Method and device for constructing multi-dimensional causal event knowledge graph

Abstract

The invention relates to the technical field of causal reasoning, and provides a method and device for constructing a multi-dimensional causal event knowledge graph. The method comprises the following steps: extracting event nodes, relation edges and various characteristics of the event nodes from data; calculating interaction strength among different event nodes, and performing causal relationship reasoning among the different event nodes according to the interaction strength; calculating the feature coupling strength between different features, and analyzing the development law of each feature coupling strength; and performing feature causal evolution reasoning of the event nodes according to the development law of the coupling strength of each feature. According to the method, the characteristics of multiple dimensions are extracted, and the interaction strength and the characteristic coupling strength are calculated by using the multiple dimensions, so that the quantification of the causal relationship is realized, and the causal relationship in the causal event is completely captured by performing causal reasoning from the two dimensions of the event node and the characteristics.

Description

Technical Field

[0001] The present invention relates to the field of causal reasoning technology, and in particular to a method and device for constructing a multidimensional causal event knowledge graph. Background Art

[0002] With the development of informatization, a large amount of unstructured text is generated every day, and many causal relationships and event logics are hidden in these texts. Traditional relational databases are inefficient in processing complex causal chains and are difficult to meet the needs of real-time analysis and intelligent decision-making. Existing methods usually rely on templates and rules for event extraction, which has great limitations and labor costs. In addition, traditional causal reasoning methods are often unable to handle dynamic and complex causal relationships. And with the development of large language models, how to integrate them with causal event knowledge graphs to improve the system's causal reasoning and prediction capabilities has become an important direction for current technological development.

[0003] In view of this, overcoming the defects of the prior art is an urgent problem to be solved in the field of this technology. Summary of the invention

[0004] The technical problem to be solved by the present invention is that existing causal reasoning methods are often unable to handle dynamic and complex causal relationships.

[0005] The present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a method for constructing a multidimensional causal event knowledge graph, comprising:

[0007] Extract event nodes, relationship edges, and various features of event nodes from the data;

[0008] Calculating the interaction strength between different event nodes, and performing causal relationship reasoning between different event nodes based on the interaction strength;

[0009] Calculate the feature coupling strength between different features and analyze the development law of each feature coupling strength; perform feature causal evolution reasoning of event nodes based on the development law of each feature coupling strength.

[0010] Preferably, the calculating the interaction strength between different event nodes specifically includes:

[0011] Calculate the feature matching degree between the first event node and the second event node

[0012] Calculate the interaction strength between the first event node and the second event node IS = k1FS + k2EW ori ; Where N is the number of feature dimensions, w ijis the preset interaction weight between the features of the i-th dimension and the features of the j-th dimension, f i,1 is the feature of the first event node in the i-th dimension, f j,2 is the feature of the second event node in the jth dimension, sim(f i,1 f j,2 ) is used to calculate f i,1 With f j,2 The similarity between them, the feature similarity calculation function, you can choose Euclidean distance or cosine similarity, k1 and k2 are preset coefficients, EW ori is the initial edge weight corresponding to the relationship edge between the first event node and the second event node.

[0013] Preferably, the performing causal relationship reasoning between different event nodes according to the interaction strength specifically includes:

[0014] According to the time characteristics of each event node, searching for the first event node and the second event node that have a sequential relationship in terms of the time characteristics;

[0015] If the interaction strength between the first event node and the second event node is greater than the preset interaction strength, it is inferred that there is a causal relationship between the first event node and the second event node; wherein the event node with an earlier time feature is the cause, and the event node with a later time feature is the result.

[0016] Preferably, the calculating the characteristic coupling strength between different characteristics specifically includes:

[0017] Calculate the characteristic coupling strength between the first feature and the second feature Where M is the number of event nodes, w 1，2 is the preset coupling weight corresponding to the first feature and the second feature, f 1,i is the first feature of the i-th event node, f 2,i is the second feature of the i-th event node, sim(f 1,i f 2,i ) is used to calculate f i,1 With f j,2 The similarity between them, the similarity function can use the normalized Euclidean distance.

[0018] Preferably, when each event node is an event node recording the development of the same event over time, the analysis of the development law of each characteristic coupling strength specifically includes:

[0019] When the feature coupling strength between the first feature and the second feature is greater than the preset coupling strength, the event node is subjected to sliding window sampling with the preset number as the sliding window size, and the sampling feature coupling strength corresponding to each sliding window sampling is calculated; wherein the sampling feature coupling strength is the feature coupling strength between the first feature and the second feature calculated using the event node of the sliding window sampling;

[0020] According to the sampling feature coupling strength corresponding to multiple sliding window samplings, the development law of the feature coupling strength over time is analyzed.

[0021] Preferably, analyzing the development law of the characteristic coupling strength over time according to the sampling characteristic coupling strength corresponding to the multiple sliding window samplings specifically includes:

[0022] Using the sampling characteristic coupling strength at time t and the sampling characteristic coupling strength at time t-1, combined with the first formula, a time decay function of the characteristic coupling strength decaying with time is derived;

[0023] Wherein, the first formula is CS(t)=β×DecayFactor(Δt)×CS(t-1);

[0024] The time decay function is DecayFactor(Δt)=exp(-λ×Δt); wherein β is a preset coefficient, CS(t) is the sampling characteristic coupling strength at time t, CS(t-1) is the sampling characteristic coupling strength at time t-1, Δ t The time interval for sliding window sampling.

[0025] Preferably, the feature causal evolution reasoning of event nodes is performed according to the development law of each feature coupling strength, specifically including:

[0026] According to the time attenuation function, the first formula and the sampling feature coupling strength at the current moment, the sampling feature coupling strength at the next moment is calculated, which is the feature causal evolution result of the event node.

[0027] Preferably, the extraction of event nodes, relationship edges and multiple features of event nodes from the data is achieved by using a preset model for data extraction, and the preset model is G=(V, E, L, M); wherein V is the event node, E is the relationship edge, L is the original feature set, and M is the mapping relationship matrix, which is used to normalize each original feature in the original feature set to obtain multiple features of the event node.

