Propagation effect evaluation system and method based on multi-platform data fusion

Abstract

The invention relates to the technical field of propagation effect evaluation, in particular to a propagation effect evaluation system and method based on multi-platform data fusion. In the propagation effect evaluation system and method based on the multi-platform data fusion, through normalization of forwarding frequency and position mapping with a user sorting index, behavior data is converted into a structured unit; the fusion capability of different platform data in the time dimension and the frequency dimension is enhanced, a sequence vector of a propagation path is established after a sequence difference value and a frequency difference value between nodes are processed through a graph neural network, and a coding structure is formed in combination with direction judgment and weight combination to reflect the seriality and the propagation intensity in the node propagation path, so that the propagation speed of the nodes is improved. Through continuous matching operation of a path sequence structure, similarity judgment is carried out on a track sequence, a core path with actual diffusion continuity is effectively screened out, aggregation processing is carried out according to platforms by combining node strength, a path bit sequence and a weighted relation among repeated indexes, regional difference values are extracted, and a propagation strength index of each platform is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of communication effect evaluation, and in particular to a communication effect evaluation system and method based on multi-platform data fusion. Background Art

[0002] The technical field of communication effect evaluation aims to conduct quantitative analysis and systematic evaluation of the coverage, transmission speed, user response level and influence of information during its dissemination on various media platforms. By constructing a communication model, setting an evaluation indicator system, and introducing data processing and analysis algorithms, it can achieve objective judgment on the effectiveness of communication activities, thereby providing decision-making support for content optimization, media strategy adjustment and marketing effect improvement.

[0003] A communication effect evaluation system based on multi-platform data fusion aims to solve the problems existing in the current communication effect evaluation, such as fragmented data sources, inconsistent evaluation results, and highly subjective evaluation dimensions. It can integrate data from multiple heterogeneous platforms, establish a unified communication effect evaluation model, and present the results with multi-dimensional indicators, thereby achieving quantitative evaluation effects on multiple dimensions such as communication breadth, depth, continuity, and user response.

[0004] Existing technologies have the problem of heterogeneous data structures, which makes it difficult to carry out analysis synchronously. In the process of path identification and influence factor extraction, they rely on static statistical values ​​such as the number of nodes, forwarding volume or coverage rate, and lack dynamic structure judgment based on behavioral paths. They are unable to reveal the true trunk path and strong node distribution relationship in the communication structure, and it is difficult to simultaneously consider the interactive influence of user communication timing and behavior frequency, resulting in high-frequency but discontinuous behaviors being mistakenly identified as high-impact nodes. Simple classification methods are generally used to divide path segments by platform or time, ignoring the reconstruction and matching of structural features in the communication trajectory, resulting in the same communication chain being divided and attributed, and lacking the ability to judge cross-platform continuity and communication scalability. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a communication effect evaluation system and method based on multi-platform data fusion.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solutions: A communication effect evaluation system based on multi-platform data fusion includes:

[0007] Node behavior decomposition module: Based on multi-platform user behavior records, it extracts time point sets, sorts target users, and normalizes forwarding frequencies. It then maps frequency values ​​to sequential indexes to generate node propagation feature sets.

[0008] Propagation trajectory construction module: Based on the node propagation feature set, the forwarding node sequence and interval time are extracted, and a graph neural network is used to construct a propagation path sequence vector based on the node sequence difference. The direction judgment and weight mapping are combined with the frequency difference to establish a path sequence relationship model;

[0009] Influence signal extraction module: Based on the path sequence relationship model, it extracts the number of upstream and downstream nodes and calculates the propagation difference. It selects nodes with a propagation difference greater than zero, performs a product judgment based on the path length difference and the frequency value, and screens out nodes that meet the conditions to obtain a high-influence signal node group;

[0010] Propagation core identification module: Based on the high-impact signal node group, the frequency of the path segments to which they belong is counted and the density is determined. The time coverage is matched and an upper limit is set. After extracting the trajectory, path distribution aggregation and sequence matching are performed to obtain the core propagation path set.

[0011] Strength index generation module: Based on the core propagation path set, extract the node strength and path order, perform weighted operations on repeated indexes, group and aggregate by platform, filter the platform difference values ​​in combination with regional identification, and obtain the propagation platform strength index.

[0012] As a further solution of the present invention, the node behavior decomposition module includes:

[0013] Time series extraction submodule: Based on multi-platform user behavior records, it extracts a set of user forwarding time points. By verifying the time format validity of the time point set and sorting it, it creates a time mapping index set in ascending order. The index result is bound to the event number and output to generate a time series index set.

[0014] User sequence sorting submodule: Based on the time series index set, the target user list is extracted. By matching the target identification value in the forwarding node with the index sequence one by one and rearranging the sequence in chronological order, the user mapping sequence is unified and integrated with the event sequence to generate a target user sequence group;

[0015] Frequency sequence mapping submodule: Based on the target user sequence group, the number of repeated user appearances in each node is counted, the frequency value normalization operation is completed and the frequency vector is established. The normalized results are then matched to the corresponding sorting index positions in sequence to construct a sequential frequency vector group and generate a node propagation feature set.

[0016] As a further solution of the present invention, the propagation trajectory construction module includes:

[0017] Node sequence generation submodule: Based on the node propagation feature set, it extracts the start and target nodes, matches the node pairs in the user sequence and reads the time index order, extracts the index difference and maps it to the node number position, constructs a structure in which the nodes are arranged in time sequence, and obtains the node position index vector;

[0018] Propagation path vector construction submodule: Based on the node position index vector, the time difference and corresponding frequency difference of each pair of nodes are extracted. The two types of differences are linearly superimposed and the node position splicing operations are performed through the graph neural network to construct a composite structure. The structure is then grouped and combined according to the path order and output to generate a path order weight structure.

