Key influence message identification method and device, equipment and storage medium

Abstract

The application discloses a key influence message identification method and device, equipment and a storage medium, and relates to the technical field of social media information processing. The method comprises the following steps: acquiring interaction messages in a target event and response relationships between the messages, constructing a message propagation graph based on the interaction messages and the response relationships; calculating a group stance distribution based on message nodes in the message propagation graph; determining a target message node from the message propagation graph, performing an anti-fact pruning operation on the target message node to obtain an anti-fact pruning graph, and calculating an anti-fact group stance distribution based on the anti-fact pruning graph; obtaining a causal attribution score of the target message node according to the group stance distribution and the anti-fact group stance distribution, and identifying a key influence message according to the causal attribution score. The application can improve the accuracy and reliability of key influence message identification.

Description

Technical Field

[0001] This application relates to the field of social media information processing technology, and in particular to methods, apparatus, devices and storage media for identifying key impact messages. Background Technology

[0002] With the widespread adoption of internet social media platforms, Weibo, Twitter, and other social media have become core arenas for the public to express their opinions and participate in discussions on public issues. As the basic unit of information dissemination, comments do not have an evenly distributed influence; a comment at a bottleneck in the dissemination process or a reply from an opinion leader can have a decisive impact on the direction of public opinion. In comment influence analysis, accurately identifying key comments with irreplaceable causal effects is a crucial foundation for improving the accuracy of public opinion trend assessment. However, existing methods cannot effectively distinguish between the true influence of comments and false exposure, leading to biases in key comment identification and consequently affecting the accuracy of public opinion trend assessment and the credibility of public opinion intervention decisions.

[0003] Currently, methods are typically used to assess the influence of comments based on statistical indicators (such as the number of reposts, likes, and followers), network structure, or content characteristics (such as semantic analysis and sentiment analysis). However, these methods are essentially descriptive statistics or correlation analysis, which can only identify "who is at the center of the network" and cannot answer counterfactual causal questions such as "how would public opinion change if this comment had never appeared?" They are difficult to distinguish between statistical correlation and real causal effects, and therefore cannot accurately locate key messages with irreplaceable causal roles.

[0004] Therefore, how to accurately identify key messages with irreplaceable causal effects is a technical problem that urgently needs to be solved.

[0005] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main objective of this application is to provide a method, apparatus, device, and storage medium for identifying key impact messages, aiming to solve the technical problem of how to accurately identify key impact messages with irreplaceable causal effects.

[0007] To achieve the above objectives, this application proposes a method for identifying key impact messages, the method comprising: Obtain the interactive messages and response relationships between messages in the target event, and construct a message propagation graph based on the interactive messages and response relationships; The group's position distribution is calculated based on the message nodes in the message propagation graph. The target message node is determined from the message propagation graph, and a counterfactual pruning operation is performed on the target message node to obtain a counterfactual pruning graph. The counterfactual group position distribution is calculated based on the counterfactual pruning graph. Based on the group position distribution and the counterfactual group position distribution, the causal attribution score of the target message node is obtained, and key influencing messages are identified based on the causal attribution score.

[0008] In one embodiment, the step of performing counterfactual pruning on the target message node to obtain a counterfactual pruning graph, and calculating the counterfactual group position distribution based on the counterfactual pruning graph includes: The counterfactual pruning graph is constructed by removing the target message node and the set of downstream nodes that have a response relationship with the target message node from the message propagation graph. Obtain the position labels of the remaining message nodes in the counterfactual pruning graph, perform one-hot encoding transformation on the position labels of the remaining message nodes, and count the remaining cumulative number corresponding to each position category; The remaining cumulative quantity corresponding to each position category is normalized based on the total number of remaining message nodes to obtain the counterfactual group position distribution.

[0009] In one embodiment, the step of removing the target message node and the set of downstream nodes that have a response relationship with the target message node from the message propagation graph includes: Starting from the target message node, perform a graph traversal search along the response relationships in the message propagation graph to locate the set of all downstream nodes that have a direct or indirect response relationship with the target message node; Remove the target message node and all nodes in the located downstream node set from the message propagation graph.

[0010] In one embodiment, the step of calculating the group position distribution based on the message nodes in the message propagation graph includes: Obtain the preset position tags of each message node in the message propagation graph, wherein the position tags include three positions: opposition, neutrality, or support. Perform one-hot encoding conversion on the position tags of each message node, and count the cumulative number corresponding to each position category; The cumulative number corresponding to each position category is normalized based on the total number of each message node to obtain the group position distribution.