[0028] In a second aspect, the present invention further provides a device for constructing a multidimensional causal event knowledge graph, which is used to implement the method for constructing a multidimensional causal event knowledge graph described in the first aspect, and the device includes:

[0029] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the processor to execute the method for constructing a multidimensional causal event knowledge graph as described in the first aspect.

[0030] In a third aspect, the present invention further provides a non-volatile computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions are executed by one or more processors to complete the method described in the first aspect.

[0031] In a fourth aspect, a chip is provided, comprising: a processor and an interface, for calling and running a computer program stored in the memory from a memory, and executing the method of the first aspect.

[0032] According to a fifth aspect, a computer program product comprising instructions is provided. When the instructions are executed on a computer or a processor, the computer or the processor executes the method according to the first aspect.

[0033] The present invention realizes the quantification of causal relationships by extracting features from multiple dimensions and using multiple dimensions to jointly calculate the interaction strength and feature coupling strength, thereby performing causal reasoning from two dimensions: event nodes and features themselves, to fully capture the causal relationships in causal events. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 It is a flowchart of a method for constructing a multi-dimensional causal event knowledge graph provided by an embodiment of the present invention;

[0036] Figure 2 It is a flowchart of a method for constructing a multi-dimensional causal event knowledge graph provided by an embodiment of the present invention;

[0037] Figure 3 It is a flowchart of a method for constructing a multi-dimensional causal event knowledge graph provided by an embodiment of the present invention;

[0038] Figure 4 It is a flowchart of a method for constructing a multi-dimensional causal event knowledge graph provided by an embodiment of the present invention;

[0039] Figure 5It is a schematic diagram of a method for constructing a multi-dimensional causal event knowledge graph provided by an embodiment of the present invention;

[0040] Figure 6 It is a schematic diagram of a method for constructing a multi-dimensional causal event knowledge graph provided by an embodiment of the present invention;

[0041] Figure 7 It is a schematic diagram of a method for constructing a multi-dimensional causal event knowledge graph provided by an embodiment of the present invention;

[0042] Figure 8 It is a schematic diagram of a method for constructing a multi-dimensional causal event knowledge graph provided by an embodiment of the present invention;

[0043] Fig. 9 It is a schematic diagram of the architecture of an apparatus for constructing a multi-dimensional causal event knowledge graph provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0045] Unless the context requires otherwise, throughout the specification and claims, the term "including" is to be interpreted as open inclusion, that is, "including, but not limited to". In the description of the specification, the terms "one embodiment", "some embodiments", "exemplary embodiments", "examples", "specific examples" or "some examples" and the like are intended to indicate that specific features, structures, materials or characteristics associated with the embodiment or example are included in at least one embodiment or example of the present disclosure. The schematic representation of the above terms does not necessarily refer to the same embodiment or example. In addition, the specific features, structures, materials or characteristics may be included in any one or more embodiments or examples in any appropriate manner, that is, although they may be carried in the embodiments or examples of the above terms due to reasons such as the order and position of appearance, it is not limited to that they can be carried in combination by one embodiment or example.

[0046] In the description of the present invention, the terms "first" and "second" are used only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, the meaning of "multiple" is two or more. In addition, for example, the same type of nouns may be described as two independent individuals by adding "A" and "B" at the end. In this case, the corresponding features defined as "A" and "B" are only used to distinguish the same type of individuals for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features.

[0047] In the description of the present invention, the expression "A and / or B" (where A and B are used to formally represent specific characteristic contents) will be involved, and the corresponding expressions include the following three combinations: only A, only B, and a combination of A and B.

[0048] As used herein, "about," "substantially," or "approximately" includes the stated value and an average value that is within an acceptable range of deviation from the particular value as determined by one of ordinary skill in the art taking into account the measurements in question and the errors associated with the measurement of the particular quantity (i.e., the limitations of the measurement system).

[0049] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0050] Embodiment 1:

[0051] The prior art usually simply maps causal events into nodes and edges in a graph, ignoring the deep integration of multi-dimensional features such as temporal dependency, spatial correlation, and subject interaction between events, and does not consider the cumulative effect of uncertainty in the process of causal chain propagation, which affects the prediction accuracy. In addition, the prior art modeling of causal relationships is too static and cannot effectively capture the dynamic changes and environmental dependencies of causal strength. The lack of in-depth modeling of the interactive relationship between multi-dimensional features in causal events limits the causal reasoning ability. In order to solve these problems, Example 1 of the present invention provides a method for constructing a multi-dimensional causal event knowledge graph, such as Figure 1 As shown, including:

[0052] In step 201, event nodes, relationship edges, and various features of event nodes are extracted from the data, that is, event nodes, relationship edges, and multi-dimensional features of event nodes are extracted from unstructured data, and a basic data structure is constructed for subsequent causal reasoning calculations.

[0053] The extraction of event nodes, relationship edges and multiple features of event nodes from the data is achieved by using a preset model for data extraction, and the preset model is G=(V, E, L, M); wherein V is the event node, E is the relationship edge, L is the original feature set, and M is the mapping relationship matrix, which is used to normalize each original feature in the original feature set to obtain multiple features of the event node.

[0054] In actual application scenarios, the data is unstructured text, and the features include time features, space features, and text features. The preset model G = (V, E, L, M) is used to represent the extraction results, where: V represents the event node set; E represents the relationship edge set between event nodes; L represents the original feature set of event nodes (such as time features, space features, text features, etc.); M represents the mapping relationship matrix, which is used to normalize the original feature set L and map the feature values ​​to a unified scale. It can be understood as follows: if there are n original features in L, then there are n mapping relationships in M. The i-th original feature obtains a normalized value after the i-th mapping relationship, which is the feature of one dimension of the event node. The features of each dimension can be measured on a unified scale after normalization, so that the subsequent calculation of interaction intensity and feature coupling intensity can be performed.

[0055] In a specific implementation, according to the preset model, the features in the unstructured text are classified based on event attribute specifications, and structured features such as time and space are extracted based on rules; the text features are vectorized and feature extracted for event descriptions and event types using a natural language model (such as BERT or Word2Vec), and the above-mentioned technology is directly used to extract event nodes, relationship edges, and multiple features of event nodes to complete the feature extraction process.