[0019] Direction weight judgment submodule: Based on the path sequence weight structure, perform node propagation direction comparison operation, judge the path direction by comparing the time difference sign and the bit sequence change value, then classify the node pairs in ascending and descending order of weight, combine the path structure information and output it, and establish a path sequence relationship model.

[0020] As a further solution of the present invention, the graph neural network is according to the formula:

[0021]

[0022] in: Indicates that in the communication effect evaluation system, the node After the The fused representation vector after the layer graph neural network is updated, represents a nonlinear activation function, Indicates the The weight matrix of the layer, the parameters are optimized by gradient descent during training, Representation node In the The representation vector of the layer, Represents the splicing operation, which is used to connect the current node representation with the neighbor aggregation information. Indicates the node The set of neighbor nodes Perform aggregation operations, Represents neighbor nodes In the The representation vector of the layer, Represents a node pair , The propagation time difference, Represents a node pair , The difference in propagation frequency, Represents a node pair , The structural tightness coefficient, Representation node The path sequence encoding, 、 、 、 Indicates the weight coefficient of each influencing factor.

[0023] As a further solution of the present invention, the graph neural network first constructs a directed graph structure between nodes, where each node of the graph corresponds to a user entity, and each edge represents a propagation event. The features of the edge include the time difference and the propagation frequency difference between the node pairs, which are then input into the graph neural network model. Graph convolution is used to aggregate the node features, and each node is weighted according to the edge features of the adjacent nodes. After superimposing the adjacent node features, a linear transformation is performed to obtain a node embedding vector. The node embedding vector is then concatenated with the matching position index vector to form a composite representation structure, and a path sequence partitioning operation is performed on the output layer of the graph neural network. Through a clustering method, the composite representation is logically combined according to the propagation path, and the embedded representation of each path is output to generate a path sequence weight structure with propagation strength and timing characteristics.

[0024] As a further solution of the present invention, the influence signal extraction module includes:

[0025] Upstream and downstream quantity extraction submodule: Based on the path sequence relationship model, a node pair set construction operation is performed. By counting the roles of the source nodes and the target nodes in all node pairs and dividing the nodes into upstream and downstream categories, the upstream and downstream quantity of each node is summarized and the difference between the previous and next quantities is calculated to obtain the upstream and downstream difference set;

[0026] Propagation strength screening submodule: Based on the upstream and downstream difference sets, a node screening operation is performed. The initial screening is completed by extracting nodes whose difference values ​​are greater than zero. The total path length of the corresponding node and the number of times the node is repeated in the path are extracted. The product calculation is performed and a fixed threshold is set on the product value to filter out low-value nodes, thereby obtaining a propagation strength node set.

[0027] Influence node identification submodule: Based on the propagation strength node set, the local density statistical operation of the nodes in the network is performed. By calculating the number of intersections between nodes that are mutually targeted and performing classification and partitioning operations, a set of nodes with the largest connection strength in each partition is extracted as the screening result to obtain a high-influence signal node group.

[0028] As a further solution of the present invention, the propagation core identification module includes:

[0029] Path frequency statistics submodule: Based on the high-impact signal node group, a path segment extraction operation is performed. By obtaining the propagation path segment to which each node belongs and counting the occurrence frequency value of each path segment, the path density value is calculated based on the number of nodes covered by each segment. The frequency value and density value are then double-filtered to obtain a set of frequency-dense path segments.

[0030] Time coverage screening submodule: Based on the frequency-intensive path segment set, it performs time interval processing operations, extracts the time indexes of the start and end nodes of the path segment and calculates the span between the indexes. It then compares the span values ​​with the set time upper limit one by one and performs an over-limit elimination operation to obtain a time-constrained path segment group;

[0031] Trajectory path extraction submodule: Based on the time-constrained path segment group, a node sequence extraction operation is performed. By constructing a trajectory set for the nodes in the path segment in chronological order and performing a structural similarity matching operation between the trajectory sets, the node sequence with high trajectory structure continuity in the matching results is used as the output set to obtain the core propagation path set.

[0032] As a further solution of the present invention, the frequency value and the density value are subjected to a double screening operation, and the operation process includes two stages. In the first stage, the frequency of each path segment in the propagation diagram is counted, and a frequency threshold is set, and all path segments with frequencies higher than the threshold are retained. In the second stage, the density value of the path segment is further judged, and a density lower limit standard is set to eliminate path segments with density lower than the standard.

[0033] As a further solution of the present invention, the intensity index generation module includes:

[0034] Node index calculation submodule: Based on the core propagation path set, it performs node extraction operations within the path, calculates the product of the position number of each node in the path and the corresponding propagation strength value, extracts the number of repeated node occurrences, and performs a weighted accumulation operation on the product result to generate a node propagation influence value set;

[0035] Platform grouping and aggregation submodule: Based on the node propagation influence value set, perform node platform identification extraction operation, read the platform identification content attached to each node to perform classification operation and build a platform index table, perform the sum operation of the node values ​​in the platform and record the corresponding total value of the platform to generate a platform influence value group;

[0036] Regional difference extraction submodule: Based on the platform influence value group, the regional identification field extraction operation is performed, and a regional classification mapping table is constructed by identifying the platform's regional label value. The platform influence value is aggregated by region and summarized to obtain the communication platform strength index.