[0011] In one embodiment, the step of obtaining the causal attribution score of the target message node based on the group position distribution and the counterfactual group position distribution includes: Calculate the distribution difference between the group's position distribution and the counterfactual group's position distribution; The distribution difference is subjected to numerical compression transformation, and the transformed value is normalized to obtain the causal attribution score of the target message node.

[0012] In one embodiment, the step of identifying key impact messages based on the causal attribution score includes: The causal attribution score is compared with a preset threshold, and the corresponding message node is determined as a key impact message based on the comparison result. The preset threshold is determined based on the ranking result of the causal attribution scores of all message nodes, or selected through cross-validation.

[0013] In one embodiment, the step of constructing a message propagation graph based on the interaction messages and response relationships includes: Each message in the interactive message is mapped to a message node; Based on the response relationship, the pointing relationship between message nodes is determined, and a directed propagation edge connecting the message nodes is constructed; The message propagation graph is constructed based on the message nodes and the directed propagation edges.

[0014] Furthermore, to achieve the above objectives, this application also proposes a critical impact message identification device, which includes: The graphing module is used to obtain the interaction messages and response relationships between messages in the target event, and to construct a message propagation graph based on the interaction messages and response relationships; The distribution module is used to calculate the group position distribution based on the message nodes in the message propagation graph; The pruning module is used to determine the target message node from the message propagation graph, perform counterfactual pruning operation on the target message node to obtain the counterfactual pruning graph, and calculate the counterfactual group position distribution based on the counterfactual pruning graph. The attribution module is used to obtain the causal attribution score of the target message node based on the group position distribution and the counterfactual group position distribution, and to identify key influencing messages based on the causal attribution score.

[0015] In addition, to achieve the above objectives, this application also proposes a critical impact message identification device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the critical impact message identification method as described above.

[0016] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the key impact message recognition method described above.

[0017] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the key impact message recognition method described above.

[0018] This application provides a method for identifying key impact messages. First, it constructs a message propagation graph by acquiring the interactive messages and response relationships between messages in a target event. Based on the message nodes in the propagation graph, it calculates the group position distribution to reflect the overall trend of public opinion. Then, it identifies a target message node from the propagation graph and performs counterfactual pruning on it, removing the target message node and all its downstream nodes to obtain a counterfactual pruning graph. Based on this graph, it calculates a counterfactual group position distribution to reflect the public opinion trend when the target message node had never appeared. Finally, based on the difference between the group position distribution and the counterfactual group position distribution, it obtains a causal attribution score for the target message node and identifies key impact messages based on this score, quantifying the strength of the true causal effect of the target message node on the direction of public opinion.

[0019] In summary, this application, by combining structured construction of the message propagation graph, counterfactual pruning operations, and measurement of differences in group position distribution, achieves a leap from "identifying the central node of the network" to "accurately identifying key impact messages with irreplaceable causal effects" in a quantifiable and reproducible manner. This overcomes the technical shortcomings of existing technologies, which, due to the lack of a causal inference perspective, can only identify the central node of the network and cannot answer the counterfactual causal questions of comment influence, thus improving the accuracy and credibility of key impact message identification. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating an embodiment of the key impact message identification method of this application. Figure 2 A simplified flowchart illustrating the group position distribution calculation method provided in Embodiment 1 of this application; Figure 3 This is a flowchart illustrating Embodiment 2 of the method for identifying key impact messages in this application. Figure 4 This is a schematic diagram of the module structure of the key impact message recognition device in the embodiments of this application; Figure 5 This is a schematic diagram of the device structure of the hardware operating environment involved in the key impact message identification method in the embodiments of this application.

[0023] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0024] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0025] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0026] The main solution of this application embodiment is as follows: obtain the interaction messages and response relationships between messages in the target event, construct a message propagation graph based on the interaction messages and response relationships; calculate the group position distribution based on the message nodes in the message propagation graph; determine the target message node from the message propagation graph, perform counterfactual pruning on the target message node to obtain a counterfactual pruning graph, and calculate the counterfactual group position distribution based on the counterfactual pruning graph; obtain the causal attribution score of the target message node according to the group position distribution and the counterfactual group position distribution, and identify key influencing messages according to the causal attribution score.

[0027] Currently, methods based on statistical indicators (such as the number of reposts, likes, and followers), network structure, or content characteristics (such as semantic analysis and sentiment analysis) are commonly used to assess the influence of comments. However, these methods are essentially descriptive statistics or correlation analysis, which can only identify "who is at the center of the network" and cannot answer counterfactual causal questions such as "how would public opinion change if this comment had never appeared?" They also struggle to distinguish between statistical correlation and real causal effects, thus failing to accurately pinpoint key messages with irreplaceable causal roles. Therefore, accurately identifying key messages with irreplaceable causal roles is a pressing technical problem that needs to be solved.