[0056] In step 202, the interaction strength between different event nodes is calculated, and causal reasoning between different event nodes is performed based on the interaction strength. That is, by calculating the interaction strength between different event nodes, the correlation between events is quantified, providing a scientific basis for causal reasoning. First, according to the multidimensional features of each event node, the features are classified according to categories such as time, space, text, and numerical values, and an adapted feature extraction method is used for alignment processing to ensure that the features are measured on a unified scale. Subsequently, the comprehensive interaction strength between event nodes is calculated by matching the multidimensional feature similarity and dimensional interaction between nodes, combined with the weights of each dimension. The calculated interaction strength IS is compared with a preset threshold. If IS is greater than the threshold, it is inferred that there may be a causal relationship between the two event nodes. At this time, relevant information is recorded, including the interaction strength value of the causal relationship, the feature matching degree FS, and the edge weight EW. ori , providing support for subsequent causal reasoning.

[0057] In step 203, the feature coupling strength between different features is calculated, and the development law of each feature coupling strength is analyzed; according to the development law of each feature coupling strength, the feature causal evolution reasoning of the event node is performed. Specifically: by calculating the feature coupling strength between different features, the correlation between features is quantified, the development law of the feature coupling strength over time is further analyzed, and the feature causal evolution of the event node is derived in combination with the law. In order to further analyze the change law of the feature coupling strength over time, the sliding window sampling method is used to calculate the feature coupling strength of each sampling, and its change trend over time is modeled. Trend analysis can be achieved through a time series model (such as ARIMA) to capture the dynamic change characteristics of the feature coupling strength. When deriving the feature causal evolution of the event node, the causal direction is analyzed in combination with the change trend of the feature coupling strength. If the coupling strength of a feature shows an obvious time series increase or decrease, it can be inferred that the feature plays a core role in the causal evolution of the event node. At the same time, the feature coupling strength is combined with the interaction strength IS between nodes to further verify the accuracy of the reasoning result.

[0058] The interaction strength can be understood as a representation of whether there is a causal relationship between event nodes, and the feature coupling strength can be understood as a representation of whether there is a causal relationship between different features.

[0059] In actual use, event nodes, relationship edges, multiple features of event nodes, the interaction strength between different event nodes, and the feature coupling strength between different features are stored in a graph database to obtain a knowledge graph; then the Cypher language is used to read the interaction strength between different event nodes and the feature coupling strength between different features from the graph database, so as to perform causal relationship reasoning between different event nodes and feature causal evolution reasoning of event nodes.

[0060] This embodiment realizes the quantification of causal relationships by extracting features from multiple dimensions and using multiple dimensions to jointly calculate the interaction strength and feature coupling strength, thereby performing causal reasoning from two dimensions: event nodes and features themselves, to fully capture the causal relationships in causal events.

[0061] In an optional implementation, the interaction strength between different event nodes is calculated, such as Figure 2 As shown, specifically including:

[0062] In step 301, the feature matching degree between the first event node and the second event node is calculated. The goal of this step is to evaluate the similarity between the first event node and the second event node in terms of different dimensional features, providing a basis for the subsequent calculation of interaction intensity.

[0063] In step 302, the interaction strength IS between the first event node and the second event node is calculated as: IS=k1FS+k2EW ori ; Where N is the number of feature dimensions, w ij is the preset interaction weight between the features of the i-th dimension and the features of the j-th dimension, f i,1 is the feature of the first event node in the i-th dimension, f j,2 is the feature of the second event node in the jth dimension, sim(f i,1 f j,2 ) is used to calculate f i,1 With f j,2 The similarity between them is expressed as f in actual use. i,1 With f j,2 The Euclidean distance or cosine similarity between them, k1 and k2 are preset coefficients, EW ori is the initial edge weight corresponding to the relationship edge between the first event node and the second event node. The preset interaction weight, preset coefficient and initial edge weight are pre-set by technicians in this field according to needs. The goal here is to calculate the interaction strength between the first event node and the second event node based on feature matching and other causal factors, and provide a quantitative basis for causal reasoning. Calculate the interaction strength between the first event node and the second event node IS = k1FS + k2EW ori ; Where IS is the interaction strength, indicating the causal relationship between two event nodes; FS is the feature matching degree calculated in step 301; EW ori It is the initial weight of the relationship edge between the first event node and the second event node. It is a quantitative value of the strength of the causal relationship between the two nodes in the initial state. Its value is in the range of 1.0 to 0.1, indicating the strength weight of the relationship. The weight can be set through empirical rules, regression models or classification models.

[0064] In actual use, there is another implementation of step 301: the calculation formula of the feature matching degree is: Among them, N represents the number of dimensions of the feature; w i is the preset weight of the i-th feature dimension, introducing the importance weight setting of the features with manual intervention; f i,1 f i,2 Respectively represent the feature values ​​of the first event node and the second event node on the i-th feature dimension; sim(f i,1 f i,2 ) represents the similarity calculation function on the i-th feature dimension, which can be implemented by methods such as cosine similarity or Euclidean distance. The feature matching degree FS is an indicator to measure the similarity between two event nodes in the feature space, and the calculation result is used for further analysis of the interaction intensity.

[0065] Performing causal relationship reasoning between different event nodes according to the interaction intensity, such as Figure 3 shown, specifically including:

[0066] In step 401, according to the time characteristics of each event node, find the first event node and the second event node with a sequential relationship in time characteristics. That is, filter out the node pairs that may have a causal relationship according to the time characteristics as the candidate set for causal reasoning. Specifically: extract the timestamp t and its context time characteristics (such as duration, repetition pattern) for each event node, and filter to obtain candidate pairs according to the causal candidate set filtering rules. The causal candidate set filtering rules are specifically: filter the event node pairs (e1, e2) whose time sequence relationship satisfies the following conditions, t1 < t2, where t1 and t2 are the timestamps of e1 and e2 respectively, and the time interval satisfies the constraint condition Δt = t2 - t1 ≤ θ t , where θ t is the allowable time interval threshold, which is set by domain requirements or determined by experiments. The node pairs (e1, e2) that meet the time conditions are stored as candidate pairs for causal reasoning for subsequent steps to process.