[0037] A communication effect evaluation method based on multi-platform data fusion is implemented based on the above-mentioned communication effect evaluation system based on multi-platform data fusion, and includes the following steps:

[0038] S1: Based on a multi-platform user behavior record dataset, extract the time index value involved in each record, select a set of user IDs with record values, sort them according to the time index order in the record, identify whether there is a continuous relationship between the user IDs with adjacent time and different IDs, count the occurrence frequency value of each user ID in the relationship and normalize the frequency value, then positionally match the normalized frequency value with the position index in the user ID sorting sequence to generate a node propagation feature set;

[0039] S2: Based on the node propagation feature set, all user ID pairs are extracted, and the difference between the time index difference and the frequency normalized value of each ID pair is obtained. The two difference value ranges are linearly combined using a graph neural network, and then spliced ​​with the user ID sorting position to form a triple structure. All triple structures are then grouped and processed in chronological order. Each group of structures is used as a path component, and the processing results are output in sequential order to obtain a path sequence relationship model;

[0040] S3: Based on the path sequence relationship model, identify the number of forward connections and backward connections of each user identifier in the path, calculate the propagation difference between the connection numbers, select user identifiers with positive propagation differences, read the position difference of the path segment corresponding to the identifier and the occurrence frequency value in the path, perform a product judgment on the two, filter out user identifiers with product values ​​higher than the screening standard, and construct a high-influence signal node group;

[0041] S4: Based on the high-influence signal node group, identify the unique number of the propagation path segment corresponding to each user identifier, count the frequency of occurrence in all propagation paths, read the number of user identifiers after deduplication in each path segment, compare the frequency value and the number of user identifiers with preset screening criteria, and only retain path segments with both the frequency value and the number of identifiers exceeding the criteria to obtain a set of frequency-intensive path segments;

[0042] S5: Based on the frequency-intensive path segment set, read the time index corresponding to the first and last user identifiers of each path segment, calculate the time span value and compare it with the time upper limit, remove the path segments that exceed the limit and obtain the retained path segment set, extract the user identifiers and sort them by time index to form a trajectory sequence set, call the structural similarity matching method to calculate the structural continuity between any two groups of trajectories, retain the trajectory sequence combinations with structural continuity higher than the matching standard, and obtain the core propagation path set.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are:

[0044] In this invention, by normalizing the forwarding frequency and mapping it with the user ranking index, the behavioral data is converted into structured units, which enhances the ability to integrate data from different platforms in the time and frequency dimensions;

[0045] In the present invention, the sequence difference and frequency difference between nodes are processed by a graph neural network to establish a sequence vector of the propagation path. Combined with direction judgment and weight combination, a coding structure is formed to reflect the sequence and propagation strength in the node propagation path.

[0046] In this invention, by counting the number of upstream and downstream connections of each node and calculating the propagation difference, supplemented by the product calculation of the path length difference and the frequency value, a precise screening of nodes with dominant propagation capabilities is achieved, which improves the accuracy of identifying key sources of propagation influence. By combining frequency statistics with time coverage constraints, a dual screening mechanism for path segments in terms of propagation density and duration is constructed, eliminating non-mainline path segments with low density or abnormal duration.

[0047] In the present invention, the similarity of the trajectory sequence is judged through the continuity matching operation of the path sequence structure, and the core paths with actual diffusion continuity are effectively screened out. Combined with the weighted relationship between node strength, path order and repeated index, aggregation processing is performed by platform, and regional difference values ​​are extracted to obtain the diffusion intensity index of each platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a system flow chart of the present invention;

[0049] Figure 2 Schematic diagram of the method steps of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to 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.

[0051] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0052] Example 1

[0053] See also Figure 1 The present invention provides a technical solution: a communication effect evaluation system based on multi-platform data fusion includes:

[0054] Node behavior decomposition module: Based on multi-platform user behavior records, it extracts time point sets, sorts target users, and normalizes forwarding frequencies. It then maps frequency values ​​to sequential indexes to generate node propagation feature sets.

[0055] Propagation trajectory construction module: Based on the node propagation feature set, it extracts the forwarding node sequence and interval time, uses a graph neural network to construct the propagation path sequence vector based on the node sequence difference, combines the frequency difference to perform direction judgment and weight mapping, and establishes a path sequence relationship model;

[0056] Influence signal extraction module: Based on the path sequence relationship model, it extracts the number of upstream and downstream nodes and calculates the propagation difference. Nodes with a propagation difference greater than zero are selected. The path length difference is multiplied by the frequency value to filter out nodes that meet the conditions and obtain a group of high-influence signal nodes.

[0057] Propagation core identification module: Based on the high-impact signal node group, the frequency of the path segments to which they belong is counted and the density is determined. The time coverage is matched and an upper limit is set. After extracting the trajectory, path distribution aggregation and sequence matching are performed to obtain the core propagation path set.

[0058] Strength index generation module: Based on the core propagation path set, node strength and path order are extracted, weighted operations are performed on repeated indexes, grouped and aggregated by platform, and platform difference values ​​are filtered in combination with regional identifiers to obtain the propagation platform strength index.

[0059] The node behavior decomposition module includes:

[0060] Time series extraction submodule: Based on multi-platform user behavior records, it extracts a set of user forwarding time points. By verifying the time format validity of the time point set and sorting it, it creates a time mapping index set in ascending order. The index result is bound to the event number and output to generate a time series index set.

[0061] User sequence sorting submodule: Based on the time series index set, the target user list is extracted. By matching the target identification value in the forwarding node with the index sequence one by one and rearranging the sequence in chronological order, the user mapping sequence is unified and integrated with the event sequence to generate a target user sequence group.

