic semantic association rule set. Each semantic association rule contains the optimized association trigger condition, association result description, and corresponding latest fit rate.

[0099] Step S140: Construct association links for industry dynamic text units based on the semantic association rule set to obtain dynamic semantic tracking links for industry dynamics.

[0100] After generating the set of semantic association rules, the association link construction module in the computer system will construct association links for industry dynamic text units based on the set of semantic association rules, thereby obtaining the dynamic semantic tracking links of industry dynamics.

[0101] Step S141: Perform node encoding processing on the industry dynamic text units, and assign a unique text node identifier to each industry dynamic text unit. The text node identifier includes source code and text sequence number.

[0102] The association link construction module first performs node encoding processing on each industry dynamic text unit. A unique text node identifier is assigned to each industry dynamic text unit. This identifier consists of a source code and a text sequence number. The source code corresponds to the source identifier of the industry dynamic text unit and is used to identify its source. The text sequence number is set to distinguish different text units from the same source. For example, if the source code of a certain industry dynamic text unit represents an official channel of a certain company, and the text sequence number is 001, then its text node identifier consists of the source code and 001.

[0103] Step S142: Based on each semantic association rule in the set of semantic association rules, filter out industry dynamic text unit pairs that meet the association triggering conditions to obtain a set of rule-matched text unit pairs.

[0104] The association link construction module filters industry dynamic text units based on each semantic association rule in the semantic association rule set. It checks whether each pair of industry dynamic text units meets the association triggering condition in the semantic association rule. If it does, the pair of industry dynamic text units is determined as a text unit pair that conforms to the rule. All the above text unit pairs form the rule-matched text unit pair set.

[0105] Step S143: Establish an association edge for each rule-matched text unit pair. The attributes of the association edge include the association rule identifier and the association confidence, and the association confidence is determined based on the fit rate of the semantic association rule.

[0106] For each rule-matching text unit pair in the rule-matching text unit pair set, the association link construction module establishes an association edge for it. The attributes of the association edge include the association rule identifier and the association confidence. The association rule identifier indicates which semantic association rule the association edge is based on, and the association confidence is determined based on the fit rate of the semantic association rule. The higher the fit rate, the higher the association confidence. For example, the fit rate of the semantic association rule can be directly used as the association confidence, or the fit rate can be converted into the association confidence through a certain conversion formula.

[0107] Step S144: Organize the text node identifiers with associated edges according to the association relationship to construct an initial semantic association graph, wherein the nodes in the initial semantic association graph are text node identifiers and the edges are associated edges.

[0108] The association link construction module organizes text node identifiers with associated edges according to their relationships to build an initial semantic association graph. In this graph, each node represents a text node identifier, i.e., an industry dynamic text unit, and the edges between nodes are the established association edges, used to represent the association relationships between industry dynamic text units.

[0109] Step S145: Perform edge filtering on the initial semantic association graph, delete the edges with association confidence lower than the preset confidence threshold, and obtain the filtered semantic association graph. Extract the association paths from the filtered semantic association graph. Each association path contains multiple text node identifiers arranged in association order, and adjacent text node identifiers are connected by association edges.

[0110] The association link construction module performs association edge filtering on the initial semantic association graph. A preset confidence threshold is set, and association edges with a confidence level below this threshold are removed from the initial semantic association graph, resulting in a filtered semantic association graph. Then, association paths are extracted from the filtered semantic association graph. An association path is a sequence of multiple text node identifiers arranged in an association order, where adjacent text node identifiers in the sequence are connected by association edges, indicating an association relationship.

[0111] Step S146: Perform semantic coherence analysis on each associated path, retain associated paths whose semantic coherence meets the preset coherence requirements, prioritize the retained associated paths according to the degree of semantic coherence, and generate dynamic semantic tracking links for industry dynamics.

[0112] The association link construction module performs semantic coherence analysis on each association path.

[0113] For example, step S1461: Extract the core semantic content of the industry dynamic text unit corresponding to each text node identifier in the associated path. The core semantic content is extracted from the industry dynamic text unit by a text summarization algorithm.

[0114] First, for each text node identifier in the association path, the association link construction module finds its corresponding industry dynamic text unit, and then extracts the core semantic content from the industry dynamic text unit through a text summarization algorithm (such as an extraction-based summarization algorithm that selects key sentences and words in the text). The core semantic content is the content that can summarize the main meaning of the text unit.