[0067] In step 402, if the interaction intensity between the first event node and the second event node is greater than the preset interaction intensity, it is inferred that there is a causal relationship between the first event node and the second event node; among them, the event node with the earlier time characteristic is the cause, and the event node with the later time characteristic is the result. The preset interaction intensity is obtained by those skilled in the art through empirical analysis. It can be understood that the first event node and the second event node with a sequential relationship in time characteristics can be considered as two event nodes that may have a causal relationship. Further judgment is made according to the interaction intensity. If the interaction intensity is large, it is considered that the relevance between the two event nodes is large, and thus it can be considered that the two have a causal relationship.

[0068] The above steps 401 - step 402 can be understood as: based on the interaction intensity IS, the following rules are used for causal relationship reasoning: First, according to the time characteristics of the event nodes, find the first event node and the second event node with a sequential relationship in time. If the interaction intensity IS between the first event node and the second event node is greater than the set parameter, it is inferred that there is a causal relationship between the two, where the event node with the earlier time is the "cause"; the event node with the later time is the "result". Its goal is to filter out the event node pairs with higher causal relevance through the quantitative calculation of the interaction intensity IS and clarify the causal direction (that is, the cause node and the result node). The time characteristic filtering provides a candidate set for causal reasoning, but the causal correlation cannot be accurately judged only by the time sequence relationship. Therefore, step 402 comprehensively considers the time constraint and multi-dimensional characteristics and calculates the interaction intensity between the nodes. The calculation of the interaction intensity synthesizes the feature matching degree FS and the initial edge weight EWori and the dynamic changes of the characteristic coupling strength CS.

[0069] In an actual application scenario, the calculation of the feature coupling strength between different features specifically includes:

[0070] Calculate the characteristic coupling strength between the first feature and the second feature Where M is the number of event nodes, w 1，2 is the preset coupling weight corresponding to the first feature and the second feature, f 1,i is the first feature of the i-th event node, f 2,i is the second feature of the i-th event node, sim(f 1,i f 2,i ) is used to calculate f i,1 With f j,2 The preset coupling weight is obtained by those skilled in the art based on empirical analysis.

[0071] The feature coupling strength captures the dynamic law of different features changing over time through the sliding window sampling method, and affects the interaction strength through the weight adjustment mechanism. The sliding window size is set according to the data characteristics and time interval. When the sliding window sampling strength is greater than the set threshold, it is considered that the feature association is enhanced. This process dynamically reflects the evolution law of multidimensional features and provides more accurate input for the calculation of interaction strength. In order to capture the law of feature development over time, the sliding window sampling method is used to perform segmented analysis on event nodes. During the sliding window sampling process, the sampling feature coupling strength in each time window is calculated and its change trend is recorded. When the sampling feature coupling strength exceeds the set threshold, it is considered that the association between features is significantly enhanced. The sliding window size and frequency are set according to the time distribution characteristics of the data to ensure the accuracy of capturing dynamic changes. By combining the dynamic change law of feature coupling strength and the quantitative calculation of interaction strength, the model can comprehensively evaluate the causal correlation between nodes. Finally, this process not only provides multi-dimensional support for the reasoning of causal relationships, but also enhances the adaptability and robustness of the system in complex scenarios.

[0072] When each event node is an event node that records the development of the same event over time, such as the event node corresponding to the change in store traffic, at this time, it can be understood that each event node has been arranged in chronological order. The development law of the coupling strength of each feature is analyzed, such as Figure 4 As shown, specifically including:

[0073] In step 501, when the feature coupling strength between the first feature and the second feature is greater than the preset coupling strength, sliding window sampling is performed on the event node with a preset number as the sliding window size, and the sampling feature coupling strength corresponding to each sliding window sampling is calculated; wherein the sampling feature coupling strength is the feature coupling strength between the first feature and the second feature calculated using the event node of the sliding window sampling.

[0074] That is, the feature coupling strength of the event node is dynamically analyzed based on the sliding window sampling method. Specifically, the sliding window size Δt is preset according to the dynamic change frequency of the event node to ensure that key feature changes are captured. The result of each sliding window sampling is the quantized value of the feature strength in the current time window, combined with the time decay factor DecayFactor (Δt) = exp (-λ × Δt), to form a comprehensive analysis of the change of feature strength over time. Among them, the sampled feature coupling strength is the feature coupling strength between the first feature and the second feature calculated by the event node using sliding window sampling; the preset number is obtained by technical personnel in this field based on empirical analysis. Sliding window sampling is used here to reduce the amount of calculation and reduce the interference of noise on the feature coupling strength analysis.

[0075] In step 502, the development law of the characteristic coupling strength over time is analyzed according to the sampling characteristic coupling strength corresponding to multiple sliding window samplings.

[0076] The analysis of the development law of the characteristic coupling strength over time according to the sampling characteristic coupling strength corresponding to the multiple sliding window samplings specifically includes:

[0077] Using the sampled characteristic coupling strength at time t and the sampled characteristic coupling strength at time t-1, combined with the first formula, a time decay function of the characteristic coupling strength that decays with time is derived; wherein the first formula is CS(t)=β×DecayFactor(Δt)×CS(t-1).

[0078] The time decay function is DecayFactor(Δt)=exp(-λ×Δt); wherein β is a preset coefficient, CS(t) is the sampling characteristic coupling strength at time t, CS(t-1) is the sampling characteristic coupling strength at time t-1, and Δt is the time interval of sliding window sampling.

[0079] By constructing time series data from the feature coupling strength of multiple sliding window samplings, we analyze its changing pattern over time. The first formula CS(t) = β × DecayFactor (Δt) × CS(t-1) is used to dynamically balance the historical observations and the current feature strength, and deduce the development trend of the feature strength. The change in feature strength within the sliding window will serve as the key basis for causal reasoning to determine the causal correlation between event nodes. Ultimately, this process combines sliding window sampling with the time decay model, which not only improves the sensitivity to short-term changes, but also enhances the robustness of the system in long-term dynamic evolution scenarios.