[0062] Frequency sequence mapping submodule: Based on the target user sequence group, it counts the number of repeated user appearances in each node, completes the frequency value normalization operation and establishes the frequency vector. It then matches the normalized results to the corresponding sorting index positions in sequence, constructs the sequential frequency vector group, and generates the node propagation feature set;

[0063] Time series extraction submodule: Based on multi-platform user behavior records, a time format standardization algorithm is used to process the user forwarding time point set, read the time field value in the forwarding record one by one, and judge the format of the year, month, day, hour, minute, and second involved in each record. The legitimacy is screened based on whether the time components are integer values ​​and within the time range. After filtering out the record items that do not meet the conditions, a legal time set is constructed, and the year field is arranged from small to large. When the year value is the same, the month, day, hour, minute, and second field values ​​are compared in sequence, and the ascending order in the set is completed in sequence. A position index is set for the sorted time set, and the index is numbered continuously starting from zero. Each time item corresponds to a unique position index. Then, a pairing relationship is established between the original event number corresponding to the time item and the index value. The number and index binding structure is output to generate a time series index set.

[0064] User sequence sorting submodule: Based on the time series index set, the target user list is extracted using the user mapping sequence construction method. Record entries with target identification values ​​are screened from the forwarding records, and the time index values ​​bound to the entries are extracted. All user identifiers are reordered from small to large according to the index values. The sorted user identifiers are assigned a mapping sequence according to their position in the sorting queue. The mapping sequence is then bidirectionally matched with the original event sequence number to form a unified position matching set between user identifiers and event records. The sequence position of each user identifier in the set and the event number position are integrated into a unified sorting structure. The integrated user sequence structure set is output to generate the target user sequence group.

[0065] Frequency sequence mapping submodule: Based on the target user sequence group, the user behavior frequency normalization algorithm is used to count and normalize the number of repetitions of each user identifier in the sequence, read each identifier value in the user sequence, and perform cumulative statistics on the same identifier value. After the statistics are completed, the cumulative frequency corresponding to each identifier value is converted with the user sequence length to obtain the normalized frequency value, and then the normalized result is matched according to the sorting index position of the user identifier in the sequence. Each frequency value is written into the position index corresponding to the user identifier. All normalized frequency values ​​are formed into a frequency vector sequence according to the sorting index structure, and the vector sequence and the sorting index are synchronously output to construct a position mapping structure set, a sequential frequency vector group is constructed, and a node propagation feature set is generated.

[0066] The propagation trajectory building blocks include:

[0067] Node sequence generation submodule: Based on the node propagation feature set, it extracts the start and target nodes. By matching node pairs in the user sequence and reading the time index order, it extracts the index difference and maps it to the node number position, constructs a structure in which nodes are arranged in time sequence, and obtains the node position index vector.

[0068] Propagation path vector construction submodule: Based on the node position index vector, it extracts the time difference and corresponding frequency difference of each pair of nodes. Through the graph neural network, the two types of differences are linearly superimposed and the node position splicing operations are performed to construct a composite structure. The structure is then grouped and combined according to the path order and output to generate a path order weight structure.

[0069] Direction weight judgment submodule: Based on the path sequence weight structure, it performs node propagation direction comparison operations, determines the path direction by comparing the time difference sign and the bit sequence change value, and then classifies the node pairs according to the ascending and descending weight order. The path structure information is combined and output to establish a path sequence relationship model;

[0070] Node sequence generation submodule: Based on the node propagation feature set, the node pair sequence construction method is used to extract the starting and target nodes. The user identifier in each record is used as the node number. All nodes are combined into pairs in the order of time index to form a node pair set. The node pairs are screened according to the screening criteria that the starting node time is earlier than the target node. The combination items with equal time index or starting node later than the target node are eliminated. The starting node number, target node number and the corresponding time index value difference are recorded for the filtered node pair structure. After reading the time index difference, the difference is mapped to the starting node number to form a time mapping structure. The node numbers in the mapping structure are rearranged according to the mapping value to generate a node sequence. The position of each node in the node sequence is sequentially numbered, and the numbers are formed into a continuous index vector output to obtain the node position index vector;

[0071] Propagation path vector construction submodule: Based on the node position index vector, the graph neural network structure processing method is used to extract the time difference and frequency difference of each pair of nodes. The time index difference and the frequency normalization value difference are read in sequence in units of node pairs. The two difference vectors are linearly integrated using the information superposition structure in the graph neural network. The linear integration uses the starting time index between the nodes as the input initial parameter, and the superposition weight is set to a one-to-one ratio of the time difference to the frequency difference. A two-dimensional mixed difference structure is output, and the structure and the node position index are spliced. The position index is used as the last dimension of the structure during splicing. The spliced ​​result is path aggregated according to the node order. The aggregation is grouped according to the judgment rule that each group of consecutive nodes is a path. All path groups are combined and output to generate a path sequence weight structure.

[0072] Direction weight judgment submodule: Based on the path sequence weight structure, the propagation direction state judgment method is used to perform node propagation direction comparison operations. The time index difference is read for each node pair and the positive and negative states are judged. At the same time, the position difference of the node pair in the path structure is extracted. The position difference is the position number of the target node in the path minus the position number of the starting node. Both values ​​are non-negative integers. If the time difference and the position difference are in the same direction, it is forward propagation. If the directions are opposite, it is reverse propagation. All node pairs are classified into forward sets and reverse sets according to the direction state, and then the path weight value corresponding to each node pair is extracted. All node pairs are sorted and classified from large to small according to the weight value, and the classification results are integrated into path structure units according to the node pair combination relationship. After outputting the integrated structure, a path sequence relationship model is established.

