Large factory cross-platform user drainage and growth promotion system based on artificial intelligence

Abstract

The invention relates to the technical field of platform drainage and promotion, in particular to a large factory cross-platform user drainage and growth promotion system based on artificial intelligence, and provides a basis for subsequent data analysis through behavior time axis alignment and interest dimension unification and attention attenuation node labeling; the interest of the low-activity platform user is effectively activated by constructing an interest energy conduction network and performing directional path planning to optimize resource allocation; by generating a three-dimensional tensor field, the spatial-temporal change and theme relevance of interest energy is captured, and a multi-dimensional analysis framework is provided for a complex interest scene; user interest energy ecological niche features are quantified based on a three-dimensional tensor field, multi-level dynamic clustering is achieved, and the user layering operation efficiency and the conversion effect are improved; by defining platform interest energy ecological niche indexes, identifying dominant platforms and complementary combinations, and optimizing platform ecological layout and synergistic effect; by generating the interest energy guide chain, the viscidity and activeness of the user to the platform are improved.

Description

Technical Field

[0001] The present invention relates to the field of platform traffic diversion and promotion technology, and in particular to an artificial intelligence-based cross-platform user traffic diversion and growth promotion system for large companies. Background Art

[0002] There are still many challenges in the current platform traffic generation and promotion management: the current user behavior data on different platforms (such as social and e-commerce platforms) is scattered and independent, making it difficult for the platform to fully understand user interests and achieve accurate traffic generation; there is resource competition among multiple platforms under large companies, and some platforms have low user activity (such as newly launched platforms or vertical field platforms), but there is a lack of effective means to activate these users; user interests change rapidly over time and in different scenarios (such as from games to fitness), but traditional recommendation systems rely on static tags and have difficulty capturing interest migration in real time, resulting in a disconnect between recommended content and user needs; at the same time, there is a lack of collaboration between platforms, resulting in insufficient ecological competitiveness (such as users repeating operations between different platforms), and there is a phenomenon that dominant platforms fail to effectively drive the development of other platforms. To this end, the present invention proposes an artificial intelligence-based cross-platform user traffic generation and growth promotion system for large companies. Summary of the Invention

[0003] The purpose of the present invention is to solve the problems in the background technology and to propose an artificial intelligence-based cross-platform user traffic diversion and growth promotion system for large companies.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions: An AI-based cross-platform user traffic generation and growth promotion system for large companies, including: a behavior timeline alignment module, an interest energy conduction module, an interest energy tensor construction module, a user hierarchical clustering module, an ecological niche platform evaluation module, and a hierarchical guidance chain generation module; Behavior Timeline Alignment Module: This module stitches user activity records across different platforms into a continuous behavior timeline, using algorithms to uniformly map interest dimensions. It uses stacked area charts to illustrate the interest migration process and annotates the attention decay nodes when user behavior jumps between platforms. Interest Energy Conduction Module: This module identifies low- and high-activity platforms based on attention decay nodes, constructs an interest energy conduction network to quantify the interest kinetic energy transfer relationship between platforms, generates a directional conduction path through path planning and energy loss calculation, and outputs the interest energy distribution. Interest Energy Tensor Construction Module: Based on the acquired interest energy distribution, it deconstructs the multi-dimensional dynamic characteristics of interest energy from the three dimensions of time, topic, and user sensitivity, and maps the multi-dimensional dynamic characteristics to the corresponding subspaces. It then fuses all subspaces based on the tensor product to generate a three-dimensional interest energy tensor field. User-level clustering module: Based on the three-dimensional interest energy tensor field, the niche sensitivity function is used to quantify the niche characteristics of user interest energy distribution, and multi-level clustering of users is obtained. The competition-symbiosis game model is used to achieve dynamic migration of user levels and confidence-weighted label output. Niche platform evaluation module: defines the platform interest energy niche indicators, calculates the position of each platform in the interest energy niche, and ranks them; at the same time, screens out the optimal platform and platform combination; Hierarchical guidance chain generation module: For users at different levels, guidance chains are constructed to optimize the interest energy distribution, and finally a weighted output of the user-platform interest energy guidance chain set is generated.