[0028] This application, by combining structured construction of message propagation graphs, counterfactual pruning operations, and measurement of differences in group position distribution, achieves a leap from "identifying network center nodes" to "accurately identifying key impact messages with irreplaceable causal effects" in a quantifiable and reproducible manner. It overcomes the technical shortcomings of existing technologies that, due to the lack of a causal inference perspective, can only identify network center nodes and cannot answer counterfactual causal questions about the influence of comments, thereby improving the accuracy and credibility of key impact message identification.

[0029] It should be noted that the executing entity in this embodiment can be a critical impact message identification system, or a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a thermal management device capable of implementing the above-mentioned functions of a critical impact message identification system. This embodiment does not specifically limit it in this way. The following uses a critical impact message identification system as the executing entity as an example to describe this embodiment and the following embodiments.

[0030] Based on this, embodiments of this application provide a method for identifying key impact messages, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the key impact message identification method of this application.

[0031] In this embodiment, the key impact message identification method includes steps S10 to S40: Step S10: Obtain the interaction messages and response relationships between messages in the target event, and construct a message propagation graph based on the interaction messages and response relationships; It should be noted that interactive messages refer to comments, posts, or replies published by users on social media platforms, including text content, publisher identifiers, and posting timestamps. Response relationships refer to the propagation connections between comments, including reply relationships and forwarding relationships. The message propagation graph is derived from a complete heterogeneous graph. ,in For a set of nodes, Represents a user node. Indicates a comment node; For the set of edges that respond; Forwarding edge set; For the set of edges to focus on.

[0032] Additionally, it should be noted that the message propagation graph uses comments as nodes and the reply and forwarding relationships between comments as directed edges. The graph structure is simple and computationally efficient. Each directed edge carries an edge type identifier to distinguish between reply and forwarding relationships.

[0033] Understandably, by constructing a message propagation graph, the discrete interactive messages and their propagation relationships in the target event are transformed into structured graph data, providing a data foundation for subsequent group position distribution calculations and counterfactual pruning operations based on graph structures.

[0034] In one feasible implementation, step S10 may include steps A11 to A13: Step A11: Map each message in the interaction message to a message node; It should be noted that a message node is the basic unit in the message propagation graph, and each interactive message uniquely corresponds to one message node. The node identifier is consistent with the unique identifier of the message. Message node It includes the text content of the message, the timestamp of publication, the publisher's identifier, and other attribute information.

[0035] Additionally, it should be noted that the mapping operation transforms each interaction message into a vertex in the graph structure, the message node. Each node corresponds to a specific interactive message, and the number of nodes equals the total number of interactive messages, Ncomment.

[0036] Step A12: Determine the pointing relationship between message nodes based on the response relationship, and construct a directed propagation edge connecting the message nodes; It should be noted that the pointing relationship refers to the direction of message propagation. For the reply relationship, the edge... Comment Forwarded comment The direction of transmission is For forwarding relationships, edges Comment Forwarded comment The direction of transmission is This reflects the actual flow of information from its source to its subsequent stages.

[0037] Additionally, it should be noted that the direction of directed propagation edges strictly follows the chronological order, that is, for any directed edge... Meeting the release time t j <t i This ensures that the direction of propagation is consistent with the time logic. Each directed propagation edge carries an edge type identifier to distinguish between two different response types: reply and forwarding relationships.

[0038] Step A13: Construct the message propagation graph based on the message nodes and the directed propagation edges; It should be noted that the message propagation graph has the characteristics of a directed acyclic graph, that is, the propagation direction between messages strictly follows the chronological order and there is no circular dependency. This characteristic ensures the topological rationality of removing message nodes and their downstream propagation chains in subsequent counterfactual pruning operations.

[0039] Step S20: Calculate the group position distribution based on the message nodes in the message propagation graph; It should be noted that the distribution of group stances refers to the statistical distribution of stance labels among all message nodes in the message propagation graph across the three categories of support, neutrality, and opposition, used to characterize the overall trend of current public opinion. Stance Labels These represent opposing, neutral, and supporting stances, respectively, derived from the pre-labeled `stance` field in the dataset. The group stance distribution is obtained by statistically normalizing the stance labels of all message nodes after one-hot encoding.

[0040] Additionally, it should be noted that the formula for calculating the distribution of group positions is:

[0041] in This is the one-hot encoded vector for the position label; Total number of comment nodes; Let be the position distribution vector on the probabilistic simplex, satisfying that the sum of its components is... .

