Dynamic knowledge graph construction method, system and equipment

Abstract

The invention discloses a dynamic knowledge graph construction method, system and device, and relates to the technical field of knowledge graph extension.The method comprises the steps that the similarity between an external text and a historical text is calculated; entities and events in the external text are extracted, the entities and the events contained in predefined entities and event words are defined as keywords, and the keyword coverage degree is calculated; when the similarity is smaller than a similarity threshold value, or the similarity is larger than or equal to the similarity threshold value and the keyword coverage degree is larger than or equal to a coverage degree threshold value, the current external text is reserved; reasoning the relationship between entities on the retained external text to obtain new knowledge; and when the new knowledge conflicts with the existing knowledge in the knowledge graph, embedding the finally reserved knowledge into the knowledge graph according to evaluation standards of intra-chain consistency and cross-chain consistency to complete updating of the knowledge graph. The method adapts to complex and changeable requirements of internal management of enterprises, and realizes efficient construction of the dynamic knowledge graph.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph expansion technology, and in particular to a method, system and device for constructing a dynamic knowledge graph. Background Art

[0002] In today's era of digital information explosion, knowledge graphs are becoming increasingly important as a powerful knowledge organization and management tool. Especially in internal enterprise management scenarios, knowledge graphs can integrate various types of dispersed information, providing strong support for decision-making and business process optimization.

[0003] In the real-world context of internal enterprise management, the construction of dynamic knowledge graphs is irreplaceable and highly practical. For example, in the case of a company's organizational structure, employees need real-time updates to clarify reporting lines, collaboration partners, and other information during their daily work. However, due to the lack of timely updates to traditional knowledge graphs, employees often lack timely access to organizational structure information. This can lead to miscommunication and work flow errors in cross-departmental collaborative projects due to misunderstandings about personnel's departments and responsibilities, thus impacting the overall efficiency of the project. In financial systems, if the knowledge graphs relied upon by financial personnel are not updated with tax policies and preferential information, they cannot ensure that the knowledge they rely on in financial processing is up to date. This can lead to companies missing out on tax incentives or even facing legal risks due to improper tax treatment. In the field of intellectual property, companies need to constantly monitor patent developments and trademark changes within the industry and promptly update their knowledge graphs to effectively protect their intellectual property, avoid infringement disputes, and seize potential collaboration and development opportunities. In vertical fields, such as the production process knowledge of chemical companies and the programming knowledge of software companies, new knowledge is promptly integrated into the knowledge graph to help companies improve their core competitiveness and promote product innovation and service optimization.

[0004] Updating knowledge graphs often requires significant human resources. Firstly, internal enterprise data sources are extensive and complex, encompassing personnel organizational structures, financial systems, intellectual property, and information closely related to the company's operations and management. For example, within the personnel organizational structure, information such as employee position changes and department adjustments must be accurately and promptly entered into the knowledge graph. Within the financial system, new interpretations of financial policies and changes in tax information, such as adjustments to tax types and preferential policies, require professional analysis and integration. However, manual processing is not only inefficient but also prone to human error, resulting in low accuracy in knowledge updates. Furthermore, with the continuous expansion of business operations and the dynamic changes in the market environment, new knowledge is constantly emerging. The speed at which human resources can update knowledge graphs far outstrips the rate of knowledge generation, resulting in extremely slow iteration of the entire knowledge graph. For example, in the vertical fields of chemical companies, new production processes, new material research and development results, and other new knowledge are constantly being generated. Similarly, in software companies, new programming techniques and methodologies are constantly being updated. Traditional knowledge graph update methods struggle to quickly incorporate this new knowledge, severely impacting the timeliness and practicality of knowledge graphs in enterprise operations and management.

[0005] In the existing technologies, the screening of new knowledge expansion and updates in knowledge graphs often uses a single judgment standard (such as screening only by text similarity threshold), or relies on fixed thresholds for static filtering, which has significant limitations: (1) Insufficient screening accuracy: A single similarity threshold makes it difficult to strike a balance between eliminating redundant information and retaining innovative knowledge. For example, if only a high similarity threshold is used for screening, new texts may be misjudged as redundant information due to subtle differences between them and historical knowledge (such as the addition of key details), resulting in the omission of valuable new knowledge. If the threshold is too low, a large amount of duplicate or low-quality information will be introduced, increasing storage and maintenance costs.

[0006] (2) Poor adaptability: Fixed thresholds cannot match diverse data source characteristics (e.g., the difference in credibility between authoritative and general sources), system operating conditions (e.g., the need to reduce data processing during high load), and knowledge importance (e.g., the need for more refined screening of knowledge related to core entities). For example, using a fixed threshold for low-credibility data sources may result in erroneous information entering the graph; using a unified standard for high-priority knowledge may introduce noise due to overly loose screening.

[0007] (3) Risk of missing key information: Traditional methods lack the ability to identify high-value content in texts. Although some new texts have a high degree of similarity with historical knowledge, they contain key event words (such as breakthrough technologies in the industry and changes in core entity attributes). If they are only filtered by similarity, they will be mistakenly deleted, causing the graph to miss important updates.