[0005] Furthermore, the user's operation records on different platforms are spliced into a continuous behavior timeline, and an algorithm is used to uniformly map the interest dimensions. A stacked area chart is used to show the interest migration process, and the attention decay nodes when the user's behavior jumps between platforms are marked. The process includes: Collect user operation records from various platforms used by the user; based on the timestamp of each operation record, splice the user operation records on each platform into a continuous behavior timeline in chronological order, and mark the type of each operation record on the behavior timeline; extract user behavior features from the user behavior timeline, these features will serve as the basis for behavioral fingerprints; use the behavioral fingerprint alignment algorithm to map user behavior features on different platforms to the same interest dimension; define a set of interest dimensions, and map user behavior features to the defined interest dimensions; Based on the aligned behavior timeline, analyze the user behavior jumps between different platforms: record the user's operation behavior when jumping from one platform to another, as well as the changes in interest dimensions before and after the jump; for each time point on the continuous behavior timeline, count the distribution of users' interest dimensions on different platforms; calculate the user activity of each interest dimension at each time point; draw a stacked area chart, with the X-axis representing the behavior timeline; the Y-axis representing the user activity of each interest dimension at the corresponding time point; a stacking area corresponds to a stacking area for each interest dimension, and the areas of different interest dimensions are stacked in sequence in the Y-axis direction to form a total area chart; mark the attention attenuation nodes when user behavior jumps between platforms on the stacked area chart.

[0006] Furthermore, based on the attention decay nodes, low-activity and high-activity platforms are identified, and an interest energy conduction network is constructed to quantify the interest kinetic energy transfer relationship between platforms. Through path planning and energy loss calculation, a directional conduction path is generated, and the interest energy distribution is output. The process includes: The process of constructing the interest energy conduction network is as follows: each platform is abstracted as a node in the graph; the connection relationship between platforms is regarded as the edge in the graph; the connection strength between nodes is calculated based on the user's switching frequency and interest similarity between different platforms, and used as the weight of the edge in the graph; the interest energy conduction coefficient is introduced, and the connection strength is multiplied by the interest energy conduction coefficient to obtain the adjusted edge weight in the graph; Based on the constructed interest energy conduction network, the Dijkstra algorithm is used to generate the interest energy conduction path from the high-activity platform to the low-activity platform; During the conduction process, the energy of the interest dimension of the high-activity platform is directed to the low-activity platform according to the interest energy conduction coefficient; an interest energy receiving monitoring point is set on each low-activity platform to record the received interest energy information in real time; at the same time, combined with the length of the interest energy conduction path, the interest energy loss algorithm is used to calculate the final interest energy arrival amount, that is, the actual amount of interest energy reaching the low-activity platform; the interest energy information received by each low-activity platform is integrated to draw an interest energy distribution map, thereby obtaining the energy distribution of each platform in the interest energy conduction network.

[0007] Furthermore, based on the obtained interest energy distribution, the multi-dimensional dynamic characteristics of interest energy are deconstructed from the three dimensions of time, topic, and user sensitivity, and the multi-dimensional dynamic characteristics are mapped to the corresponding subspaces. The process of fusing all subspaces based on the tensor product to generate a three-dimensional interest energy tensor field includes: The gated loop compression unit is used to decompose the user interest energy sequence into time granularity, generate three elastic weight channels for short-term, medium-term and long-term, and quantify the temporal dynamic characteristics of interest energy; the quantum entanglement entropy is calculated based on the topic gravity algorithm to quantify the energy distribution and competitive relationship of each interest topic on different platforms; the user interest sensitivity fingerprint matrix is constructed through dynamic time regularization to quantify the user's sensitivity gradient to different topics; the three subspaces of time elastic weight, topic energy proportion and user sensitivity gradient are fused through tensor product to generate a three-dimensional interest energy tensor field.

[0008] Furthermore, based on the three-dimensional interest energy tensor field, the niche sensitivity function is used to quantify the niche characteristics of user interest energy distribution, resulting in multi-level clustering of users. The competition-symbiosis game model is used to achieve dynamic migration of user hierarchies and confidence-weighted label output. The specific process includes: The niche distance metric is obtained by calculation, and the clustering weight factor is automatically adjusted using the interest energy entropy of the three-dimensional interest energy tensor field; When the interest energy entropy value exceeds the preset dynamic migration threshold, fuzzy boundary clustering is triggered, allowing users to dynamically migrate between different levels; Organize and output user stratification labels, namely high-interest energy active layer labels, medium-interest energy fluctuation layer labels, and low-interest energy dormant layer labels; among them, each user stratification label carries a dynamic confidence weight and is calculated through the payoff function of the competition-symbiosis game model.