[0115] Step S1462: Convert each core semantic content into a semantic vector to obtain a sequence of semantic vectors corresponding to the associated path. Calculate the cosine similarity between adjacent semantic vectors in the semantic vector sequence to obtain a sequence of adjacent semantic similarities.

[0116] The association link construction module converts each core semantic content into a semantic vector. This can be achieved using a pre-trained language model that maps text content into multi-dimensional semantic vectors. The semantic vectors corresponding to all core semantic content in the association path are arranged sequentially to obtain a sequence of semantic vectors for the association path. Then, the cosine similarity between any two adjacent semantic vectors in this sequence is calculated to obtain an adjacent semantic similarity sequence. Each element in this adjacent semantic similarity sequence reflects the degree of similarity between the core semantics of two adjacent text units in the association path.

[0117] Step S1463: Calculate the average value of adjacent semantic similarity sequences as an overall coherence index of the associated path; extract the common semantic components of all semantic vectors in the associated path, calculate the similarity between each semantic vector and the common semantic components, and obtain the component similarity sequence.

[0118] The association link construction module calculates the average value of adjacent semantic similarity sequences. This average value serves as an overall coherence indicator of the association path, reflecting the overall semantic coherence of the association path. Simultaneously, common semantic components of all semantic vectors in the association path are extracted using methods such as principal component analysis. Then, the similarity between each semantic vector and these common semantic components is calculated, resulting in a component similarity sequence. This component similarity sequence reflects the degree of fit between each semantic vector and the common semantic components.

[0119] Step S1464: Calculate the average value of the component similarity sequence as the semantic consistency index of the association path; construct a semantic coherence scoring model based on the overall coherence index and the semantic consistency index, wherein the semantic coherence scoring model is a weighted sum of the overall coherence index and the semantic consistency index.

[0120] The average value of the component similarity sequence is calculated and used as a semantic consistency index for the association path, reflecting the degree of consistency among the semantic vectors in the association path on common semantic components. Then, a semantic coherence scoring model is constructed based on the overall coherence index and the semantic consistency index. This semantic coherence scoring model is a weighted sum of the overall coherence index and the semantic consistency index, where the weights of the two indices are determined according to their importance in evaluating semantic coherence.

[0121] Step S1465: Calculate the semantic coherence score for each associated path using the semantic coherence scoring model. The semantic coherence score is used to determine whether the associated path meets the preset coherence requirements.

[0122] The associated link construction module calculates the semantic coherence score of each associated path through a semantic coherence scoring model. This score is compared with a preset coherence requirement (i.e., a preset scoring threshold), and the associated paths with semantic coherence scores reaching or exceeding this threshold are retained, that is, the associated paths with semantic coherence meeting the preset coherence requirement. Then, according to the level of the semantic coherence scores, the retained associated paths are sorted by priority, and the sorted associated paths together constitute the dynamic semantic tracking link of industry dynamics.

[0123] Step S150: Based on the dynamic semantic tracking link, extract the core semantic development direction of industry dynamics and determine the core tracking dimension of industry dynamics tracking.

[0124] After obtaining the dynamic semantic tracking link, the core dimension determination module in the computer system will, based on this dynamic semantic tracking link, extract the core semantic development direction of industry dynamics, so as to determine the core tracking dimension of industry dynamics tracking.

[0125] Step S151: Conduct semantic topic mining on each associated path in the dynamic semantic tracking link to obtain the path semantic topic corresponding to each associated path, and the path semantic topic is extracted from the core semantic content of the associated path through a topic modeling algorithm.

[0126] The core dimension determination module conducts semantic topic mining on each associated path in the dynamic semantic tracking link.

[0127] For example, step S1511: Extract the complete text content of the industry dynamics text units corresponding to all text node identifiers in each associated path.

[0128] The core dimension determination module first extracts the complete text content of the industry dynamics text units corresponding to all text node identifiers in each associated path, and these complete text contents will be used as the original data for semantic topic mining.

[0129] Step S1512: Perform text cleaning processing on the complete text content, perform word segmentation on the text content after cleaning, and split the text content after cleaning into multiple independent lexical units.

[0130] Perform text cleaning processing on the extracted complete text content, including removing special symbols, punctuation marks, stop words (such as meaningless words like "of", "is", "in", etc.) in the text, etc., to obtain the text content after cleaning. Then, perform word segmentation on the text content after cleaning and split it into multiple independent lexical units. For example, for Chinese text, a word segmentation method based on a dictionary or statistics can be used for splitting.