[0008] Furthermore, as knowledge graphs continue to expand and update, conflicts between new and existing knowledge are inevitable. These conflicts arise from various sources (e.g., differing credibility of sources), changes in information timeliness (e.g., policy adjustments, factual updates), and biased perspectives (e.g., differing descriptions of the same event from multiple perspectives). Existing technologies have significant limitations in addressing knowledge conflicts: (1) Simplified conflict resolution logic: Most methods use static rules of "either this or that", such as retaining the latest knowledge, retaining authoritative source knowledge, or manual judgment. This approach ignores the contextual relevance of knowledge. For example, a new piece of knowledge from a common source that is highly consistent with the recent event chain may be mistakenly discarded due to the low priority of the source; and an outdated authoritative source knowledge may be retained due to its authoritative attribute, resulting in the graph containing invalid information.

[0009] (2) Lack of utilization of knowledge relevance: Existing technologies often compare conflicting knowledge features (such as timestamps and source weights) in isolation, without considering the event chain structure in which the knowledge exists. For example, if two conflicting pieces of "product release time" information are separated from the "product development-testing-release" event chain to which they belong, it is difficult to determine their correctness through isolated comparison alone. (3) Improper handling of cross-event chain conflicts: When conflicting knowledge belongs to different event chains, existing technologies often use equal weight comparison without considering the quality differences of the event chains themselves. For example, when a piece of knowledge from a fragmented, low-credibility event chain conflicts with a piece of knowledge from a complete, high-credibility event chain, equal comparison may result in the correct knowledge of the high-quality chain being discarded. In summary, existing technologies have problems such as low accuracy, poor adaptability, and omission of key information in new information screening, and problems of incorrect retention or correct abandonment in new knowledge integration and conflict resolution, which reduces the quality of knowledge graphs. Summary of the Invention

[0010] In order to solve the above problems, the present invention proposes a method, system and equipment for constructing a dynamic knowledge graph, which adopts a dynamic dual-threshold trigger mechanism to construct a hybrid architecture of lightweight change detection triggering, large model precise focusing and knowledge incremental cascade propagation, which can adapt to the complex and changing needs of internal enterprise management and realize the efficient construction of dynamic knowledge graphs.

[0011] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for constructing a dynamic knowledge graph, comprising: Calculate the similarity between the obtained external text and the historical text of the corresponding type; Extract entities and events from external texts, define entities and events contained in predefined entity and event words as keywords, and calculate keyword coverage; When the similarity is less than the similarity threshold, or the similarity is greater than or equal to the similarity threshold and the keyword coverage is greater than or equal to the coverage threshold, the current external text is retained; wherein the similarity threshold and the coverage threshold are dynamically adjusted according to the system load, the credibility of the data source and the importance of the keyword; Reasoning about the relationships between entities in the retained external text to obtain new knowledge; When new knowledge conflicts with existing knowledge in the knowledge graph, determine whether the new knowledge and the conflicting knowledge belong to the same event chain. If so, calculate the consistency score between the new knowledge and other knowledge in the event chain, and retain the current new knowledge when the consistency score is greater than or equal to the consistency threshold; if not, calculate the consistency score between the new knowledge and the conflicting knowledge and other knowledge in their respective event chains, and weight the consistency scores of the new knowledge and the conflicting knowledge based on the chain quality, and retain the knowledge with the higher score after weighting; The final retained knowledge is embedded into the knowledge graph to complete the update of the knowledge graph.

[0012] As an optional implementation, the system load is calculated by taking a weighted sum of CPU usage, memory usage, and the ratio of the number of currently processed requests to the maximum number of requests to be processed; The data source credibility is calculated based on the historical accuracy, update frequency and expert rating of the data source; The keyword importance is as follows: different weights are given to predefined entities and event words to determine the weight of each keyword in the external text, and the maximum weight among them is taken as the keyword importance.

[0013] As an optional implementation, the data source credibility is: ; in, is the credibility of the data source; n is the amount of data to be evaluated in the data source; is the historical accuracy of the i-th data to be evaluated; is the update frequency of the i-th data to be evaluated; Rating the experts; 、 and are the weight coefficients of historical accuracy, update frequency and expert rating respectively, and .

[0014] As an optional implementation method, the update frequency is: calculate the average value of the intervals between the release times of all historical data related to the data to be evaluated in the data source, take the interval between the release time of the first historical data related to the data to be evaluated and the release time of the last data as the total duration, and calculate the update frequency as the ratio of the average value to the total duration.

[0015] As an optional implementation method, the historical accuracy rate is: the similarity between the data to be evaluated and the official data is calculated. If it is greater than a set threshold, it is considered accurate, thereby determining the amount of data judged to be accurate, and the ratio of the amount of data judged to be accurate to the amount of all historical data related to the data to be evaluated is taken as the historical accuracy rate.

[0016] As an optional implementation, the process of dynamically adjusting the similarity threshold and coverage threshold according to system load, data source credibility, and keyword importance is as follows: ; ; in, and is the adjusted similarity threshold and coverage threshold; and are the similarity threshold and coverage threshold before adjustment; is the load threshold, is the credibility threshold, and are load factors, and are all credibility coefficients, is the importance coefficient, 、 and They are system load, data source credibility and keyword importance.

[0017] As an optional implementation, the reasoning process further includes: first using predefined rules for reasoning, and then using the large model for reasoning if reasoning cannot be achieved using the predefined rules.

[0018] As an optional implementation, the events that arrive continuously in the knowledge graph are organized into event chains according to time, causality, and co-reference relationships; Conflict involves the same causes leading to opposite effects and the same effects being caused by opposite consequences.