[0080] The time decay function is DecayFactor(Δt)=exp(-λ×Δt); where β is a preset coefficient, CS(t) represents the sliding window sampling feature coupling strength at time t, as the feature quantization result of time decay. CS(t-1) represents the sliding window sampling feature coupling strength at time t-1, which is used to balance the impact of historical observations on current reasoning, and Δt is the time interval of sliding window sampling.

[0081] The time decay function describes the law of the change of the characteristic coupling strength over time, and can quantify the change trend of the characteristic correlation degree over time. That is, the time decay function is the law of the change of the corresponding characteristic coupling strength over time, so it can reflect the change of the correlation degree between two characteristics over time.

[0082] On the contrary, according to the development law of the characteristic coupling strength, combined with the time decay function and the sampling characteristic coupling strength at the current moment, the sampling characteristic coupling strength at the next moment is derived as the characteristic causal evolution result of the event node. This method predicts the characteristic coupling strength at future moments through dynamic analysis, providing reliable time evolution support for causal reasoning. That is, the characteristic causal evolution reasoning of the event node is carried out according to the development law of each characteristic coupling strength, specifically including: according to the time decay function, the first formula and the sampling characteristic coupling strength at the current moment, the sampling characteristic coupling strength at the next moment is calculated, which is the characteristic causal evolution result of the event node, that is, according to the characteristic coupling strength at the current moment, combined with the development law, the characteristic coupling strength at a certain moment in the future can be calculated.

[0083] Embodiment 2:

[0084] The present invention is based on the method described in Example 1, combined with specific application scenarios, and uses technical descriptions in related scenarios to illustrate the implementation process of the present invention in characteristic scenarios.

[0085] First, this embodiment provides a multi-level feature organization structure (i.e., the preset model in Embodiment 1), which realizes the deep fusion of multi-dimensional features of causal events by constructing composite nodes and multiple associated edges in the graph database. Specifically:

[0086] G=(V,E,L,M)

[0087] Where: V: causal event node set

[0088] E: Causal edge set

[0089] L: a set of feature layers, such as {L_time, L_space, L_actor, L_intensity, L_context}

[0090] M: Inter-layer mapping relationship matrix

[0091] At the node modeling level, each causal event node is designed as a multi-layer structure. The core layer carries basic event information, such as event ID, type and description; the temporal layer describes the temporal characteristics of the event, which includes not only the specific time of occurrence, but also the duration of the event, repetition pattern and other temporal laws; the spatial layer describes the geographical distribution characteristics of the event, including the location of occurrence, scope of influence, spatial diffusion pattern, etc.; the subject layer records the entity information involved in the event, including entity type, attributes, interaction relationship, etc.; the intensity layer quantifies the impact of the event, including direct impact intensity, attenuation characteristics, etc.; the context layer stores background information such as environmental conditions and triggering factors.

[0092] At the edge relationship modeling level, causal relationships are no longer simple one-way connections, but include multiple dimensions of association descriptions: temporal association: describing the temporal dependency between events, such as sequence, time interval, etc.; spatial association: describing the geographical location relationship between events, such as spatial distance, direction, etc.; subject association: expressing the interaction mode between participating entities, such as direct effect, indirect effect, etc.; intensity association: quantifying the transmission characteristics of causal effects, such as attenuation laws, threshold conditions, etc.; context association: recording the transmission effects of environmental factors, such as common triggering conditions, environmental constraints, etc. In actual storage, such as Figure 5 As shown, the feature layer set is stored in the extended attributes of the event node.

[0093] This multi-level feature organization method brings two key advantages: 1. It provides a more complete expression of causal events, overcoming the problems of scattered feature descriptions and weak correlation in traditional methods; 2. It provides rich feature support for subsequent causal reasoning, enabling the system to analyze and judge causal relationships from multiple dimensions.

[0094] After extracting event nodes, relationship edges, and various features of event nodes, they are stored in the graph database. In the graph database, the multi-dimensional features of causal events are stored through a custom node attribute structure. Each event node contains multi-level feature attributes. The following is an example of creating an event node using the Cypher language:

[0095]

[0096]

[0097] Causal relationships are represented by edges with multi-dimensional attributes. Here is an example of using Cypher to create causal relationships:

[0098]

[0099] In addition, the following values ​​are calculated and also stored in the graph database to form a knowledge graph, so that the corresponding values ​​can be read from the graph database during subsequent causal reasoning.

[0100] The values ​​calculated include:

[0101] 1. Feature coupling strength, through which a multi-layer coupling network structure can be used to express the multi-dimensional characteristics of causal events and their interactive relationships:

[0102]

[0103] Where M is the number of event nodes, w 1，2 is the preset coupling weight corresponding to the first feature and the second feature, f 1,i is the first feature of the i-th event node, f 2,i is the second feature of the i-th event node, sim(f 1,i f 2,i ) is used to calculate f i,1 With f j,2 The similarity between them can be obtained by Figure 6 The algorithm implementation shown is as follows: The function receives a single dimension as an input parameter and calculates the interaction effect by traversing all other dimensions. For each different dimension, the interaction parameters with the current dimension are obtained from the predefined DimensionInteractionMatrix, including weights and influence coefficients. The interaction scores of all dimensions are accumulated, and finally normalized through NormalizeInteraction to output the overall synergy score of the dimension and other dimensions.

[0104] 2. Interaction intensity between different event nodes:

[0105]

[0106] IS=k1FS+k2EW ori

[0107] Among them, N is the number of dimensions of the feature, w ijis the preset interaction weight between the features of the i-th dimension and the features of the j-th dimension, f i,1 is the feature of the first event node in the i-th dimension, f j,2 is the feature of the second event node in the jth dimension, sim(f i,1 f j,2 ) is used to calculate f i,1 With f j,2 The similarity between them, k1 and k2 are preset coefficients, EW ori is the initial edge weight corresponding to the relationship edge between the first event node and the second event node.

[0108] Its use is as Figure 7 The algorithm implementation of the framework shown in the figure receives two event nodes as input. First, the weight coefficient of each dimension is initialized through InitializeDimensionWeights. Then, each dimension feature is traversed and calculated: CalculateFeatureMatch is called to calculate the feature matching degree, and CalculateDimensionInteraction is used to evaluate the interaction score between dimensions. The two are multiplied by the weight to obtain the comprehensive score of the dimension. Then, the topological relationship strength between events is evaluated and the feature interaction impact is calculated. Finally, the CombineScores function is used to perform a weighted combination of the scores and normalize them to obtain the final causal strength score.