[0073] Graph neural network, according to the formula:

[0074]

[0075] in: Indicates that in the communication effect evaluation system, the node After the The fused representation vector after the layer graph neural network is updated, represents a nonlinear activation function, Indicates the The weight matrix of the layer, the parameters are optimized by gradient descent during training, Representation node In the The representation vector of the layer, Represents the splicing operation, which is used to connect the current node representation with the neighbor aggregation information. Indicates the node The set of neighbor nodes Perform aggregation operations, Represents neighbor nodes In the The representation vector of the layer, Represents a node pair , The propagation time difference, Represents a node pair , The difference in propagation frequency, Represents a node pair The structural tightness coefficient, Representation node The path sequence encoding, 、 、 、 Indicates the weight coefficient of each influencing factor;

[0076] Execution process: First, for any node to be updated , collect the set of neighbor nodes connected in the propagation path from multiple platforms , extract each neighbor node The current representation vector of , then record the compute node pair based on the timestamp in the multi-platform data The propagation time difference , and calculate the difference in propagation frequency of nodes per unit time through event log statistics , and then calculate the structural tightness coefficient , which is determined by the ratio of the number of common neighbors of a node to the number of adjacent sets, reflecting the degree of coupling in the propagation path structure between platforms and determining the node The order of the bits in the propagation path , and input the sine function to generate the bit sequence code , and then the vectors for the above neighbors are represented , time difference , frequency difference , structural compactness and positional coupling · Multiply by the weight coefficient 、 、 、 All weights are initially set to a normalized distribution, and are dynamically adjusted based on the loss function feedback during the model training phase to minimize the evaluation error. The weighted results are summed to form the neighbor aggregation feature, which is then combined with the node itself to represent For splicing, through the weight matrix Map and input nonlinear activation function , the propagation representation vector of the output node in the next layer , iterates in all propagation path nodes in turn, and finally outputs the complete path sequence weight structure to evaluate the multi-platform propagation effect.

[0077] Graph neural network, first, constructs a directed graph structure between nodes. Each node in the graph corresponds to a user entity, and each edge represents a propagation event. The edge features include the time difference and propagation frequency difference between node pairs. Then, it is input into the graph neural network model, and graph convolution is used to aggregate the node features. Each node is weighted according to the edge features of the adjacent nodes. After superimposing the adjacent node features, a linear transformation is performed to obtain the node embedding vector. Then the node embedding vector is spliced ​​with the matching position index vector to form a composite representation structure, and the path sequence partitioning operation is performed in the output layer of the graph neural network. Through the clustering method, the composite representation is logically combined according to the propagation path, and the embedded representation of each path is output to generate a path sequence weight structure with propagation strength and timing characteristics.

[0078] The impact signal extraction module includes:

[0079] Upstream and downstream quantity extraction submodule: Based on the path sequence relationship model, it constructs a node pair set. By counting the roles of the source and target nodes in all node pairs and dividing the nodes into upstream and downstream categories, it summarizes the upstream and downstream quantities of each node and calculates the difference between the previous and next quantities to obtain the upstream and downstream difference set.

[0080] Propagation strength screening submodule: Based on the upstream and downstream difference sets, node screening operations are performed. The initial screening is completed by extracting nodes with a difference greater than zero. The total path length of the corresponding node and the number of times the node is repeated in the path are then extracted. The product calculation is performed and a fixed threshold is set on the product value to filter out low-value nodes to obtain the propagation strength node set.

[0081] Influence node identification submodule: Based on the propagation strength node set, it performs local density statistics of nodes in the network. By calculating the number of intersections between nodes that are mutually targeted and performing classification and partitioning operations, it extracts the set of nodes with the strongest connection strength in each partition as the screening result, and obtains the high-influence signal node group;

[0082] Upstream and downstream number extraction submodule: Based on the path sequence relationship model, the node role division statistical method is used to construct the node pair set. Each pair of nodes in the path encoding result is used as a record item to extract the two fields of source node number and target node number. The source node number is recorded as the output node, and the target node number is recorded as the input node. The source node number in all records is uniquely removed and the upstream node set is established. The target node number is uniquely removed and the downstream node set is established. The number of times each node appears as a source node is counted to obtain the node upstream count value, and the number of times each node appears as a target node is counted to obtain the downstream count value. The difference between the upstream count value and the downstream count value is calculated for each node, and the upstream and downstream count difference entry of each node is constructed. All node entries are integrated and the node number and the corresponding upstream and downstream difference pair are output to obtain the upstream and downstream difference set.

[0083] Propagation intensity screening submodule: Based on the upstream and downstream difference sets, the propagation intensity threshold calculation method is used to perform node screening. First, a preliminary screening is performed on the nodes with upstream and downstream difference greater than zero, and the nodes with difference less than or equal to zero are directly eliminated. Only positive node numbers are retained. The path number set of the retained node numbers is read, and the number of nodes in each path in the set is counted to obtain the total path length. The number of times each node appears in the path is independently counted to obtain the repetition frequency value. The total path length and the node repetition frequency are then multiplied and calculated. The product value is recorded as the node propagation measurement value. A fixed screening threshold is set in all node measurement value sets. All node numbers less than the threshold are deleted. The list of retained node numbers is output to obtain the propagation intensity node set.

[0084] Influence node identification submodule: Based on the propagation intensity node set, the density statistics method based on the mutual target relationship is used to identify local network connections. Each node in the node number set is taken as the starting node, and the target node set in the path structure is counted. The node numbers that meet the mutual target conditions are found to point to the original starting node again. The node numbers that meet the mutual target conditions are extracted as bidirectional connection units, and the number of times each node appears in the bidirectional connection unit is calculated as the connection density count value. All node density values ​​are grouped and sorted, and nodes with the same density count value are classified into the same partition. The propagation weights in the path structure between the nodes in each partition are compared and the node number group corresponding to the maximum value is taken as the strongest connection set of the partition. All the strongest connection sets are merged and the identification results are output to obtain a high-influence signal node group.

[0085] The core identification module of communication includes:

[0086] Path frequency statistics submodule: Based on the high-impact signal node group, path segment extraction is performed. By obtaining the propagation path segment to which each node belongs and counting the occurrence frequency value of each path segment, the path density value is calculated based on the number of nodes covered by each segment. The frequency value and density value are then double-filtered to obtain a set of frequency-dense path segments.