[0019] As an optional implementation, the chain quality is obtained by weighted calculation based on the credibility of the data source of each event in the event chain and the length of the event chain.

[0020] In a second aspect, the present invention provides a dynamic knowledge graph construction system, comprising: a similarity calculation module configured to calculate the similarity between the acquired external text and the historical text of the corresponding type; a coverage calculation module configured to extract entities and events from external text, define entities and events contained in predefined entity and event words as keywords, and calculate keyword coverage; The filtering module is configured to retain the current external text when the similarity is less than a similarity threshold, or when the similarity is greater than or equal to the similarity threshold and the keyword coverage is greater than or equal to the coverage threshold; wherein the similarity threshold and the coverage threshold are dynamically adjusted according to system load, data source credibility, and keyword importance; The reasoning module is configured to reason about the relationship between entities on the retained external text to obtain new knowledge; The fusion module is configured to determine whether the new knowledge and the conflicting knowledge belong to the same event chain when there is a conflict between the new knowledge and the existing knowledge in the knowledge graph. If so, the consistency score of the new knowledge and the other knowledge in the event chain is calculated, and the current new knowledge is retained when the consistency score is greater than or equal to the consistency threshold. If not, the consistency scores of the new knowledge and the conflicting knowledge with the other knowledge in their respective event chains are calculated separately, and the consistency scores of the new knowledge and the conflicting knowledge are weighted based on the chain quality, and the knowledge with the higher weighted score is retained; The update module is configured to embed the final retained knowledge into the knowledge graph to complete the update of the knowledge graph.

[0021] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0022] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a method, system and equipment for constructing a dynamic knowledge graph, which can adapt to the complex and changeable needs of internal enterprise management, realize the efficient construction of dynamic knowledge graphs, break the traditional large-scale model's full-scale processing of streaming data, adopt a dynamic dual-threshold trigger mechanism, and construct a hybrid architecture of lightweight change detection triggering, large-scale model precise focusing and knowledge incremental cascade propagation. The dynamic dual-threshold trigger mechanism is used for efficient data filtering and preliminary processing, quickly captures high-value fragments, greatly reduces the large model load from the source, reduces computational complexity, and applies time series event chain embedding and incremental fusion to the screened high-value data streams for deep event chain modeling, conflict resolution and fusion, thereby reducing computational complexity. It is suitable for scenarios with data torrents, uneven change distribution and stringent real-time requirements.

[0023] This paper proposes a method based on temporal event chain embedding and incremental fusion. Traditional methods handle the update of each event / relationship in isolation, resulting in a large amount of repeated calculations and context loss. Therefore, the method of the present invention explicitly models the temporal dependencies of event chains. Through incremental event chain embedding and conflict resolution, it significantly improves the efficiency and consistency of information fusion, enhances knowledge quality and reasoning capabilities, and is particularly suitable for fields with dense events, complex logical chains, and strong temporal dependencies, such as news narrative analysis, business process monitoring, financial transaction monitoring, and social media trend tracking. It deepens understanding while ensuring efficiency.

[0024] To solve the problems of low accuracy, poor adaptability, and omission of key information in the existing technology for screening new information, the present invention proposes a dual-dimensional screening and dynamic adjustment mechanism of similarity threshold and keyword coverage threshold, which has the following beneficial effects: (1) Improve the accuracy and comprehensiveness of new knowledge screening. By using the collaborative judgment of similarity threshold and keyword coverage threshold, we can not only filter out duplicate or low-relevance information through similarity (reducing redundancy), but also capture high-value content (such as important knowledge containing key event words) through keyword coverage, avoiding the problem of mistakenly deleting valuable knowledge or mistakenly including redundant information due to a single standard. For example, text with high similarity to historical knowledge but containing changes in core entity attributes can be retained if the keyword coverage meets the standard, ensuring that the graph is promptly included in key updates. (2) Dynamic adaptation to diverse scenario requirements. Similarity thresholds and coverage thresholds can be adjusted in real time based on system load, data source credibility, entity importance, etc., improving the flexibility of the screening mechanism. This dynamic adaptation capability enables knowledge graph updates to adapt to different business scenarios and enhance practicality. (3) Reduce manual intervention costs and improve update efficiency. The dual-threshold quantitative judgment logic reduces the need for manual review of boundary data. At the same time, it ensures judgment consistency through clear screening rules and accelerates the iteration speed of the knowledge graph. (4) Optimize the quality of knowledge graphs and enhance their application value. Through precise screening, knowledge graphs can effectively eliminate redundant and low-quality information while enriching high-value knowledge, thereby improving the knowledge density and accuracy of the graph. This allows applications based on knowledge graphs, such as intelligent retrieval and decision analysis, to rely on more reliable underlying data, reducing decision biases caused by erroneous or redundant knowledge, and enhancing the application value of knowledge graphs in various fields.

[0025] (5) Enhance the robustness of knowledge updating. The introduction of keyword coverage provides a supplementary judgment dimension for texts with similar semantics but containing new information, avoiding the masking of knowledge increments due to similar text expression styles, and improving the knowledge graph's ability to capture subtle knowledge changes.