[0131] Step S1513: Perform part-of-speech tagging on the vocabulary units, filter out vocabulary units of noun, verb and adjective types to form an effective vocabulary set, and call the pre-trained word embedding model to perform vector transformation on each vocabulary unit in the effective vocabulary set to obtain a vocabulary vector set.

[0132] The word segments obtained are labeled with part-of-speech tags to determine the part of speech of each word segment, such as noun, verb, adjective, adverb, etc. Then, word segments with the parts of speech of noun, verb, and adjective are selected to form a valid vocabulary set, as these words usually carry more semantic information. Next, a pre-trained word embedding model (such as Word2Vec, GloVe, etc.) is used to perform vector transformation on each word segment in the valid vocabulary set, mapping each word segment to a multi-dimensional vector, resulting in a vocabulary vector set.

[0133] Step S1514: Input the vocabulary vector set into the input layer of the topic modeling model, and perform feature extraction and dimensionality reduction on the vocabulary vector set through the hidden layer of the topic modeling model to obtain topic feature vectors.

[0134] The set of lexical vectors is input into the input layer of a topic modeling model (such as LDA or LSA). The hidden layer of the topic modeling model extracts features from the input set of lexical vectors, captures the latent semantic relationships between words, and performs dimensionality reduction processing to reduce the dimensionality of the data, resulting in more concise topic feature vectors that reflect the core topic information.

[0135] Step S1515: Calculate the topic probability distribution of the topic feature vector through the output layer of the topic modeling model to generate a path semantic topic distribution. The path semantic topic distribution contains multiple topics and their corresponding probability values. Select the topic with the highest probability value in the path semantic topic distribution as the path semantic topic corresponding to the associated path, and complete the semantic topic mining of each associated path.

[0136] The output layer of the topic modeling model calculates the topic probability distribution on the topic feature vectors to generate a path semantic topic distribution. This path semantic topic distribution contains multiple possible topics and their corresponding probability values. A higher probability value indicates a better match between the topic and the semantic content of the associated path. The topic with the highest probability value in the path semantic topic distribution is selected as the path semantic topic corresponding to that associated path, thus completing the semantic topic mining for each associated path.

[0137] Step S152: Count the occurrence frequency of path semantic topics corresponding to all associated paths, and generate a topic frequency statistics list, which includes path semantic topics and their corresponding occurrence frequency.

[0138] The core dimension determination module counts the occurrence frequency of path semantic topics corresponding to all associated paths. For each path semantic topic, it records the number of times it appears in all associated paths, and then compiles the path semantic topics and their corresponding occurrence frequencies into a topic frequency statistics list.

[0139] Step S153: Filter high-frequency path semantic topics according to the topic frequency statistics list. The high-frequency path semantic topics refer to path semantic topics that appear more than a preset frequency threshold.

[0140] The core dimension determination module filters high-frequency path semantic topics based on the topic frequency statistics list. A preset frequency threshold is set, and path semantic topics that appear more than the threshold in the topic frequency statistics list are identified as high-frequency path semantic topics. These topics appear frequently in the dynamic semantic tracking link and may represent important directions in industry dynamics.

[0141] Step S154: Perform semantic clustering processing on the high-frequency path semantic topics. Use a hierarchical clustering algorithm to group high-frequency path semantic topics with high semantic similarity into the same topic cluster. Each topic cluster contains multiple high-frequency path semantic topics.

[0142] The core dimension determination module performs semantic clustering on high-frequency path semantic topics. First, it calculates the semantic similarity between each high-frequency path semantic topic, which can be achieved by converting topics into vectors and then calculating the similarity between the vectors. Then, a hierarchical clustering algorithm is used to gradually merge high-frequency path semantic topics with high semantic similarity from bottom to top, grouping them into the same topic cluster, ultimately forming multiple topic clusters. Each topic cluster contains multiple semantically similar high-frequency path semantic topics.

[0143] Step S155: Extract the cluster core semantics of each topic cluster, which is obtained by merging the core words of all high-frequency path semantic topics in the topic cluster.

[0144] For each topic cluster, the core dimension determination module extracts its cluster core semantics. It identifies the core words for each high-frequency path semantic topic in the topic cluster. These core words are the words that best represent the topic. Then, these core words are merged, and duplicate words are removed to form the cluster core semantics that can summarize the main content of the topic cluster.