[0087] Time coverage screening submodule: Based on the frequency-intensive path segment set, it performs time interval processing operations. By extracting the time indexes of the start and end nodes of the path segment and calculating the span between the indexes, it compares the span values ​​with the set time upper limit one by one and performs an over-limit elimination operation to obtain the time-constrained path segment group.

[0088] Trajectory path extraction submodule: Based on the time-constrained path segment group, it performs node sequence extraction operations. By constructing a trajectory set for the nodes in the path segment in chronological order and performing structural similarity matching operations between the trajectory sets, the node sequence with high trajectory structure continuity in the matching results is used as the output set to obtain the core propagation path set;

[0089] Path frequency statistics submodule: Based on the high-impact signal node group, the path node attribution frequency calculation method is used to perform path segment extraction operations. The path number corresponding to each node is read one by one, and a mapping table of nodes and paths is constructed. The number of associated nodes for each path number is accumulated and the node set under the path number is recorded. The number of times the path number appears in all node records in the set is counted as the path frequency value. The node set under the path number is uniquely counted to obtain the path node number value. The path frequency value and the node number value are then paired and recorded for the path number. The frequency screening threshold and density screening threshold are set. Path numbers with frequency values ​​greater than the frequency threshold and node number values ​​greater than the density threshold are screened out and marked as valid path segments. All valid path segment numbers and corresponding node sets are output to generate a frequency-dense path segment set.

[0090] Time coverage screening submodule: Based on the frequency-intensive path segment set, the time span comparison screening method is used to perform time interval processing operations. The time index values ​​of the first node and the last node in each path segment are read, and the time indexes corresponding to the first and last nodes are combined according to the path segment number. The time index difference of the combination result is calculated as the time span record. A fixed time upper limit is set, and the time span of each path segment is compared with the time upper limit one by one. Only the path segment numbers with time span values ​​less than or equal to the upper limit are retained. The corresponding node sequence set is then extracted and output according to the path number. The set numbers of all path segments that meet the time conditions are summarized to generate a time-constrained path segment group.

[0091] Trajectory path extraction submodule: Based on the time-constrained path segment group, the trajectory sequence structure continuity matching algorithm is used to perform node sequence extraction operations. The node set under each path segment number is sorted in ascending order according to the time index value to generate a trajectory sequence set. All trajectory sequence sets are paired, and structural similarity comparison is performed on each pair of trajectories. In the structural similarity comparison operation, the two trajectories are compared one by one according to the node number, and the number of identical node numbers at the corresponding index position is calculated. The number of matches is then divided by the length of the shorter trajectory to form a matching ratio value. Trajectory pairs with matching ratio values ​​higher than the set threshold in all pairs are recorded as structural continuity pairs, and any trajectory sequence in each structural continuity pair is extracted as a representative trajectory. The trajectory set numbers and node orders of all qualified trajectories are output to obtain the core propagation path set.

[0092] The frequency value and density value are double-screened. The operation process includes two stages. In the first stage, the frequency of each path segment in the propagation map is counted, and a frequency threshold is set to retain all path segments with a frequency higher than the threshold. In the second stage, the density value of the path segment is further judged, and the lower limit standard of the density is set to eliminate the path segments with a density lower than the standard.

[0093] The intensity index generation module includes:

[0094] Node index calculation submodule: Based on the core propagation path set, it extracts nodes within the path, calculates the product of the position number of each node in the path and the corresponding propagation strength value, extracts the number of repeated node occurrences and performs a weighted accumulation operation on the product result to generate a node propagation influence value set;

[0095] Platform grouping and aggregation submodule: Based on the node propagation influence value set, it extracts the node platform identifier, reads the platform identifier content attached to each node, performs classification operations and builds a platform index table, performs the sum operation of the node values ​​within the platform and records the corresponding total value of the platform to generate a platform influence value group;

[0096] Regional Difference Extraction Submodule: Based on the platform influence value group, it extracts the regional identification field, builds a regional classification mapping table by identifying the platform's region label value, aggregates the platform influence value by region, and summarizes it to obtain the communication platform strength index;

[0097] Node index calculation submodule: Based on the core propagation path set, the node propagation position weighted superposition algorithm is used to extract nodes within the path. The node list under each path set number is numbered in chronological order, the position number of each node in the path is marked and a number mapping table is established. The propagation intensity value recorded in the path structure corresponding to each node is extracted, and the node position number and the corresponding propagation intensity value are read and multiplied. The product result is used as the initial influence value of the node. Repeated statistical operations are then performed on the repeated node numbers in all path sets. The number of occurrences of each node number is used as a weighted multiple value. All product results of the same node number are accumulated according to the weight, and the set of all node numbers and accumulated result values ​​is output to generate a node propagation influence value set;

[0098] Platform grouping and aggregation submodule: Based on the node propagation influence value set, the node platform identifier is extracted using the platform identifier field clustering and summation method. The platform identifier field content is read item by item for the combination of node number and influence value, and the node numbers with the same platform identifier field are classified into the same platform classification structure. An index mapping table between platform identifiers and node numbers is established. The influence values ​​of all node numbers belonging to each platform classification are summed up. The summed results are paired with the platform identifiers one by one and recorded and processed according to the platform identifier number. The summed value structure under all platforms is summarized and the node influence value results corresponding to each platform are output to generate a platform influence value group.