[0109] The algorithm mainly focuses on effectively integrating multi-dimensional feature information into the causal graph reasoning process, breaking through the limitation of traditional causal reasoning that only relies on the topological relationship between nodes. By constructing a multi-level reasoning mechanism, the algorithm can simultaneously consider the feature matching degree, correlation strength and interaction between dimensions of events, thereby achieving a more comprehensive and in-depth causal relationship analysis.

[0110] The core advantage of this algorithm is that it establishes a comprehensive reasoning framework that organically integrates the topological relationship of the graph structure with the multi-dimensional information of event features. Through the collaborative analysis mechanism of feature dimensions, it not only significantly improves the accuracy and reliability of causal relationship discovery, but also enhances the interpretability of reasoning results, providing more scientific technical support for the identification and verification of causal relationships.

[0111] This embodiment also proposes an adaptive causal strength measurement method for the dynamic evolution characteristics of causal relationships in real-life scenarios. This method captures the changing pattern of causal event characteristics in real time, dynamically evaluates and updates the strength of causal relationships, and thus accurately describes the evolution process of causal relationships. There are two implementation methods for this process:

[0112] The first implementation method is to calculate the time decay function DecayFactor(Δt)=exp(-λ×Δt), and build a dynamic evaluation model of causal strength based on the time decay function and state transition probability:

[0113] CS(t)=β×DecayFactor(Δt)×CS(t-1), where β is the preset coefficient, CS(t) is the sampling characteristic coupling strength at time t, CS(t-1) is the sampling characteristic coupling strength at time t-1, Δ t The time interval for sliding window sampling.

[0114] The dynamic evaluation model dynamically calculates the strength value of the causal relationship by considering both the historical status and the current new information. Specifically, at each time point, the system combines the causal strength (historical information) of the previous moment and the new causal strength (new information) currently observed, and obtains the updated causal strength value by weighted average. The calculation of the new causal strength involves a comprehensive evaluation of multiple dimensions such as time series, space, subject association, influence, and context. At the same time, the system also takes into account the time decay effect, that is, the influence of historical information will gradually weaken over time. This dynamic evaluation mechanism can accurately capture the real-time changes in the strength of causal relationships, and is particularly suitable for complex scenarios where causal relationships change dynamically with time, environment and other factors, such as disease transmission, financial risk transmission and other fields.

[0115] The second implementation method is to describe the evolution of causal relationships by constructing a state transition matrix:

[0116] By constructing a state transition matrix and a transition probability formula, the system solves the limitations of static causal modeling and provides dynamic update capabilities. This design ensures that the causal model can simultaneously maintain compliance with historical laws and respond to the impact of environmental changes in a timely manner, thereby achieving an accurate portrayal of the dynamic evolution of causal relationships in complex scenarios. Especially in scenarios with emergencies, complex transmission networks, or frequent dynamic environmental changes, this method can provide greater adaptability and reliability for causal reasoning.

[0117] Traditional causal relationship models are mostly based on static data and cannot effectively capture the dynamic evolution characteristics of relationship strength over time or environmental changes. By introducing the state transfer formula P(t+1)=T(t)×P(t), the system can dynamically update the state of causal relationships as time passes, reflecting the time sensitivity of causal relationships. The state transfer formula indicates that between time t and t+1, the state vector P(t) of the current causal relationship evolves to the state P(t+1) at the next moment through the state transfer matrix T(t).

[0118] Among them, T(t) is the state transfer matrix, which is used to capture the evolution law of causal relationships in the time dimension, and P(t) is the causal relationship state vector at the current moment, which represents the quantitative strength of the causal relationship between all event nodes in the system. P(t+1) can be understood as the probability matrix of whether multiple event nodes occur at time t+1, and P(t) can be understood as the matrix of whether multiple event nodes occur at time t. If there are n event nodes, then e(i)=0 means that event node i has not occurred at time t, and e(i)=1 means that event node i has occurred at time t. T ij (t) is the state transition probability of event node i triggering event node j at time t.

[0119] This embodiment introduces the state transition probability formula, and the model is extended from a static, single causal relationship description to a dynamic, comprehensive causal relationship modeling, providing higher accuracy and timeliness support for causal reasoning.

[0120] State transition probability calculation:

[0121] T ij (t) = θ × H ij +(1-θ)C ij (t)

[0122] Among them, the historical frequency H ij Provide stable statistical laws accumulated over a long period of time to ensure that the modeling has sufficient historical basis and stability. Specifically: H ij is the probability that event i leads to event j in history, time(i->j) is the number of times event i has led to event j in history, and time(i) is the number of times event i has occurred in history; the real-time environment transition probability C ij (t) reflects the current dynamic environmental factors (such as time characteristics, spatial characteristics, subject changes, etc.), so that the causal relationship modeling can respond to environmental changes in real time. θ is the balance parameter. In actual use, C ij (t) can be the feature matching degree between event i and event j, that is, C ij (t) = FS(i, j).

[0123] State transition formula and state transition probability formula: The state transition formula describes the dynamic evolution of the overall system and depends on the state transition matrix T(t). The state transition probability formula defines the composition of the matrix T(t), ensuring that the state transition modeling combines historical laws with real-time dynamic changes.

[0124] This implementation describes the evolution of causal relationships by constructing a state transition matrix, where the matrix elements represent the transition probabilities between different states. When calculating these transition probabilities, the system takes into account the influence of historical statistical data and current environmental conditions. Specifically, each transition probability is a weighted combination of the historical transition frequency and the possibility of transition in the current environment. This enables the system to maintain the inheritance of historical laws while responding promptly to new environmental changes. In this way, the system can effectively track the changing trajectory of the strength of causal relationships and predict their future evolution trends, providing important support for causal analysis in complex dynamic scenarios.