[0145] Step S156: Define preliminary tracking dimensions based on the cluster core semantics. Each preliminary tracking dimension corresponds to a topic cluster, and the name of the preliminary tracking dimension is determined by the cluster core semantics.

[0146] The core dimension determination module defines preliminary tracking dimensions based on the core semantics of each thematic cluster. Each preliminary tracking dimension corresponds to a thematic cluster, and its name is determined according to the cluster's core semantics, accurately reflecting the industry dynamics represented by that thematic cluster. For example, if the core semantics of a thematic cluster is "development of new energy vehicle battery technology," then the corresponding preliminary tracking dimension could be named "New Energy Vehicle Battery Technology Development Dimension."

[0147] Step S157: Analyze the semantic correlation between different initial tracking dimensions, and calculate the semantic correlation coefficient between any two initial tracking dimensions. The semantic correlation coefficient is determined based on the similarity of the core semantics of the cluster.

[0148] The core dimension determination module analyzes the semantic correlation between different initial tracking dimensions. It calculates the similarity between the cluster core semantics corresponding to any two initial tracking dimensions, and then converts this similarity into a semantic correlation coefficient. The higher the semantic correlation coefficient, the stronger the semantic correlation between the two initial tracking dimensions.

[0149] Step S158: Merge the initial tracking dimensions whose semantic correlation coefficient exceeds the preset correlation threshold to obtain the core tracking dimensions for industry dynamic tracking. Each core tracking dimension includes the topic cluster and cluster core semantics corresponding to the merged initial tracking dimensions.

[0150] The core dimension determination module merges preliminary tracking dimensions whose semantic correlation coefficients exceed a preset correlation threshold. These preliminary tracking dimensions are semantically closely related, and after merging, they form a more comprehensive core tracking dimension. Each core tracking dimension includes the topic clusters corresponding to all the merged preliminary tracking dimensions and the core semantics of these topic clusters, thereby determining the core tracking dimension for industry dynamic tracking.

[0151] Step S160: Integrate and process the semantic information in the dynamic semantic tracking link according to the core tracking dimension to generate an industry dynamic analysis report.

[0152] After determining the core tracking dimensions, the report generation module in the computer system will integrate and process the semantic information in the dynamic semantic tracking link according to these core tracking dimensions, and finally generate an industry dynamic analysis report.

[0153] Step S161: Assign each associated path in the dynamic semantic tracking link to the corresponding core tracking dimension. Each core tracking dimension corresponds to a subset of associated paths, and the subset of associated paths contains all associated paths that semantically match the core tracking dimension.

[0154] The report generation module assigns each associated path in the dynamic semantic tracking chain to the corresponding core tracking dimension. By comparing the path semantic theme of the associated path with the cluster core semantic of the core tracking dimension, all associated paths that match the semantics of a certain core tracking dimension are grouped into the associated path subset corresponding to that core tracking dimension. Each core tracking dimension has a corresponding associated path subset.

[0155] Step S162: Extract semantic information for each subset of associated paths, extracting the core semantic content, association attributes, and path priority of each associated path to obtain a set of dimensional semantic information.

[0156] For each subset of associated paths, the report generation module extracts semantic information. It extracts core semantic content (i.e., the core semantic content of each text unit in the path), association attributes (such as association rule identifiers, association confidence, etc.), and path priority (i.e., the priority ranking of the path in the dynamic semantic tracking chain) from each associated path. It then gathers this extracted information to form a set of dimensional semantic information corresponding to each core tracking dimension.

[0157] Step S163: Perform structured processing on the dimensional semantic information set to construct a dimensional semantic data table, which includes a path identifier field, a core semantic content field, an association attribute field, and a path priority field.

[0158] The report generation module performs structured processing on the dimensional semantic information set to construct a dimensional semantic data table. This table contains multiple fields, including a path identifier field to record the unique identifier of the associated path; a core semantic content field to store the core semantic content of the associated path; an association attribute field to record the association attributes of the associated path; and a path priority field to record the path priority of the associated path. This structured processing makes the dimensional semantic information clearer and more organized.

[0159] Step S164: Perform frequency statistics on the core semantic content in the dimensional semantic data table, calculate the number of times each core semantic content appears in the dimensional semantic data table, and generate the core semantic frequency distribution.