[0042] Understandably, by aggregating discrete stance labels into a group stance distribution, the public opinion trend is quantified into a computable vector form, providing a comparable benchmark for subsequent counterfactual pruning operations.

[0043] In one feasible implementation, please refer to Figure 2 , Figure 2 This is a simplified flowchart of the group position distribution calculation method provided in Embodiment 1 of this application; step S20 may include steps A21 to A23: Step A21: Obtain the preset position labels of each message node in the message propagation graph, wherein the position labels include three positions: opposition, neutrality, or support. It should be noted that the stance label These represent opposing, neutral, and supporting stances, respectively. For message nodes without stance labels, the stance can be inferred from the message text using a pre-trained stance classifier.

[0044] Additionally, it should be noted that the acquisition of position tags is completed in advance during the offline phase and does not consume computing resources for online inference.

[0045] Step A22: Perform one-hot encoding conversion on the position tags of each message node, and count the cumulative number corresponding to each position category; It should be noted that one-hot encoding conversion will change the position label. Convert to three-dimensional one-heat vector Among them, the opposition 1 is mapped to [1,0,0], neutral 0 is mapped to [0,1,0], and support +1 is mapped to [0,0,1]. The one-hot vectors of each message node are accumulated along each dimension to obtain the cumulative number corresponding to each stance category.

[0046] Additionally, it should be noted that the cumulative statistics can be efficiently completed under a large-scale parallel computing framework, making it suitable for large-scale public opinion events containing tens of thousands of messages.

[0047] Step A23: Normalize the cumulative number corresponding to each position category based on the total number of each message node to obtain the group position distribution; It should be noted that the normalization process involves dividing the cumulative number of each position category by the total number of message nodes. The population position distribution vector on the probability simplex is obtained as follows:

[0048] in, ∈[0,1] 3 The three components represent the proportions of opposing, neutral, and supporting positions in the overall message, and the sum of the components is 1.

[0049] Additionally, it should be noted that the distribution of group positions, as a quantitative representation of the current overall public opinion situation, will be used as a benchmark for measuring the causal effect of individual messages in subsequent steps.

[0050] Step S30: Determine the target message node from the message propagation graph, perform counterfactual pruning on the target message node to obtain a counterfactual pruning graph, and calculate the counterfactual group position distribution based on the counterfactual pruning graph; It should be noted that the target message node is a message node selected from the message propagation graph to be evaluated. Counterfactual pruning refers to removing the target message node from the message propagation graph. And all its downstream nodes, simulating a counterfactual scenario where "this message node never appeared," is equivalent to the do-operator intervention in Pearl's causal inference framework. Counterfactual pruning diagram It is the graph structure remaining after removing the target message node and all its downstream nodes. Counterfactual group position distribution. It is the statistical distribution of the position labels of the remaining message nodes in the counterfactual pruning diagram across the three categories of support, neutrality, and opposition, used to characterize the public opinion trend under the condition of "assuming that the message node never appeared".

[0051] Additionally, it should be noted that counterfactual pruning operations... middle, For comments The depth-reachable set contains all direct or indirect responses. The comments and all comments spread through the chain of reposts. The counterfactual group's position distribution. The calculation formula is:

[0052] Understandably, since the construction of counterfactual scenarios relies on intervention operations on the graph structure, this step explicitly models the removal effect of the target message node as a change in the graph structure through structured counterfactual pruning operations, and recalculates the position distribution based on the remaining nodes after pruning, thereby constructing a counterfactual benchmark of "this message node never appeared", providing a comparable counterfactual reference for subsequently quantifying the true causal effect of this message node on the trend of public opinion.

[0053] Step S40: Based on the group position distribution and the counterfactual group position distribution, obtain the causal attribution score of the target message node, and identify key influencing messages based on the causal attribution score; It should be noted that the group stance distribution represents the overall trend of current public opinion, while the counterfactual group stance distribution represents the counterfactual public opinion trend under the condition that "the target message node never appeared." The difference between the two reflects the actual impact of removing the target message node on the direction of public opinion. Causal attribution score is a quantitative indicator that measures the strength of the causal contribution of the target message node to the direction of public opinion; a higher score indicates a more significant impact of the message node on public opinion. Key impact messages refer to message nodes that have an irreplaceable causal role and a decisive influence on the direction of public opinion.

[0054] Additionally, it should be noted that the causal attribution score is obtained by numerically compressing and normalizing the distribution difference, facilitating comparisons between different events or message nodes. The preset threshold is used to convert continuous scores into discrete recognition results.