[0125] In actual use, this embodiment also updates the weights of causal edges by evaluating the cumulative effects of changes in multidimensional features. Specifically, the system will continuously monitor the changes in each dimensional feature (such as time, space, subject features, etc.), and evaluate the actual impact of these changes based on the current environmental conditions. For the changes in each feature dimension, the system will assign different weights according to its importance and calculate its impact on the causal relationship. The impact effects of these sub-dimensions will be accumulated, combined with the time decay factor, and ultimately used to update the weight values ​​of the causal edges. This mechanism can promptly reflect the impact of feature changes on the strength of causal relationships, ensuring that the edge weights in the causal graph always reflect the latest causal association status.

[0126] Based on the cumulative effect of multi-dimensional feature changes, the edge weights are dynamically adjusted: Figure 8 The algorithm shown in the figure implements the dynamic adjustment of causal edge weights through three input parameters: edge, features, and time. First, call CalculateFeatureChanges to calculate the change in the feature in the time dimension, and then use EvaluateContextImpact to evaluate the impact of the environment on the weight update. The core part is to loop through each feature change, obtain the corresponding dimension weight (GetFeatureWeight), and calculate the effect (CalculateEffect) in combination with the environmental impact, and accumulate the overall effect. Finally, use the UpdateWeight function to calculate and return the updated edge weight value based on the current weight, cumulative effect, and time decay factor.

[0127] The calculated values ​​are stored in the graph database. When performing causal reasoning, the database is queried using the Cypher language, such as:

[0128] 1. Implement causal relationship matching query based on multi-dimensional features:

[0129] cypher

[0130] MATCH(e1:Event)-[r:CAUSES]->(e2:Event)

[0131] WHERE e1.timeFeatures.timestamp < e2.timeFeatures.timestamp

[0132] WITH e1,e2,r,

[0133] point.distance(e1.spatialFeatures.location,e2.spatialFeatures.location) as spatial_dist,

[0134] duration.between(e1.timeFeatures.timestamp,e2.timeFeatures.timestamp) as temporal_dist

[0135] WHERE spatial_dist <= e1.spatialFeatures.radius

[0136] AND temporal_dist <= duration('P7D')

[0137] RETURN e1,r,e2,

[0138] spatial_dist,temporal_dist,

[0139] gds.alpha.similarity.euclidean(e1.intensityFeatures,e2.intensityFeatures) as intensity_similarity

[0140] 2. Implement path discovery for multi-hop causal relationships:

[0141] cypher

[0142] MATCH path=(start:Event)-[r:CAUSES*1..3]->(end:Event)

[0143] WHERE start.eventId = 'E001'

[0144] WITH path,relationships(path) as rels,

[0145] reduce(s=1.0,r in relationships(path)|s*r.strength)aspath_strength

[0146] WHERE path_strength>=0.5

[0147] RETURN path,path_strength

[0148] ORDER BY path_strength DESC

[0149] 3. Implement causal strength update based on time decay:

[0150] cypher

[0151] MATCH(e1:Event)-[r:CAUSES]->(e2:Event)

[0152] WHERE.last_update <datetime('2024-01-01T00:00:00')

[0153] WITH e1,e2,r,

[0154] r.strength*exp(-0.1*duration.between(r.last_update,datetime()).days)as decayed_strength

[0155] SET r.strength=decayed_strength,

[0156] r.last_update = datetime()

[0157] 4. Record and query the evolution history of causal relationships:

[0158] cypher

[0159] MATCH(e1:Event)-[r:CAUSES]->(e2:Event)

[0160] WHEREr.strength_changed=true

[0161] CREATE(h:HistoricalStrength{

[0162] timestamp:datetime(),

[0163] old_strength:r.previous_strength,

[0164] new_strength:r.strength,

[0165] change_reason:r.change_factors

[0166] })

[0167] CREATE(r)-[:STRENGTH_HISTORY]->(h)

[0168] These implementations show how to implement causal event modeling, reasoning, and dynamic evolution capture of multi-dimensional features in graph databases. By using Cypher, the native query language of graph databases, complex causal analysis and dynamic update operations can be efficiently implemented. Each implementation part takes into account the data storage structure, query efficiency, and real-time requirements, providing reliable technical support for the entire causal event knowledge graph system.

[0169] After testing, it is found that this embodiment has improved multiple indicators compared with the prior art. In terms of knowledge completeness, the improved indicators are as follows:

[0170] Entity coverage: This solution reaches 92% through multi-dimensional feature modeling, which is a significant improvement over the traditional method (75%).

[0171] Relationship coverage: 88% for this solution, compared to 70% for the traditional approach

[0172] Attribute completeness: This solution supports multi-dimensional feature expression, with a completeness of 85%, while traditional methods are only 60%

[0173] In terms of knowledge accuracy, this embodiment improves the following indicators:

[0174] Entity accuracy: 95% (traditional method 85%)

[0175] Relationship accuracy: 90% (traditional method 78%)

[0176] Comprehensive F1 value: 0.86 (traditional method 0.72)

[0177] Knowledge Consistency

[0178] Logical consistency: 0.95

[0179] Timing consistency: 0.92

[0180] Spatial consistency: 0.88

[0181] In terms of reasoning ability evaluation, this embodiment improves the following indicators:

[0182] Using the standard evaluation method based on Link Prediction, we get:

[0183] Prediction Accuracy

[0184] Hit@1: 0.82 (traditional method 0.65)

[0185] Hit@5: 0.89 (traditional method 0.75)

[0186] MRR (Mean Reciprocal Rank): 0.85 (traditional method 0.70)

[0187] In terms of path discovery, this embodiment improves the following indicators:

[0188] Path integrity: 0.88 (traditional method 0.72)

[0189] Path reliability: 0.85 (traditional method 0.68)

[0190] This embodiment solves the shortcomings of existing methods in dealing with complex causal relationships, dynamic updates, and efficient queries. By using a graph database to model multidimensional causal events, combining a causal graph inference algorithm with multidimensional features and a dynamically evolving causal relationship capture method, the problem of insufficient expression of causal event features in the prior art and the difficulty in accurately quantifying the strength of causal relationships are solved. This embodies deep innovation in multidimensional causal relationship modeling. It is suitable for various application scenarios that require causal analysis, such as intelligent decision-making, risk management, etc.