[0160] The report generation module performs frequency statistics on the core semantic content in the dimensional semantic data table. For each core semantic content, it counts the number of times it appears in the dimensional semantic data table, and then combines the core semantic content and its corresponding occurrence count to generate a core semantic frequency distribution. This core semantic frequency distribution reflects the frequency of occurrence of different core semantic content in that core tracking dimension.

[0161] Step S165: Based on the frequency distribution and path priority of the core semantics, construct a semantic development trend map of the core tracking dimension. The semantic development trend map is used to show the frequency changes and path priority distribution of different core semantic content under the core tracking dimension.

[0162] The report generation module constructs a semantic development trend chart for the core tracking dimensions based on the frequency distribution of core semantics and path priority. In this chart, the horizontal axis can represent time or path order, and the vertical axis can represent frequency of occurrence. Different lines or bars display the changes in the frequency of occurrence of different core semantic content; simultaneously, different colors or markers represent the distribution of path priority. The semantic development trend chart can intuitively display the development and changes of dynamic semantics in the industry under the core tracking dimensions.

[0163] Step S166: Perform trend analysis on the semantic development trend map of each core tracking dimension, and summarize the semantic change direction and key focus of industry dynamics under the core tracking dimensions.

[0164] The report generation module performs trend analysis on the semantic development trend map of each core tracking dimension. By observing the changing patterns of the frequency of occurrence and the distribution of path priority of core semantic content in the semantic development trend map, it summarizes the direction of semantic changes in industry dynamics under that core tracking dimension. For example, it identifies which core semantic contents are increasing in frequency, which are decreasing, and the industry dynamic trends reflected by these changes. At the same time, it determines the key content to focus on under that core tracking dimension, that is, the industry dynamic information corresponding to the core semantic contents with high frequency and high priority.

[0165] Step S167: Organize the dimension semantic data tables, core semantic frequency distribution, semantic development trend chart and trend analysis results of all core tracking dimensions according to the preset report framework, and supplement the overall semantic summary of industry dynamics. The overall semantic summary of industry dynamics is generated based on the trend analysis results of all core tracking dimensions.

[0166] The report generation module organizes the semantic data tables, frequency distributions, trend charts, and trend analysis results of all core tracking dimensions according to a preset report framework. This framework can include separate analysis sections for each core tracking dimension, with each section presenting the aforementioned information for that dimension sequentially. Simultaneously, based on the trend analysis results of all core tracking dimensions, it integrates and generates a comprehensive semantic summary of industry dynamics. This summary outlines the overall development trends of the entire industry, key areas of common concern, and the relationships between the core tracking dimensions.

[0167] Step S168: Merge the organized content with the overall semantic summary content to generate an industry dynamic analysis report, which includes a core tracking dimension analysis section and an overall semantic summary section.

[0168] The report generation module combines the content organized according to the preset report framework with the overall semantic summary to form a complete industry dynamics analysis report. This industry dynamics analysis report includes a core tracking dimension analysis section and an overall semantic summary section. The core tracking dimension analysis section details the specific analysis results under each core tracking dimension, while the overall semantic summary section provides an overall overview and summary of industry dynamics, offering users comprehensive and systematic industry dynamics analysis information.

[0169] Figure 2 The illustration shows exemplary hardware and software components of an AI-based industry dynamic tracking semantic analysis system 100 that can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the AI-based industry dynamic tracking semantic analysis system 100 and to perform the functions in this application.

[0170] The AI-based industry dynamic tracking semantic analysis system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the AI-based industry dynamic tracking semantic analysis method of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0171] For example, the AI-based industry dynamic tracking semantic analysis system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the AI-based industry dynamic tracking semantic analysis system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The AI-based industry dynamic tracking semantic analysis system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0172] For ease of explanation, only one processor is described in the AI-based industry dynamic tracking semantic analysis system 100. However, it should be noted that the AI-based industry dynamic tracking semantic analysis system 100 of this application may also include multiple processors, and therefore the steps performed by one processor as described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the AI-based industry dynamic tracking semantic analysis system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0173] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned industry dynamic tracking semantic analysis method based on artificial intelligence is implemented.