[0099] Regional difference extraction submodule: Based on the platform impact value group, the platform regional aggregation index algorithm is used to extract the regional identification field, read the regional affiliation label attached to each platform identification one by one, classify and organize the platform identification according to the regional affiliation label, and construct a mapping relationship matrix between the regional identification and the platform identification. After reading the corresponding platform impact value in each region, the summation processing is performed, and each platform impact value is set as a single input. The total statistics of the platform values ​​in the region are performed, and the accumulated values ​​corresponding to each region are combined with the regional label to form a structure list. The result record structure of all regions is summarized and uniformly coded to obtain the communication platform strength index.

[0100] A communication effect evaluation method based on multi-platform data fusion is implemented based on the above-mentioned communication effect evaluation system based on multi-platform data fusion, and includes the following steps:

[0101] S1: Based on a multi-platform user behavior record dataset, extract the time index value involved in each record, select a set of user IDs with record values, sort them according to the time index order in the record, identify whether there is a continuous relationship between the user IDs with adjacent time and different IDs, count the occurrence frequency value of each user ID in the relationship and normalize the frequency value, then positionally match the normalized frequency value with the position index in the user ID sorting sequence to generate a node propagation feature set;

[0102] S2: Based on the node propagation feature set, all user ID pairs are extracted. The difference between the time index difference and the frequency normalized value of each ID pair is obtained. The two difference value ranges are linearly combined using a graph neural network. These are then concatenated with the user ID ranking to form a triple structure. All triple structures are then grouped in chronological order. Each group of structures is used as a path component. The processing results are output in sequential order to obtain a path sequence relationship model.

[0103] S3: Based on the path sequence relationship model, identify the number of forward connections and backward connections for each user ID in the path, calculate the propagation difference between the connection numbers, select user IDs with positive propagation differences, read the position difference of the path segment corresponding to the ID and the occurrence frequency value in the path, perform a product judgment on the two, filter out user IDs with product values ​​higher than the screening criteria, and construct a high-influence signal node group;

[0104] S4: Based on the high-influence signal node group, identify the unique number of the propagation path segment corresponding to each user ID, count the frequency of occurrence in all propagation paths, read the number of user IDs after deduplication in each path segment, compare the frequency value and the number of user IDs with the preset screening criteria, and only retain the path segments with both the frequency value and the number of IDs exceeding the criteria to obtain a set of frequency-intensive path segments;

[0105] S5: Based on the frequency-dense path segment set, read the time index corresponding to the first and last user identifiers of each path segment, calculate the time span value and compare it with the time upper limit, remove the path segments that exceed the limit and obtain the retained path segment set, extract the user identifiers and sort them by time index to form a trajectory sequence set, call the structural similarity matching method to calculate the structural continuity between any two groups of trajectories, retain the trajectory sequence combinations with structural continuity higher than the matching standard, and obtain the core propagation path set.

[0106] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A communication effect evaluation system based on multi-platform data fusion, characterized by: The system comprises: Node behavior decomposition module: Based on multi-platform user behavior records, it performs time point set extraction, target user sorting, and forwarding frequency normalization, mapping frequency values ​​to sequential indexes to generate node propagation feature sets; Propagation trajectory construction module: Based on the node propagation feature set, the forwarding node sequence and interval time are extracted, and a graph neural network is used to construct a propagation path sequence vector based on the node sequence difference. The direction judgment and weight mapping are combined with the frequency difference to establish a path sequence relationship model; Influence signal extraction module: Based on the path sequence relationship model, the number of upstream and downstream nodes and the propagation difference are calculated, and nodes with propagation difference greater than zero are screened. The high-influence signal node group is obtained by combining the product of the path length difference and the frequency value; Propagation core identification module: Based on the high-impact signal node group, the frequency of the path segments to which they belong is counted and the density is determined. The time coverage is matched and an upper limit is set. After extracting the trajectory, path distribution aggregation and sequence matching are performed to obtain the core propagation path set. Strength index generation module: Based on the core propagation path set, extract node strength and path order, repeated index weighting, group and aggregate by platform, combine regional identification to filter platform difference values, and generate a propagation platform strength index.

2. The communication effect evaluation system based on multi-platform data fusion according to claim 1 is characterized in that: The node behavior decomposition module includes: Time series extraction submodule: Based on multi-platform user behavior records, extract the user forwarding time point set, verify the time format and sort in ascending order, establish a time mapping index and bind it to the event number to generate a time series index set; User sequence sorting submodule: extracts the target user list based on the time series index set, matches the target identifier of the forwarding node with the index sequence, and rearranges them in chronological order to obtain a target user sequence group; Frequency sequence mapping submodule: Based on the target user sequence group, count the number of repeated user appearances and normalize them, establish a frequency vector, match the normalized result to the sorting index position, and generate a node propagation feature set.

3. The communication effect evaluation system based on multi-platform data fusion according to claim 1 is characterized in that: The propagation trajectory construction module includes: Node sequence generation submodule: Based on the node propagation feature set, it extracts the start and target nodes, matches the node pairs in the user sequence and reads the time index order, extracts the index difference and maps it to the node number position, constructs a structure in which the nodes are arranged in time sequence, and obtains the node position index vector; Propagation path vector construction submodule: Based on the node position index vector, the time difference and corresponding frequency difference of each pair of nodes are extracted. The two types of differences are linearly superimposed and the node position splicing operations are performed through the graph neural network to construct a composite structure. The structure is then grouped and combined according to the path order and output to generate a path order weight structure. Direction weight judgment submodule: Based on the path sequence weight structure, perform node propagation direction comparison operation, judge the path direction by comparing the time difference sign and the bit sequence change value, then classify the node pairs in ascending and descending order of weight, combine the path structure information and output it, and establish a path sequence relationship model.