[0055] Understandably, by calculating the difference between the distribution of group positions and the distribution of counterfactual group positions, the impact of a single message on the direction of public opinion can be quantified into a calculable causal attribution score, thereby elevating influence analysis from the level of "statistical correlation" to the level of "causal inference," and making the identification of key influencing messages have a quantifiable basis for judgment.

[0056] In one feasible implementation, step S40 may include steps A41 to A43: Step A41: Calculate the distribution difference between the group position distribution and the counterfactual group position distribution; It should be noted that the distribution difference is used to quantify the distribution of the group's position. Distribution of positions of the aforementioned counterfactual groups The degree of difference between them. In this embodiment, Kullback-Leibler divergence (KL divergence for short) is used as a measure of difference to calculate the position distribution of the counterfactual group. Relative to the group's position distribution The KL divergence, used as the original causal attribution score, indicates that the removal of the target message node has a more significant impact on the distribution of group positions, meaning the causal effect of that message node is stronger. The original causal attribution score is... The calculation formula is:

[0057] The agreement and (When the latter occurs, it is truncated with a larger constant).

[0058] Additionally, it should be noted that KL divergence is an asymmetric information-theoretic metric used to measure the difference between two probability distributions. In this embodiment, the group's position distribution is used as an example. Based on this, calculate the counterfactual group position distribution. Compared to The greater the deviation, the greater the impact of removing the message node on the direction of public opinion.

[0059] Step A42: Perform numerical compression transformation on the distribution difference quantity, and normalize the transformed value to obtain the causal attribution score of the target message node. It should be noted that the numerical compression transformation is used to logarithmically compress the original causal attribution score to smooth the long-tailed distribution characteristics present in the real data. The normalization process is used to map the compressed value to the [0,1] interval to obtain the causal attribution score of the target message node. Wherein, the causal attribution score... The calculation formula is:

[0060] in This is a logarithmic compression function used to compress the dynamic range of high-scoring regions; the denominator is the maximum compressed value among all comments, ensuring... This facilitates the use of the query-related scores as a supervisory signal for regression tasks or for ranking learning.

[0061] Additionally, it should be noted that the logarithmic compression transformation solves the problem of scores being concentrated in the low-value region due to the long-tailed distribution of the original scores, allowing nodes with most scores close to zero to obtain distinguishable scores. The min-max normalization process makes the causal attribution scores between different message nodes comparable, facilitating subsequent determinations using a unified preset threshold.

[0062] Step A43: Compare the causal attribution score with a preset threshold, and determine whether the corresponding message node is a key impact message based on the comparison result. The preset threshold is determined based on the ranking result of the causal attribution scores of all message nodes, or selected through cross-validation. It should be noted that the preset threshold is denoted as This is used to convert continuous causal attribution scores into binary judgment results. When the causal attribution score of a message node is greater than the preset threshold... At that time, the message node is determined to be a critical message with significant impact.

[0063] Additionally, it should be noted that the preset threshold... The Top-K quantile (e.g., the threshold corresponding to K=10) can be set based on the ranking of causal attribution scores of all message nodes, or the optimal threshold can be selected through cross-validation to balance the accuracy and recall of the identification. The flexibility of the preset threshold allows this method to adapt to the needs of different application scenarios for the granularity of key impact message identification.

[0064] This embodiment provides a method for identifying key impact messages. First, by acquiring the interactive messages and response relationships between messages in a target event, a message propagation graph is constructed. Based on the message nodes in the message propagation graph, the distribution of group positions is calculated to reflect the overall trend of public opinion. Then, a target message node is identified from the message propagation graph, and a counterfactual pruning operation is performed on the target message node to remove the target message node and all its downstream nodes, resulting in a counterfactual pruning graph. Based on the counterfactual pruning graph, a counterfactual group position distribution is calculated to reflect the public opinion trend when the target message node had never appeared. Furthermore, based on the difference between the group position distribution and the counterfactual group position distribution, a causal attribution score for the target message node is obtained. Key impact messages are identified based on this causal attribution score to quantify the strength of the true causal effect of the target message node on the trend of public opinion.

[0065] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3Step S30 also includes steps S31 to S33: Step S31: Remove the target message node and the set of downstream nodes that have a response relationship with the target message node from the message propagation graph to construct a counterfactual pruning graph; It should be noted that the downstream node set refers to all comment nodes that can be reached from the target message node along the response relationships (including reply relationships and forwarding relationships) in the message propagation graph, i.e. All direct or indirect downstream nodes. A counterfactual pruning graph removes the target message node from the original message propagation graph. The remaining graph structure after all its downstream nodes is used to simulate a counterfactual scenario where "the message node never appeared".