[0174] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. An artificial intelligence-based industry dynamic tracking semantic analysis method, characterized in that, The method comprises: acquiring an industry dynamic text set containing a plurality of industry dynamic text units with source identifiers; constructing a source semantic hierarchical mapping of the industry dynamic text units based on the source identifiers to obtain a hierarchical association relationship between the industry dynamic text units; calling a pre-trained dynamic semantic tracking model to perform dynamic association rule generation processing on the hierarchical association relationship to generate a semantic association rule set of the industry dynamics; constructing an association link of the industry dynamic text units according to the semantic association rule set to obtain a dynamic semantic tracking link of the industry dynamics; extracting a core semantic development direction of the industry dynamics based on the dynamic semantic tracking link to determine a core tracking dimension of the industry dynamic tracking; integrating and processing semantic information in the dynamic semantic tracking link according to the core tracking dimension to generate an industry dynamic analysis report. 2.The AI-based industry dynamics tracking semantic analysis method of claim 1, wherein, The method comprises: parsing the source identifier of each industry dynamic text unit to extract source subject information, source field information and source publishing channel information in the source identifier; constructing a source subject hierarchical system based on the source subject information, the source subject hierarchical system containing a plurality of hierarchical subject nodes, each subject node corresponding to a unique subject code, and different subject nodes being connected through subject association strength; classifying the industry dynamic text units according to the source field information to obtain field text groups divided by fields, each field text group corresponding to at least one subject node in the source subject hierarchical system; extracting field semantic features of the industry dynamic text units in each field text group, the field semantic features containing field core vocabulary and field expression style features; calculating semantic similarity between different field text groups based on the field semantic features to generate a field semantic similarity matrix, elements in the field semantic similarity matrix representing the semantic similarity between two field text groups; constructing a source semantic hierarchical mapping based on the source subject hierarchical system and the field semantic similarity matrix, the source semantic hierarchical mapping containing a bidirectional association relationship between the subject hierarchical dimension and the field semantic dimension; extracting a hierarchical association relationship between the industry dynamic text units from the source semantic hierarchical mapping, the hierarchical association relationship containing an intra-hierarchical text unit association relationship and a cross-hierarchical text unit association relationship. 3.The AI-based industry dynamics tracking semantic analysis method of claim 2, wherein, The method comprises: performing deduplication processing on the source subject information of all industry dynamic text units to obtain a unique source subject set; extracting subject attribute information of each unique source subject, the subject attribute information containing subject type, subject size and subject industry influence; performing first-level classification on the unique source subjects based on the subject type to obtain a plurality of first-level subject categories, each first-level subject category containing a plurality of unique source subjects with the same subject type; According to the main body scale, the unique source main body in each first-level main body category is classified into a second-level main body subcategory, and a plurality of second-level main body subcategories are obtained, each second-level main body subcategory containing a plurality of unique source main bodies with the same main body scale; Based on the main body industry influence, the unique source main body in each second-level main body subcategory is ranked in a third level, and the hierarchical position of each unique source main body in the second-level main body subcategory is determined; Each unique source main body is assigned a corresponding main body code, and the main body code contains a first-level main body category identifier, a second-level main body subcategory identifier, and a hierarchical position identifier; The main body association strength between different unique source main bodies is calculated, and the main body association strength is determined based on the main body cooperation history information and the main body information interaction frequency; The unique source main body with the main body association strength is organized according to the hierarchical classification result, and a source main body hierarchical system is constructed, wherein each main body node in the source main body hierarchical system contains a main body code, main body attribute information, and main body association strength of associated main body nodes. 4.The AI-based industry dynamics tracking semantic analysis method of claim 1, wherein, The pre-trained dynamic semantic tracking model is called to generate the dynamic association rules of the hierarchical association relationship, and a set of industry dynamic semantic association rules is generated, including: The hierarchical association relationship is input into the rule initialization module of the dynamic semantic tracking model to extract the association attribute information in the hierarchical association relationship, and the association attribute information includes the association type, the association frequency, and the association semantic direction; Based on the association attribute information, an initial association rule pool is constructed, and each initial association rule in the initial association rule pool contains an association trigger condition and an association result description, and the association trigger condition is set based on the association type and the association frequency; Each initial association rule in the initial association rule pool is evaluated by the rule optimization module of the