[0026] To address the issues of incorrect retention or correct abandonment in new knowledge integration and conflict resolution, the present invention adopts a conflict resolution logic based on event chain attribution judgment, consistency score calculation, and chain quality weighting, which has the following beneficial effects: (1) Improve the accuracy of conflict judgment. Introducing event chains as a reference framework for conflict handling upgrades the correctness judgment of knowledge from isolated comparison to contextual collaborative verification. For conflicts within the same event chain, the compatibility of new knowledge is judged by the consistency score between the new knowledge and other knowledge in the chain, avoiding misjudgment due to a single feature. For conflicts across event chains, chain quality is weighted to ensure that knowledge from high-quality chains is retained first.

[0027] (2) Dynamically adapt to conflict handling in complex scenarios. Differentiate between the two types of conflicts: within the same event chain and across event chains, and adopt differentiated handling strategies: within the same event chain, quickly filter compatible new knowledge through consistency thresholds to simplify the conflict handling process for similar events; across event chains, balance the credibility of knowledge from different sources through chain quality weighting, which can resolve the conflict problem between outdated knowledge from authoritative sources and the latest knowledge from non-authoritative sources.

[0028] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0030] Figure 1 Flowchart of the dynamic knowledge graph construction method provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0033] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "comprise" and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0035] Example 1 like Figure 1 As shown, this embodiment provides a method for constructing a dynamic knowledge graph, including: S101: Calculating the similarity between the acquired external text and the historical text of the corresponding type; S102: extracting entities and events from external text, defining entities and events contained in predefined entity and event words as keywords, and calculating keyword coverage; S103: When the similarity is less than the similarity threshold, or the similarity is greater than or equal to the similarity threshold and the keyword coverage is greater than or equal to the coverage threshold, retain the current external text; wherein the similarity threshold and the coverage threshold are dynamically adjusted according to the system load, the credibility of the data source, and the importance of the keyword; S104: Reasoning about the relationship between entities in the retained external text to obtain new knowledge; S105: When new knowledge conflicts with existing knowledge in the knowledge graph, determine whether the new knowledge and the conflicting knowledge belong to the same event chain. If so, calculate the consistency score between the new knowledge and other knowledge in the event chain, and retain the current new knowledge when the consistency score is greater than or equal to the consistency threshold. If not, calculate the consistency scores between the new knowledge and the conflicting knowledge and other knowledge in their respective event chains, and assign weights to the consistency scores of the new knowledge and the conflicting knowledge based on the chain quality, and retain the knowledge with the higher weighted score. S106: Embed the final retained knowledge into the knowledge graph to complete the update of the knowledge graph.

[0036] In this embodiment, the external text may be obtained by crawling a crawler engine and may be field dynamics, news, or other external text information. For example, in a financial system, text information such as new financial policy interpretations, tax information changes, tax adjustments, and preferential information may be obtained.

[0037] After acquiring the external text, the lightweight Sentence-BERT model is used. Sentence-BERT is a pre-trained model based on BERT (Bidirectional Encoder Representations from Transformers). It encodes both the external text and the corresponding historical text into fixed-length embedding vector representations. The lightweight Sentence-BERT model is used to generate semantic embedding vectors (dense vector representations of the text), and the model output dimension can be set to 384. These embedding vectors effectively capture the semantic information of the text data and provide a basis for subsequent similarity calculations.

[0038] For example, for the external text "A company releases a new mobile phone" and the corresponding historical text "A company launches innovative products", the pre-trained model converts them into corresponding embedding vectors. Each dimension of these embedding vectors contains features related to the semantics of the text.

[0039] After encoding the acquired external text and the corresponding type of historical text into embedding vectors, the similarity between the two embedding vectors is calculated; the similarity ranges from -1 to 1, where a larger value indicates that the semantics of the two texts are closer, and -1 indicates opposite semantics.

[0040] It is understandable that common similarity measurement methods such as cosine similarity and Euclidean distance are used to calculate the similarity between the embedded vector of the obtained external text and the embedded vector of the corresponding type of historical text. Taking cosine similarity as an example, its calculation formula is: ; in, is similarity; is the embedding vector of the external text; is the embedding vector of the corresponding type of historical text.

[0041] In this embodiment, an existing named entity recognition (NER) tool is used to identify entities (such as names of people, places, organizations, etc.) in external text. For keyword extraction of non-named entities, TF-IDF (Term Frequency-Inverse Document Frequency, a commonly used weighting technology for information retrieval and data mining), RAKE (Rapid Automatic keyword extraction) algorithm, or graph-based methods (PageRan) can be used to extract event words.

[0042] For example, for the external text "Company A holds a new product launch conference in City B", the entities "Company A" and "City B" can be extracted, and the event "holding a new product launch conference" can be extracted.

[0043] A set of high-value entities and events is predefined within the corresponding research domain or application scenario. These typically refer to entities and activities, behaviors, or state changes within a specific domain. For example, in the financial sector, high-value entities may include large financial institutions and major listed companies; high-value events may involve major policy adjustments, corporate mergers and acquisitions, and IPOs.

[0044] Then, determine whether the extracted entity and event words are included in the predefined entity and event words; if so, define the extracted entity and event words as keywords; Calculate keyword coverage based on the quotient of the number of keywords in the external text and the total number of predefined entity and event words ;Right now .

[0045] In this embodiment, dynamic dual thresholds are set, namely, a similarity threshold and a coverage threshold.

[0046] Specifically: For similarity threshold : According to the similarity distribution between historical text pairs, set an appropriate percentile (such as 90% percentile) as the similarity threshold , which means that only new external texts with a similarity below this percentile compared to historical texts of the corresponding type will be considered "sufficiently different" to trigger further processing.