[0066] Additionally, it should be noted that counterfactual pruning is equivalent to the do-operator intervention in Pearl's causal inference framework. This means severing the causal path of the message node to all its downstream nodes. (Counterfactual pruning diagram) The expression is:

[0067] in, The difference operation represents the set of nodes. This refers to the set of downstream nodes located in step S311. The set of nodes that were removed.

[0068] Understandably, by removing the target message node and all its downstream nodes, the discussion structure when the message node never appeared was simulated, providing a graph structure basis for subsequent calculation of the counterfactual group's position distribution, thereby enabling the causal effect of the message node on the direction of public opinion.

[0069] In one feasible implementation, step S31 may include steps A311~A312: Step A311: Starting from the target message node, perform a graph traversal search along the response relationship in the message propagation graph to locate the set of all downstream nodes that have a direct or indirect response relationship with the target message node. It should be noted that response relationships include reply relationships and forwarding relationships. The graph traversal search can employ breadth-first search (BFS) or depth-first search (DFS) algorithms, starting from the target message node. Starting from the beginning, traverse all reachable comment nodes along the directed edge to obtain the depth-reachable set. This collection contains all direct and indirect responses. The comment nodes, and all comment nodes that spread through the forwarding chain.

[0070] Additionally, it should be noted that the comments... The depth-reachable set is:

[0071] This set fully characterizes the causal influence domain of the target message node in the propagation structure, that is, all nodes that may be affected. Comment nodes that are directly or indirectly affected.

[0072] Step A312: Remove the target message node and all nodes in the located downstream node set from the message propagation graph. It should be noted that the removal operation will target the message node. and all downstream nodes located in step A311 Removed from the message propagation graph, the remaining set of nodes is The remaining edge set consists of edges in the original edge set whose two endpoints have not been removed.

[0073] Additionally, it should be noted that after the removal operation, the counterfactual pruning diagram... Only comment nodes that have no causal relationship with the target message node are retained to simulate the public opinion discussion structure when the message node has never appeared.

[0074] Step S32: Obtain the position labels of the remaining message nodes in the counterfactual pruning graph, perform one-hot encoding transformation on the position labels of the remaining message nodes, and count the remaining cumulative number corresponding to each position category. It should be noted that the remaining message nodes refer to the counterfactual pruning graph. All comment nodes retained in the original message propagation graph, excluding the target message node and all its downstream nodes. Position tags. These represent opposing, neutral, and supporting stances, derived from the pre-labeled `stance` field in the dataset. The remaining cumulative number refers to the statistical number of remaining message nodes across each stance category.

[0075] Additionally, it should be noted that the remaining cumulative quantity is calculated as follows: one-hot encoding transformation is performed on the position label of each remaining message node in the counterfactual pruning graph. Then, the summation is performed across all dimensions.

[0076] Understandably, by statistically analyzing the distribution of position labels among the remaining message nodes in the counterfactual pruning diagram, we can obtain a quantitative representation of public opinion in terms of position dimension under the condition that "the target message node has never appeared," providing a data foundation for subsequent calculations of the position distribution of counterfactual groups.

[0077] Step S33: Based on the total number of remaining message nodes, normalize the remaining cumulative quantity corresponding to each position category to obtain the counterfactual group position distribution. It should be noted that the distribution of counterfactual group positions This is the statistical distribution of the position labels of the remaining message nodes in the counterfactual pruning diagram across three categories: support, neutral, and opposition. It is used to characterize public opinion under the condition that "the target message node never appeared." The normalization process involves dividing the cumulative number of remaining nodes in each position category by the total number of remaining message nodes. We obtain the counterfactual group position distribution vector on the probability simplex.

[0078] Additionally, it should be noted that the distribution of counterfactual group positions... The calculation formula is:

[0079] in, This represents the total number of remaining message nodes. ∈[0,1] 3 For the remaining message nodes The one-hot encoded vector of the position label, ∈[0,1] 3 Let be the position distribution vector on the probability simplex, and the sum of its components is 1.

[0080] It is understandable that the counterfactual group position distribution obtained through normalization has the same mathematical form as the group position distribution obtained in step S20. The difference between the two can be used to quantify the true causal effect of the target message node on the trend of public opinion, providing a causal inference basis for subsequent identification of key influencing messages.