4. The communication effect evaluation system based on multi-platform data fusion according to claim 3 is characterized in that: The graph neural network is based on the formula: ； in: Indicates that in the communication effect evaluation system, the node After the The fused representation vector after the layer graph neural network is updated, represents a nonlinear activation function, Indicates the The weight matrix of the layer, the parameters are optimized by gradient descent during training, Representation node In the The representation vector of the layer, Represents the splicing operation, which is used to connect the current node representation with the neighbor aggregation information. Indicates the node The set of neighbor nodes Perform aggregation operations, Represents neighbor nodes In the The representation vector of the layer, Represents a node pair , The propagation time difference, Represents a node pair , The difference in propagation frequency, Represents a node pair , The structural tightness coefficient, Representation node The path sequence encoding, 、 、 、 Indicates the weight coefficient of each influencing factor.

5. The communication effect evaluation system based on multi-platform data fusion according to claim 3 is characterized in that: The graph neural network first constructs a directed graph structure between nodes, where each node of the graph corresponds to a user entity, and each edge represents a propagation event. The features of the edge include the time difference and the propagation frequency difference between the node pairs. The edges are then input into the graph neural network model, and graph convolution is used to aggregate the node features. Each node is weighted according to the edge features of the adjacent nodes. After superimposing the adjacent node features, a linear transformation is performed to obtain a node embedding vector. The node embedding vector is then concatenated with the matching position index vector to form a composite representation structure, and a path sequence partitioning operation is performed on the output layer of the graph neural network. Through a clustering method, the composite representation is logically combined according to the propagation path, and the embedded representation of each path is output to generate a path sequence weight structure with propagation strength and timing characteristics.

6. The communication effect evaluation system based on multi-platform data fusion according to claim 1 is characterized in that: The impact signal extraction module includes: Upstream and downstream quantity extraction submodule: Based on the path sequence relationship model, it constructs a node pair set, counts the roles of source nodes and target nodes, divides upstream and downstream categories, summarizes the quantities and calculates the difference between the previous and next quantities to generate an upstream and downstream difference set; Propagation strength screening submodule: Based on the upstream and downstream difference set, screen nodes with a difference greater than zero, extract their total path length and number of repetitions, calculate the product and compare it with the threshold, eliminate low-value nodes, and obtain a propagation strength node set; Influence node identification submodule: Based on the propagation strength node set, the number of intersections between nodes that are mutually targeted is counted, classified by partition, and a group of nodes with the largest connection strength in each partition is extracted as a high-influence signal node group.

7. The communication effect evaluation system based on multi-platform data fusion according to claim 1 is characterized in that: The propagation core identification module includes: Path frequency statistics submodule: Based on the high-impact signal node group, a path segment extraction operation is performed. By obtaining the propagation path segment to which each node belongs and counting the occurrence frequency value of each path segment, the path density value is calculated based on the number of nodes covered by each segment. The frequency value and density value are then double-filtered to obtain a set of frequency-dense path segments. Time coverage screening submodule: Based on the frequency-intensive path segment set, it performs time interval processing operations, extracts the time indexes of the start and end nodes of the path segment and calculates the span between the indexes. It then compares the span values ​​with the set time upper limit one by one and performs an over-limit elimination operation to obtain a time-constrained path segment group; Trajectory path extraction submodule: Based on the time-constrained path segment group, a node sequence extraction operation is performed. By constructing a trajectory set for the nodes in the path segment in chronological order and performing a structural similarity matching operation between the trajectory sets, the node sequence with high trajectory structure continuity in the matching results is used as the output set to obtain the core propagation path set.

8. The communication effect evaluation system based on multi-platform data fusion according to claim 6 is characterized in that: The double screening operation of frequency value and density value includes two stages. In the first stage, the frequency of each path segment in the propagation map is counted, and a frequency threshold is set to retain all path segments with a frequency higher than the threshold. In the second stage, the density value of the path segment is further judged, and a lower limit standard of density is set to eliminate the path segments with a density lower than the standard.

9. The communication effect evaluation system based on multi-platform data fusion according to claim 1 is characterized in that: The intensity index generation module includes: Node index calculation submodule: Based on the core propagation path set, extract the nodes in the path, read the sequence number of each node and calculate the product of the propagation strength value, combine the weighted accumulation of the number of repeated occurrences of the node, and generate a node propagation influence value set; Platform grouping and aggregation submodule: Based on the node propagation influence value set, extract the node platform identifier, classify and construct the platform index table, calculate the sum of the node values ​​in the platform, and generate the platform influence value group; Regional difference extraction submodule: Based on the platform influence value group, extract the platform's region label, build a regional classification mapping table, aggregate and summarize the platform influence values ​​within the region, and obtain the communication platform strength index.

10. A communication effect evaluation method based on multi-platform data fusion, characterized in that: The communication effect evaluation system based on multi-platform data fusion according to any one of claims 1 to 9 is implemented, comprising the following steps: S1: Extract the time index and user ID of multi-platform user behavior records, count the frequency of user ID occurrence and normalize it, and generate a node propagation feature set based on the time index; S2: Extracting user identification pairs based on the node propagation feature set, calculating the time index difference and frequency difference of the user identification pairs, fusing the two types of differences using a graph neural network, grouping and generating a path formation sequence, and outputting a path sequence relationship model; S3: Calculate the number of previous and next connections of the user based on the path sequence relationship model to obtain a propagation difference, select users with a positive propagation difference, and construct a high-influence signal node group based on the product of the path position difference and the occurrence frequency; S4: Based on the high-influence signal node group, count the occurrence frequency and number of users of the path segments, retain the path segments whose occurrence frequency and number of users are both higher than the set standard, and obtain a frequency-intensive path segment set; S5: Extract the time index of the first and last users to calculate the time span, remove the excessive path segments in the frequency-intensive path segments based on the time span and construct a trajectory sequence, filter out trajectory sequences with high structural continuity through structural similarity, and obtain the core propagation path set.

Images