[0047] Therefore, the specific judgment logic is: if the similarity between the obtained external text and the historical text of the corresponding type is greater than or equal to the similarity threshold, the new external text is considered to be highly similar to the historical text, and the current information has not been updated, so it may not be retained or may only require a simple update operation, without starting a comprehensive downstream processing flow.

[0048] If the similarity between the obtained external text and the historical text of the corresponding type is lower than the similarity threshold, the external text is considered to be sufficiently novel, and the importance of the key entities or events it contains needs to be further evaluated.

[0049] For coverage threshold When the acquired external text contains high-value entities or events, even if the overall similarity is high, downstream processing should be considered. This is because even subtle changes may carry important information.

[0050] Therefore, the judgment logic based on the dynamic dual threshold is as follows: First, the similarity is evaluated. If the similarity is less than the similarity threshold, the current external text is directly retained because the new external text data is novel enough.

[0051] If the similarity is greater than or equal to the similarity threshold, the keyword coverage is evaluated. If the keyword coverage is greater than or equal to the coverage threshold, the current external text is retained and the next step of processing is triggered. This is because even if the overall similarity is high, the new external text data may contain important updates or changes.

[0052] Other cases do not require special treatment.

[0053] According to the above logical analysis, the following conditions are constructed : .

[0054] In this embodiment, although a fixed threshold can be directly set for the setting of the similarity threshold and the coverage threshold based on the experience of domain experts or the needs of specific application scenarios, a fixed threshold cannot guarantee that high-value information is processed first during traffic peaks. Therefore, this embodiment designs dynamic adjustment of the dual thresholds based on system load, data source credibility, and entity importance to ensure that high-value information is processed first during traffic peaks.

[0055] The adjustment rules are as follows: (1) System load.

[0056] The system load is calculated by CPU usage, memory usage, and the ratio of the number of currently processed requests to the maximum number of requests that can be handled: ; Among them, CPU represents CPU usage, memory represents memory usage, and request represents the ratio of the current number of processed requests to the maximum number of requests that can be handled. 、 and are the weight coefficients of the three respectively, and .

[0057] When the system load is higher than the set load threshold, increase and To reduce the amount of downstream processing data, thereby reducing the burden on the system; when the system load is lower than the load threshold, reduce and This allows more data to pass through and be processed further.

[0058] (2) Credibility of data source.

[0059] For example, if the external text obtained is data crawled from a website, the credibility of the data source can be evaluated based on its historical accuracy, update frequency, and expert ratings: ; Where n is the amount of data to be evaluated in the data source; represents the historical accuracy of the i-th data to be evaluated; Indicates the update frequency of the i-th data to be evaluated; Indicates expert rating, that is, an expert gives a rating on whether the data is important / authoritative; , and Represent the weight coefficients of historical accuracy, update frequency and expert rating respectively, and .

[0060] Among them, the historical accuracy, update frequency and expert scores of the data source can be normalized before calculation to unify the measurement.

[0061] The update frequency can be calculated by obtaining the release time of all historical data related to the data to be evaluated in the data source, calculating the average value based on the interval between each release time, taking the interval between the release time of the first historical data related to the data to be evaluated and the release time of the last data as the total duration, and calculating the update frequency as the ratio of the average value to the total duration.

[0062] Among them, based on objective and authoritative reference standards such as official data, industry specifications, and expert consensus, the accuracy of information crawled from a third-party website is verified, and the similarity between the data to be evaluated and the official data (or industry standards, etc.) is calculated. If it is greater than the set threshold, it is considered accurate. The amount of data judged to be accurate is determined, and the ratio of it to all historical data related to the data to be evaluated is used as the historical accuracy rate.

[0063] It is understandable that the above calculation process can be adaptively modified by those skilled in the art according to specific fields and application scenarios, and is not specifically limited.

[0064] When the credibility of a data source is greater than the set credibility threshold, it is considered a high-credibility data source and the credibility is reduced. and , because these sources may contain more valuable information; on the contrary, for low-credibility data sources, increasing and , triggering processing only when the data is significantly different or contains high-value content.

[0065] (3) Keyword importance imp.

[0066] For predefined entities and event words, different weights are given according to their importance in the knowledge graph. At the same time, the same entities are compared to determine the number of times their attributes change within a set time period. If the number exceeds the set threshold, it is considered a high-frequency change entity and a higher attention weight can be set because changes in these entities may indicate important information updates or trend changes. This can be customized according to the actual application scenario without specific limitations.

[0067] Each entity generally contains multiple attributes. For example, for a tax type, its attributes include validity period, preferential treatment, exemption method, tax object, etc.

[0068] Then, the weight of each keyword in the acquired external text is determined, and the maximum weight among them is used as the keyword importance.

[0069] If the keyword importance is greater than the set importance threshold, it means that the new data contains entities / events of high importance. Even if the similarity is high, the importance should be appropriately reduced. To ensure that these data are processed in a timely manner. On the contrary, for data that does not contain or only contains low-importance entities, the Reduce unnecessary processing overhead.

[0070] Alternatively, the importance of keywords can be determined based on factors such as the importance score of the keyword's role in the business, the impact score, and the user attention score; ; Among them, RIB represents the importance score of the keyword's role in the business, impact represents the impact score of the keyword on other processes or entities, and UA represents the user's attention score to the keyword. 、 and represents the weight coefficient, and .