[0081] This embodiment provides a method for identifying key impact messages, particularly a specific implementation of a counterfactual pruning operation. In this embodiment, firstly, starting from the target message node, a graph traversal search is performed along the response relationships in the message propagation graph to locate the set of all downstream nodes that have direct or indirect response relationships with the target message node. Then, the target message node and all nodes in the located downstream node set are removed from the message propagation graph. This removal operation constructs a counterfactual pruning graph, simulating a counterfactual scenario where "the message node never appeared," providing a graph structure foundation for subsequently quantifying the causal effect of a single message on public opinion trends. Next, the position labels of the remaining message nodes in the counterfactual pruning graph are obtained. One-hot encoding is performed on the position labels of the remaining message nodes, and the remaining cumulative quantity corresponding to each position category is counted. This statistical operation transforms the pruned graph structure information into calculable position statistics. This provides a data foundation for the subsequent calculation of the counterfactual group position distribution. Then, based on the total number of remaining message nodes, the remaining cumulative quantity corresponding to each position category is normalized to obtain the counterfactual group position distribution. This normalization process ensures that the counterfactual group position distribution and the group position distribution obtained in step S20 have the same mathematical form, thus ensuring the mathematical validity and comparability of the KL divergence calculation between the two. Finally, the counterfactual group position distribution and the group position distribution are used together for the subsequent calculation of causal attribution scores and the identification of key impact messages. The difference between the counterfactual group position distribution and the group position distribution quantifies the true causal effect of target message nodes on public opinion trends, thus providing a quantifiable causal inference basis for the accurate identification of key impact messages. Therefore, this embodiment, through the coordination of steps S31 to S33, transforms the abstract counterfactual causal inference problem into a computable and reproducible graph structure pruning and distribution statistics problem, enabling the causal effect of a single message on public opinion trends to be accurately quantified. This overcomes the shortcomings of existing technologies where counterfactual causal inference remains at the theoretical level and is difficult to execute accurately on real data, thus improving the accuracy and reproducibility of identifying key impact messages.

[0082] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the key impact message identification method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0083] This application also provides a key impact message identification device; please refer to [reference needed]. Figure 4 The key impact message identification device includes: The graphing module 10 is used to obtain the interaction messages and response relationships between messages in the target event, and to construct a message propagation graph based on the interaction messages and response relationships; Distribution module 20 is used to calculate the group position distribution based on the message nodes in the message propagation graph; The pruning module 30 is used to determine the target message node from the message propagation graph, perform counterfactual pruning operation on the target message node to obtain the counterfactual pruning graph, and calculate the counterfactual group position distribution based on the counterfactual pruning graph. The attribution module 40 is used to obtain the causal attribution score of the target message node based on the group position distribution and the counterfactual group position distribution, and to identify key influencing messages based on the causal attribution score.

[0084] The critical impact message identification device provided in this application, employing the critical impact message identification method in the above embodiments, can solve the technical problem of how to accurately identify critical impact messages with irreplaceable causal effects. Compared with the prior art, the beneficial effects of the critical impact message identification device provided in this application are the same as those of the critical impact message identification method provided in the above embodiments, and other technical features in the critical impact message identification device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0085] This application provides a critical impact message identification device, which includes: 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 at least one processor to enable the at least one processor to perform the critical impact message identification method in the first embodiment described above.

[0086] The following is for reference. Figure 5 The diagram illustrates a structural schematic of a key impact message recognition device suitable for implementing embodiments of this application. The key impact message recognition device in embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), tablets, PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The key impact message recognition device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0087] like Figure 5As shown, the critical impact message identification device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the critical impact message identification device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the critical impact message identification device to communicate wirelessly or wiredly with other devices to exchange data. While the figures show critical impact message identification devices with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0088] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0089] The critical impact message identification device provided in this application, employing the critical impact message identification method described in the above embodiments, can solve the technical problem of how to accurately identify critical impact messages with irreplaceable causal effects. Compared with the prior art, the beneficial effects of the critical impact message identification device provided in this application are the same as those of the critical impact message identification method described in the above embodiments, and other technical features in this critical impact message identification device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0090] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0091] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0092] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the key impact message recognition method in the above embodiments.

[0093] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0094] The aforementioned computer-readable storage medium may be included in the critical impact message identification device; or it may exist independently and not assembled into the critical impact message identification device.

[0095] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a key impact message identification device, the key impact message identification device: acquires interactive messages and response relationships between messages in a target event; constructs a message propagation graph based on the interactive messages and response relationships; calculates a group position distribution based on message nodes in the message propagation graph; determines a target message node from the message propagation graph; performs counterfactual pruning on the target message node to obtain a counterfactual pruning graph; calculates a counterfactual group position distribution based on the counterfactual pruning graph; obtains a causal attribution score for the target message node based on the group position distribution and the counterfactual group position distribution; and identifies key impact messages based on the causal attribution score.