dynamic semantic tracking model, and the initial association rules with an adaptation evaluation result exceeding a preset adaptation threshold are filtered out to form a candidate association rule set; Rule conflict detection is performed on the candidate association rules in the candidate association rule set to identify candidate association rule pairs with semantic conflicts; The candidate association rule pairs with semantic conflicts are subjected to rule fusion processing, the association trigger condition and the association result description are adjusted based on the association semantic direction and the association frequency, and a fused association rule is generated; The fused association rule and the candidate association rules without conflicts are integrated to obtain a preliminary association rule set, and the rule iteration module of the dynamic semantic tracking model is called to dynamically iterate and optimize the preliminary association rule set, the rule parameters are adjusted according to the hierarchical association relationship of the latest industry dynamic text unit, and a set of industry dynamic semantic association rules is generated. 5.The AI-based industry dynamics tracking semantic analysis method of claim 4, wherein, The adaptation evaluation of each initial association rule in the initial association rule pool by the rule optimization module of the dynamic semantic tracking model includes: A historical industry dynamic data set is obtained, and the historical industry dynamic data set contains a plurality of historical industry dynamic text units and corresponding historical hierarchical association relationships; Each initial association rule in the initial association rule pool is applied to the historical industry dynamic data set, and the number of times each initial association rule is triggered in the historical industry dynamic data set is counted. For each initial association rule, the actual association result corresponding to the historical industry dynamic data triggering the initial association rule is extracted, and the actual association result is compared with the association result description of the initial association rule; The application accuracy of each initial association rule is calculated, and the application accuracy is the proportion of the number of times that the actual association result is consistent with the association result description to the total number of triggering times; The application accuracy of each initial association rule in different field historical industry dynamic data is extracted, and a rule field adaptation distribution is generated; An adaptation degree evaluation index system is constructed based on the application accuracy and the rule field adaptation distribution, and the adaptation degree evaluation index system includes an overall accuracy index and a field adaptability index; The adaptation degree score of each initial association rule is calculated according to the adaptation degree evaluation index system, and the adaptation degree score is obtained by weighted summation of the overall accuracy index score and the field adaptability index score; The adaptation degree score of each initial association rule is compared with a preset adaptation threshold, and the initial association rule whose adaptation degree score exceeds the preset adaptation threshold is recorded. 6.The AI-based industry dynamics tracking semantic analysis method of claim 4, wherein, The rule iteration module calling the dynamic semantic tracking model performs dynamic iteration optimization on the preliminary association rule set, including: A latest industry dynamic text unit set is obtained, and hierarchical association relationship analysis is performed on the latest industry dynamic text unit set to obtain a latest hierarchical association relationship; Each preliminary association rule in the preliminary association rule set is applied to the latest hierarchical association relationship, and the hit number and the miss number of each preliminary association rule in the latest hierarchical association relationship are counted; The latest adaptation rate of each preliminary association rule is calculated, and the latest adaptation rate is the proportion of the hit number to the total application number; For the preliminary association rule with a latest adaptation rate lower than a preset iteration threshold, the miss reason is analyzed, and the miss reason includes a too strict association triggering condition and an association result description deviation; The parameters of the preliminary association rule are adjusted according to the miss reason. If the miss reason is that the association triggering condition is too strict, the frequency requirement in the association triggering condition is relaxed. If the miss reason is the association result description deviation, the semantic expression in the association result description is corrected. The adjusted preliminary association rule is re-applied to the latest hierarchical association relationship, and the adjusted latest adaptation rate is calculated; The parameter adjustment and adaptation rate calculation steps are repeatedly performed until the latest adaptation rate of the preliminary association rule exceeds the preset iteration threshold, and an optimized association rule is obtained. The semantic association rule set of the industry dynamic is generated by integrating all the optimized association rules, and each semantic association rule in the semantic association rule set includes an optimized association triggering condition, an association result description and a latest adaptation rate. 7.The AI-based industry dynamics tracking semantic analysis method of claim 1, wherein, According to the semantic association rule set, the industry dynamic text unit is associated to construct a dynamic semantic tracking link of the industry dynamic, including: The industry dynamic text unit is subjected to node coding processing, and each industry dynamic text unit is assigned a unique text node identifier, and the text node identifier includes a source code and a text serial number; Based on each semantic association rule in the semantic association rule set, a rule matching text unit pair set is obtained by screening industry dynamic text unit pairs meeting the association triggering condition. An association edge is established for each rule matching text unit pair, and an attribute of the association edge includes an association rule identifier