[0071] It is understandable that the importance score of the keyword role in the business, the scope of influence score, the user attention score, etc. can all be customized by those skilled in the art according to the actual application scenario needs, and no detailed description is given.

[0072] In summary, the process of dynamically adjusting the similarity threshold and coverage threshold based on system load, data source credibility, and keyword importance is as follows: ; ; in, and is the adjusted similarity threshold and coverage threshold; Indicates the load threshold, represents the confidence threshold, and are load factors, and are all credibility coefficients, is the importance coefficient.

[0073] This embodiment method builds a lightweight change detection trigger system by calculating semantic similarity, identifying key entities and events, and calculating keyword coverage. It employs a dual-threshold trigger mechanism to effectively filter out repetitive or unimportant information while ensuring timely response to truly valuable updates or changes. This system not only enables rapid response to changes in data streams, but is also resource-efficient.

[0074] In this embodiment, the retained external text is invoked to perform deep reasoning verification on the large model. The reasoning purpose is to obtain the relationship between entities and perform dynamic evolution of the relationship. It is understandable that the trained commonly used large model can be directly invoked for reasoning.

[0075] In addition, the reasoning part can also be designed as follows: first, reasoning is performed through lightweight graph neural networks or predefined rules, such as pairing easily conceivable relationships; for example, "A acquires B" automatically infers that "B becomes a subsidiary of A", and thus "B's CEO may resign".

[0076] If reasoning cannot be achieved through lightweight graph neural networks or predefined rules, a large model is used to reason about the relationships between entities.

[0077] The process can be designed as: (1) Event triggering: detecting new relationship changes or events that affect propagation.

[0078] (2) Preliminary classification: Direct relevance assessment: Determine whether the event involves a direct and clearly defined relationship (e.g., "A acquires B"). If so, proceed to step (3); otherwise, proceed directly to step (5).

[0079] (3) Simple processing: Use predefined rules or lightweight graph neural networks to quickly process events; if the processing result meets all the expected outputs (for example, "B becomes a subsidiary of A"), the process ends; if there is uncertainty or further verification is required, proceed to step (4).

[0080] (4) Local verification: For the results generated by simple processing, verify their rationality through simple logical checks or comparison with known cases. If the verification passes, the process ends; otherwise, it is marked as a complex case and enters step (5).

[0081] (5) Analysis of complex situations: Key Entity Involvement Assessment: Determine whether the incident involves key entities or has the potential for widespread impact. If it involves key entities or has the potential for significant impact, proceed directly to step (6); otherwise, proceed to the next step.

[0082] (6) Resource availability check: Evaluate the current system resource status (such as computing power, time constraints, etc.). If the resources allow for deeper analysis, call the large model for deep reasoning; otherwise, record the event for subsequent processing and try to make the best guess using the existing information.

[0083] (7) Deep Reasoning: Calling the big model to perform detailed causal analysis, multi-level impact assessment, or other required advanced analytical tasks. Update the relationship map or make corresponding decisions based on the insights provided by the big model.

[0084] (8) Feedback loop: New knowledge or pattern learning results obtained from the large model are fed back to the lightweight graph neural network or rule base to optimize future simple processing capabilities.

[0085] To summarize: If the event involves a direct and clearly defined relationship, use a lightweight graph neural network or rule-based inference engine, and if the processing result is reasonable, end the process. Otherwise, if the processing result is uncertain, perform local verification. If it involves important entities or has the potential for significant impact, and resources allow, invoke a large model. Otherwise, if resources are limited, log the event and try a best guess. Finally, use the new knowledge to improve the simple processing tool. This process provides a basic framework that can be further refined and adjusted according to specific application scenarios and needs.

[0086] In this embodiment, consecutive events in the knowledge graph are organized into event chains based on temporal, causal, and coreference relationships. For example, a company's financial report releases a surge in stock price, leading to an analyst upgrade, forming an event chain. Similarly, when Company A releases a new product, market demand for the product surges, leading to an expansion of production, forming an event chain based on causal relationships. This can be accomplished using a lightweight temporal dependency parser or a small temporal relationship classification model, which will not be detailed here.

[0087] Timing-aware chain embedding maintains a dynamic vector representation for each event chain, using time-decayed vector averaging, an efficient vector arithmetic approach for incrementally trained lightweight RNNs (Recurrent Neural Networks): ; in, is the mean vector, represents the vector corresponding to the i-th event, represents the time decay factor, Represents the timestamp corresponding to the i-th event, Represents a reference event node. You can choose the earliest event time, the current time, or other meaningful time points.

[0088] When new knowledge conflicts with existing knowledge in the knowledge graph, first determine whether the new knowledge and the conflicting knowledge belong to the same event chain; Among them, contradiction is a relationship between events, which usually includes two situations: one is that the same cause leads to opposite results, and the other is that the same result is caused by the opposite result.

[0089] If they belong to the same event chain, the consistency score of the new knowledge and other knowledge in the event chain is calculated (i.e., similarity calculation); if the consistency score is greater than or equal to the set consistency threshold (such as 80%), the current new knowledge is retained, indicating that it is accepted first even if it conflicts with isolated facts; for example, "A denies the merger and acquisition" appearing in the "M&A negotiation chain" may be a negotiation strategy and is more credible than isolated reports.