[0096] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0097] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0098] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0099] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described key impact message identification method, thereby solving the technical problem of how to accurately identify key impact messages with irreplaceable causal effects. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the key impact message identification method provided in the above embodiments, and will not be repeated here.

[0100] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the key impact message recognition method described above.

[0101] The computer program product provided in this application solves the technical problem of how to accurately identify key impact messages with irreplaceable causal effects. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the key impact message identification method provided in the above embodiments, and will not be repeated here.

[0102] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for identifying key impact messages, characterized in that, The method includes: Obtain the interactive messages and response relationships between messages in the target event, and construct a message propagation graph based on the interactive messages and response relationships; The group's position distribution is calculated based on the message nodes in the message propagation graph. The target message node is determined from the message propagation graph, and a counterfactual pruning operation is performed on the target message node to obtain a counterfactual pruning graph. The counterfactual group position distribution is calculated based on the counterfactual pruning graph. Based on the group position distribution and the counterfactual group position distribution, the causal attribution score of the target message node is obtained, and key influencing messages are identified based on the causal attribution score.

2. The method according to claim 1, characterized in that, The steps of performing counterfactual pruning on the target message node to obtain a counterfactual pruning graph, and calculating the counterfactual group position distribution based on the counterfactual pruning graph include: The counterfactual pruning graph is constructed by removing the target message node and the set of downstream nodes that have a response relationship with the target message node from the message propagation graph. Obtain the position labels of the remaining message nodes in the counterfactual pruning graph, perform one-hot encoding transformation on the position labels of the remaining message nodes, and count the remaining cumulative number corresponding to each position category; The remaining cumulative quantity corresponding to each position category is normalized based on the total number of remaining message nodes to obtain the counterfactual group position distribution.

3. The method according to claim 2, characterized in that, The step of removing the target message node and the set of downstream nodes that have a response relationship with the target message node from the message propagation graph includes: Starting from the target message node, perform a graph traversal search along the response relationships in the message propagation graph to locate the set of all downstream nodes that have a direct or indirect response relationship with the target message node; Remove the target message node and all nodes in the located downstream node set from the message propagation graph.

4. The method according to claim 1, characterized in that, The step of calculating the group position distribution based on the message nodes in the message propagation graph includes: Obtain the preset position tags of each message node in the message propagation graph, wherein the position tags include three positions: opposition, neutrality, or support. Perform one-hot encoding conversion on the position tags of each message node, and count the cumulative number corresponding to each position category; The cumulative number corresponding to each position category is normalized based on the total number of each message node to obtain the group position distribution.

5. The method according to claim 1, characterized in that, The step of obtaining the causal attribution score of the target message node based on the group position distribution and the counterfactual group position distribution includes: Calculate the distribution difference between the group's position distribution and the counterfactual group's position distribution; The distribution difference is subjected to numerical compression transformation, and the transformed value is normalized to obtain the causal attribution score of the target message node.

6. The method according to claim 1, characterized in that, The step of identifying key impact messages based on the causal attribution score includes: The causal attribution score is compared with a preset threshold, and the corresponding message node is determined as a key impact message based on the comparison result. The preset threshold is determined based on the ranking result of the causal attribution scores of all message nodes, or selected through cross-validation.

7. The method according to claim 1, characterized in that, The step of constructing a message propagation graph based on the interaction messages and response relationships includes: Each message in the interactive message is mapped to a message node; Based on the response relationship, the pointing relationship between message nodes is determined, and a directed propagation edge connecting the message nodes is constructed; The message propagation graph is constructed based on the message nodes and the directed propagation edges.

8. A key impact message identification device, characterized in that, The device includes: The graphing module is used to obtain the interaction messages and response relationships between messages in the target event, and to construct a message propagation graph based on the interaction messages and response relationships; The distribution module is used to calculate the group position distribution based on the message nodes in the message propagation graph; The pruning module is used to determine the target message node from the message propagation graph, perform counterfactual pruning operation on the target message node to obtain the counterfactual pruning graph, and calculate the counterfactual group position distribution based on the counterfactual pruning graph. The attribution module is used to obtain the causal attribution score of the target message node based on the group position distribution and the counterfactual group position distribution, and to identify key influencing messages based on the causal attribution score.

9. A key impact message recognition device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the critical impact message identification method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the critical impact message identification method as described in any one of claims 1 to 7.

Images