and an association confidence, which is determined based on an adaptation rate of the semantic association rule; Text node identifiers with association edges are organized according to an association relationship to construct an initial semantic association graph, wherein nodes in the initial semantic association graph are text node identifiers, and edges are association edges; An association edge screening process is performed on the initial semantic association graph, and association edges with an association confidence lower than a preset confidence threshold are deleted to obtain a screened semantic association graph, and an association path is extracted from the screened semantic association graph, wherein each association path includes a plurality of text node identifiers arranged in an association order, and adjacent text node identifiers are connected by an association edge; A semantic coherence analysis is performed on each association path, and an association path with a semantic coherence meeting a preset coherence requirement is retained, the retained association paths are prioritized according to a semantic coherence degree, and a dynamic semantic tracking link of an industry trend is generated. 8.The AI-based industry dynamics tracking semantic analysis method of claim 1, wherein, The dynamic semantic tracking link is used to extract a core semantic development direction of the industry trend, and a core tracking dimension of the industry trend tracking is determined, including: A semantic topic mining is performed on each association path in the dynamic semantic tracking link to obtain a path semantic topic corresponding to each association path, which is extracted from a core semantic content of the association path by a topic modeling algorithm; A topic frequency statistics list is generated by counting the number of occurrences of the path semantic topics corresponding to all association paths, and the topic frequency statistics list includes the path semantic topics and corresponding occurrence numbers; High-frequency path semantic topics are screened according to the topic frequency statistics list, and the high-frequency path semantic topics refer to path semantic topics with an occurrence number exceeding a preset frequency threshold; A semantic clustering process is performed on the high-frequency path semantic topics, and a hierarchical clustering algorithm is used to classify high-frequency path semantic topics with a high semantic similarity into a same topic cluster, and each topic cluster includes a plurality of high-frequency path semantic topics; A cluster core semantic of each topic cluster is extracted, which is obtained by merging core vocabularies of all high-frequency path semantic topics in the topic cluster; Preliminary tracking dimensions are defined based on the cluster core semantics, each preliminary tracking dimension corresponds to a topic cluster, and a name of the preliminary tracking dimension is determined by the cluster core semantic; A semantic association degree between different preliminary tracking dimensions is analyzed, and a semantic association coefficient of any two preliminary tracking dimensions is calculated, which is determined based on a similarity of the cluster core semantics; Preliminary tracking dimensions with a semantic association coefficient exceeding a preset association threshold are merged to obtain core tracking dimensions of the industry trend tracking, and each core tracking dimension includes a topic cluster corresponding to the merged preliminary tracking dimension and a cluster core semantic. 9.The industry dynamics tracking semantic analysis method based on artificial intelligence according to claim 1, wherein, The semantic information in the dynamic semantic tracking link is integrated according to the core tracking dimensions to generate an industry trend analysis report, including: assigning each associated path in the dynamic semantic tracking link to a corresponding core tracking dimension, each core tracking dimension corresponding to an associated path subset containing all associated paths matching the core tracking dimension semantics; performing semantic information extraction on each associated path subset to extract core semantic content, associated relationship attributes, and path priority in each associated path, obtaining a dimension semantic information set; performing structured processing on the dimension semantic information set to construct a dimension semantic data table containing a path identification field, a core semantic content field, an associated relationship attribute field, and a path priority field; performing frequency statistics on the core semantic content in the dimension semantic data table to calculate the number of occurrences of each core semantic content in the dimension semantic data table, generating a core semantic frequency distribution; based on the core semantic frequency distribution and the path priority, constructing a semantic development trend chart of the core tracking dimension, the semantic development trend chart used to show the frequency change of different core semantic contents and the path priority distribution under the core tracking dimension; performing trend analysis on the semantic development trend chart of each core tracking dimension to summarize the semantic change direction and key content of industry dynamics under the core tracking dimension; organizing the dimension semantic data table, the core semantic frequency distribution, the semantic development trend chart, and the trend analysis result of all core tracking dimensions according to a preset report framework, supplementing the overall semantic summary content of industry dynamics, the overall semantic summary content of industry dynamics being integrated and generated based on the trend analysis result of all core tracking dimensions; merging the organized content and the overall semantic summary content to generate an industry dynamic analysis report, the industry dynamic analysis report containing a core tracking dimension analysis part and an overall semantic summary part.

10. An artificial intelligence-based industry dynamic tracking semantic analysis system, characterized by, The industry dynamic tracking semantic analysis system based on artificial intelligence includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to realize the industry dynamic tracking semantic analysis method based on artificial intelligence in any one of claims 1-9.

Images