[0090] If the consistency score is less than the set consistency threshold, the current new knowledge is not accepted.

[0091] If they do not belong to the same event chain, that is, when there is a cross-chain conflict, the consistency scores of the new knowledge and the conflicting knowledge with other knowledge in their respective event chains are calculated separately.

[0092] The chain quality of the event chain containing new knowledge and the event chain containing conflicting knowledge is evaluated separately. Specifically, the chain quality is calculated based on the weighted data source credibility of each event in the event chain and the length of the event chain. The data source credibility and the event chain length can be weighted after normalization. The data source credibility has been explained when dynamically adjusting the similarity threshold and coverage threshold, so it will not be repeated here.

[0093] Combining the consistency score and chain quality factors, the consistency scores of new knowledge and conflicting knowledge are weighted based on the chain quality to generate a comprehensive score, and the knowledge with higher scores after weighting is retained.

[0094] The above rules ensure both intra-chain consistency and cross-chain coordination during the update process, while improving the accuracy and reliability of the entire system. It is understood that these rules can be adjusted and optimized appropriately based on specific application scenarios.

[0095] In this embodiment, the remaining knowledge is embedded into the knowledge graph, which is used to efficiently update the temporal relationships in the knowledge graph and infer potential impacts. The chain embedding itself can be used as a high-level feature for downstream tasks, such as predicting the next event in the chain.

[0096] This embodiment proposes embedding and incremental fusion based on temporal event chains. Traditional methods treat updates to each event / relationship in isolation, resulting in extensive duplication and context loss. Therefore, this embodiment explicitly models the temporal dependencies of event chains. Through incremental event chain embedding and conflict resolution, it significantly improves information fusion efficiency and consistency, enhancing knowledge quality and reasoning capabilities. This approach is particularly suitable for fields with dense events, complex logical chains, and strong temporal dependencies, such as news narrative analysis, business process monitoring, financial transaction monitoring, and social media trend tracking. It deepens understanding while ensuring efficiency.

[0097] In addition, this embodiment breaks the traditional large-scale model of full-scale processing of streaming data, adopts a dynamic dual-threshold trigger mechanism, and constructs a hybrid architecture of lightweight change detection triggering, large-scale model precise focusing, and knowledge incremental cascade propagation. It adopts a dynamic dual-threshold trigger mechanism for efficient data filtering and preliminary processing, quickly captures high-value fragments, significantly reduces the large model load from the source, reduces computational complexity, and applies time series event chain embedding and incremental fusion to the screened high-value data streams for deep event chain modeling, conflict resolution, and fusion, thereby reducing computational complexity. It is suitable for scenarios with data torrents, uneven change distribution, and stringent real-time requirements.

[0098] Example 2 This embodiment provides a dynamic knowledge graph construction system, including: a similarity calculation module configured to calculate the similarity between the acquired external text and the historical text of the corresponding type; a coverage calculation module configured to extract entities and events from external text, define entities and events contained in predefined entity and event words as keywords, and calculate keyword coverage; The filtering module is configured to retain the current external text when the similarity is less than a similarity threshold, or when the similarity is greater than or equal to the similarity threshold and the keyword coverage is greater than or equal to the coverage threshold; wherein the similarity threshold and the coverage threshold are dynamically adjusted according to system load, data source credibility, and keyword importance; The reasoning module is configured to reason about the relationship between entities on the retained external text to obtain new knowledge; The fusion module is configured to determine whether the new knowledge and the conflicting knowledge belong to the same event chain when there is a conflict between the new knowledge and the existing knowledge in the knowledge graph. If so, the consistency score of the new knowledge and the other knowledge in the event chain is calculated, and the current new knowledge is retained when the consistency score is greater than or equal to the consistency threshold. If not, the consistency scores of the new knowledge and the conflicting knowledge with the other knowledge in their respective event chains are calculated separately, and the consistency scores of the new knowledge and the conflicting knowledge are weighted based on the chain quality, and the knowledge with the higher weighted score is retained; The update module is configured to embed the final retained knowledge into the knowledge graph to complete the update of the knowledge graph.

[0099] It should be noted that the systems provided in the above embodiments are only illustrated by the division of the above functional modules when implementing their functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the systems and method embodiments provided in the above embodiments are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0100] In further embodiments, there is also provided: An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed by the processor, wherein when the computer instructions are executed by the processor, the method described in Example 1 is performed. For the sake of brevity, no further details are given here.

[0101] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), off-the-shelf field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0102] The memory may include a read-only memory and a random access memory, and provides instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0103] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is performed.

[0104] The method in Example 1 can be directly implemented as a hardware processor, or can be implemented using a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, it will not be described in detail here.

[0105] A computer program product includes a computer program, which implements the method described in embodiment 1 when executed by a processor.

[0106] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions contained in program modules, which are executed in a device on a real or virtual processor of a target to perform the process / method described above. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided between program modules as needed. The machine-executable instructions for the program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0107] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the computer or other programmable data processing device, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0108] In the context of the present invention, computer program code or related data can be carried by any appropriate carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals include electrical, optical, radio, acoustic, or other forms of propagated signals, such as carrier waves, infrared signals, and the like.

[0109] Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with this embodiment can be implemented using electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0110] It should be noted that the collection and processing of relevant data in this invention, when applied in practice, should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of authorization of laws and regulations and the personal information subject.