[0191] Embodiment 3:

[0192] like Fig. 9 , which is a schematic diagram of the architecture of the apparatus for constructing a multidimensional causal event knowledge graph according to an embodiment of the present invention. The apparatus for constructing a multidimensional causal event knowledge graph according to this embodiment includes one or more processors 21 and a memory 22. Fig. 9 A processor 21 is taken as an example.

[0193] The processor 21 and the memory 22 may be connected via a bus or other means. Fig. 9 The example of connecting through bus is taken in the following.

[0194] The memory 22, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs and non-volatile computer executable programs, such as the method for constructing a multi-dimensional causal event knowledge graph in Example 1. The processor 21 executes the method for constructing a multi-dimensional causal event knowledge graph by running the non-volatile software programs and instructions stored in the memory 22.

[0195] The memory 22 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 22 may optionally include a memory remotely arranged relative to the processor 21, and these remote memories may be connected to the processor 21 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0196] The program instructions / modules are stored in the memory 22, and when executed by the one or more processors 21, the method for constructing a multi-dimensional causal event knowledge graph in the above-mentioned embodiment 1 is executed.

[0197] It is worth noting that the information interaction, execution process, etc. between the modules and units within the above-mentioned devices and systems are based on the same concept as the processing method embodiment of the present invention. The specific contents can be found in the description of the method embodiment of the present invention and will not be repeated here.

[0198] A person skilled in the art may understand that all or part of the steps in the various methods of the embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0199] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing a multidimensional causal event knowledge graph, characterized in that: include: Extract event nodes, relationship edges, and various features of event nodes from the data; Calculating the interaction strength between different event nodes, and performing causal relationship reasoning between different event nodes based on the interaction strength; Calculate the feature coupling strength between different features and analyze the development law of each feature coupling strength; perform feature causal evolution reasoning of event nodes based on the development law of each feature coupling strength.

2. The method for constructing a multidimensional causal event knowledge graph according to claim 1, characterized in that: The calculation of the interaction strength between different event nodes specifically includes: Calculate the feature matching degree between the first event node and the second event node Calculate the interaction strength between the first event node and the second event node IS = k1FS + k2EW ori ; Where N is the number of feature dimensions, w ij is the preset interaction weight between the features of the i-th dimension and the features of the j-th dimension, f i,1 is the feature of the first event node in the i-th dimension, f j,2 is the feature of the second event node in the jth dimension, sim(f i,1 f j,2 ) is used to calculate f i,1 With f j,2 The similarity between them, k1 and k2 are preset coefficients, EW ori is the initial edge weight corresponding to the relationship edge between the first event node and the second event node.

3. The method for constructing a multidimensional causal event knowledge graph according to claim 1, characterized in that: The reasoning of the causal relationship between different event nodes according to the interaction strength specifically includes: According to the time characteristics of each event node, searching for the first event node and the second event node that have a sequential relationship in terms of the time characteristics; If the interaction strength between the first event node and the second event node is greater than the preset interaction strength, it is inferred that there is a causal relationship between the first event node and the second event node; wherein the event node with an earlier time feature is the cause, and the event node with a later time feature is the result.

4. The method for constructing a multidimensional causal event knowledge graph according to claim 1, characterized in that: The calculating of the characteristic coupling strength between different characteristics specifically includes: Calculate the characteristic coupling strength between the first feature and the second feature Where M is the number of event nodes, w 1，2 is the preset coupling weight corresponding to the first feature and the second feature, f 1,i is the first feature of the i-th event node, f 2,i is the second feature of the i-th event node, sim(f 1,i f 2,i ) is used to calculate f i,1 With f j,2 The similarity between .

5. The method for constructing a multidimensional causal event knowledge graph according to claim 1, characterized in that: When each event node is an event node that records the development of the same event over time, the analysis of the development law of each characteristic coupling strength specifically includes: When the feature coupling strength between the first feature and the second feature is greater than the preset coupling strength, the event node is subjected to sliding window sampling with the preset number as the sliding window size, and the sampling feature coupling strength corresponding to each sliding window sampling is calculated; wherein the sampling feature coupling strength is the feature coupling strength between the first feature and the second feature calculated using the event node of the sliding window sampling; According to the sampling feature coupling strength corresponding to multiple sliding window samplings, the development law of the feature coupling strength over time is analyzed.

6. The method for constructing a multidimensional causal event knowledge graph according to claim 5, characterized in that: The analysis of the development law of the characteristic coupling strength over time according to the sampling characteristic coupling strength corresponding to the multiple sliding window samplings specifically includes: Using the sampling characteristic coupling strength at time t and the sampling characteristic coupling strength at time t-1, combined with the first formula, a time decay function of the characteristic coupling strength decaying with time is derived; Wherein, the first formula is CS(t)=β×DecayFactor(Δt)×CS(t-1); The time decay function is DecayFactor(Δt)=exp(-λ×Δt); wherein β is a preset coefficient, CS(t) is the sampling characteristic coupling strength at time t, CS(t-1) is the sampling characteristic coupling strength at time t-1, and Δt is the time interval of sliding window sampling.

7. The method for constructing a multidimensional causal event knowledge graph according to claim 6, characterized in that: According to the development law of each feature coupling strength, the feature causal evolution reasoning of the event node is performed, which specifically includes: According to the time attenuation function, the first formula and the sampling feature coupling strength at the current moment, the sampling feature coupling strength at the next moment is calculated, which is the feature causal evolution result of the event node.

8. The method for constructing a multidimensional causal event knowledge graph according to any one of claims 1 to 7, characterized in that: The extraction of event nodes, relationship edges and multiple features of event nodes from the data is achieved by using a preset model for data extraction, and the preset model is G=(V, E, L, M); wherein V is the event node, E is the relationship edge, L is the original feature set, and M is the mapping relationship matrix, which is used to normalize each original feature in the original feature set to obtain multiple features of the event node.

9. A non-volatile computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, which are executed by one or more processors to complete the method for constructing a multidimensional causal event knowledge graph as described in any one of claims 1-8.

10. A device for constructing a multi-dimensional causal event knowledge graph, characterized in that: include: at least one processor; And, a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the processor to execute the method for constructing a multidimensional causal event knowledge graph as described in any one of claims 1-8.

Images