[0111] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A method for constructing a dynamic knowledge graph, characterized in that: include: Calculate the similarity between the obtained external text and the historical text of the corresponding type; Extract entities and events from external texts, define entities and events contained in predefined entity and event words as keywords, and calculate keyword coverage; When the similarity is less than the similarity threshold, or the similarity is greater than or equal to the similarity threshold and the keyword coverage is greater than or equal to the coverage threshold, the current external text is retained; wherein the similarity threshold and the coverage threshold are dynamically adjusted according to the system load, the credibility of the data source and the importance of the keyword; Reasoning about the relationships between entities in the retained external text to obtain new knowledge; When new knowledge conflicts with existing knowledge in the knowledge graph, determine whether the new knowledge and the conflicting knowledge belong to the same event chain. If so, calculate the consistency score between the new knowledge and other knowledge in the event chain, and retain the current new knowledge when the consistency score is greater than or equal to the consistency threshold; if not, calculate the consistency score between the new knowledge and the conflicting knowledge and other knowledge in their respective event chains, and weight the consistency scores of the new knowledge and the conflicting knowledge based on the chain quality, and retain the knowledge with the higher score after weighting; The final retained knowledge is embedded into the knowledge graph to complete the update of the knowledge graph.

2. A method for constructing a dynamic knowledge graph according to claim 1, characterized in that: The system load is calculated by weighted summing of CPU usage, memory usage, and the ratio of the number of currently processed requests to the maximum number of requests that can be handled. The data source credibility is calculated based on the historical accuracy, update frequency and expert rating of the data source; The keyword importance is as follows: different weights are given to predefined entities and event words to determine the weight of each keyword in the external text, and the maximum weight among them is taken as the keyword importance.

3. A method for constructing a dynamic knowledge graph according to claim 2, characterized in that: The data source credibility is: ； in, is the credibility of the data source; n is the amount of data to be evaluated in the data source; is the historical accuracy of the i-th data to be evaluated; is the update frequency of the i-th data to be evaluated; Rating the experts; 、 and are the weight coefficients of historical accuracy, update frequency and expert rating respectively, and .

4. A method for constructing a dynamic knowledge graph according to claim 3, characterized in that: The update frequency is calculated as follows: the average of the release time intervals of all historical data related to the data to be evaluated in the data source is calculated, and the interval between the release time of the first and last historical data related to the data to be evaluated is used as the total time interval. The update frequency is calculated as the ratio of the average value to the total time interval. The historical accuracy rate is as follows: the similarity between the data to be evaluated and the official data is calculated. If it is greater than the set threshold, it is considered accurate. The amount of data judged to be accurate is determined, and the ratio of the amount of data judged to be accurate to the amount of all historical data related to the data to be evaluated is the historical accuracy rate.

5. A method for constructing a dynamic knowledge graph according to claim 2, characterized in that: The process of dynamically adjusting the similarity threshold and coverage threshold based on system load, data source credibility, and keyword importance is as follows: ； ； in, and is the adjusted similarity threshold and coverage threshold; and are the similarity threshold and coverage threshold before adjustment; is the load threshold, is the credibility threshold, and are load factors, and are all credibility coefficients, is the importance coefficient, 、 and They are system load, data source credibility and keyword importance.

6. A method for constructing a dynamic knowledge graph according to claim 1, characterized in that: The reasoning process also includes: first using predefined rules for reasoning, and then using the large model for reasoning if reasoning cannot be achieved through the predefined rules.

7. A method for constructing a dynamic knowledge graph according to claim 1, characterized in that: Organize the consecutively arriving events in the knowledge graph into event chains according to time, causality, and coreference relationships; Conflict involves the same causes leading to opposite effects and the same effects being caused by opposite consequences.

8. A method for constructing a dynamic knowledge graph according to claim 1, characterized in that: Chain quality is obtained by weighted calculation based on the credibility of the data source of each event in the event chain and the length of the event chain.

9. A dynamic knowledge graph construction system, characterized in that: include: a similarity calculation module configured to calculate the similarity between the acquired external text and the historical text of the corresponding type; a coverage calculation module configured to extract entities and events from external text, define entities and events contained in predefined entity and event words as keywords, and calculate keyword coverage; The filtering module is configured to retain the current external text when the similarity is less than a similarity threshold, or when the similarity is greater than or equal to the similarity threshold and the keyword coverage is greater than or equal to the coverage threshold; wherein the similarity threshold and the coverage threshold are dynamically adjusted according to system load, data source credibility, and keyword importance; The reasoning module is configured to reason about the relationship between entities on the retained external text to obtain new knowledge; The fusion module is configured to determine whether the new knowledge and the conflicting knowledge belong to the same event chain when there is a conflict between the new knowledge and the existing knowledge in the knowledge graph. If so, the consistency score of the new knowledge and the other knowledge in the event chain is calculated, and the current new knowledge is retained when the consistency score is greater than or equal to the consistency threshold. If not, the consistency scores of the new knowledge and the conflicting knowledge with the other knowledge in their respective event chains are calculated separately, and the consistency scores of the new knowledge and the conflicting knowledge are weighted based on the chain quality, and the knowledge with the higher weighted score is retained; The update module is configured to embed the final retained knowledge into the knowledge graph to complete the update of the knowledge graph.

10. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 8 is completed.

Images