[0009] Furthermore, we define the platform interest energy niche index, calculate the position of each platform in the interest energy niche, and rank them. At the same time, the process of screening the optimal platform and platform combination includes: The calculation of each platform's position in the interest energy niche includes the calculation of platform interest energy concentration, platform interest energy complementarity, and platform interest energy radiation range. For platform interest energy concentration, the ratio of each platform's energy share on each interest topic to the total energy share of all platforms on that topic is calculated. For platform interest energy complementarity, the degree of difference between different platforms in interest energy topics is analyzed to quantify the complementarity between platforms. For platform interest energy radiation range, the ability of each platform's interest energy to attract and drive the interests of users on other platforms is evaluated. Construct a non-cooperative game payoff matrix and solve the optimal strategy combination between platforms to maximize the niche index: regard the platform as an intelligent agent in the game, define the strategy space as a combination of interest-themed promotions, set the payoff function as the weighted sum of each niche index, and iterate through the evolutionary stable strategy to screen out the Nash equilibrium strategy; after the evolutionary stable strategy converges, sort the platforms according to the niche index value; output the platform niche ranking results, and identify complementary platform pairs, that is, platform combinations with strong niche complementarity.

[0010] Furthermore, for different levels of users, guidance chains are constructed to optimize the interest energy distribution. The process of finally outputting a weighted user-platform interest energy guidance chain set includes: Z1. Targeting users in the high-interest energy active layer: Analyze the high-interest energy topics of these users on different platforms, determine the resonance points of interest energy between platforms, and generate corresponding guidance chains. Specifically, based on the interest resonance peak detection of the tensor field, find the phase coherence points of the tensor field between complementary pairs of platforms to generate phase coherence guidance chains. Z2. For users in the medium interest energy fluctuation layer: Monitor the changes in interest energy of users in the medium interest energy fluctuation layer on different platforms. When the user's interest energy on one platform decreases, guide them to transfer to a platform with increasing interest energy. Combined with the analysis results of the platform's interest energy radiation range, generate an interest energy migration guidance chain for the user. Specifically, set up a pulse neural navigator to encode the user's interest energy decrease event into a pulse signal. When the user's interest energy falls below the preset trigger threshold, trigger a topological shortcut search within the platform's radiation range. Z3. For users in the dormant layer with low interest energy: select platforms with high interest energy concentration and potential correlation with users' historical interests from the platform's interest energy niche. Based on users' historical behavior data and interest energy distribution, build an interest energy activation guidance chain for users. Specifically, build an interest embryo cultivation mechanism, select dormant topic seeds that match users' historical sensitivity fingerprints from the platform's niche, and generate an interest energy activation guidance chain through a diffusion model. Z4. The guidance chains at different layers are weighted by confidence weights, and the user-platform interest energy guidance chain is output to obtain the final set of each interest energy guidance chain.

[0011] Compared with the existing technology, the beneficial effects of the present invention are as follows: through behavioral timeline alignment and fingerprint mapping, the differences in interest expression across platforms are eliminated, providing a standardized data basis for subsequent analysis; using stacked area charts and attention decay node annotations, the dynamic migration process of user interests between different platforms is intuitively presented; by constructing an interest energy conduction network, the efficiency of interest kinetic energy transfer from high-activity platforms to low-activity platforms is quantified; through path planning and loss calculation, a directional conduction path is generated to guide the precise allocation of interest energy between platforms; through time granularity decomposition and topic energy quantification, the multi-dimensional dynamic change law of interest energy is revealed; based on subspace mapping and tensor product fusion, a three-dimensional interest energy tensor field is constructed to provide a reference for complex interest energy. Interest analysis provides a multi-dimensional perspective; quantifies the distribution of user interest energy through the niche sensitivity function, and realizes the precise division of high, medium and low interest energy levels; utilizes the competition-symbiosis game model and interest energy entropy to realize the dynamic migration of user levels and the output of confidence-weighted labels; quantifies the position of the platform in the niche by defining indicators such as interest energy concentration, complementarity and radiation range; identifies platforms with obvious advantages and platform combinations with strong complementarity based on the ranking of niche indicators, and supports the optimization of platform cooperation strategies; generates different guide chains for users at different levels to optimize the distribution of user interest energy; supports personalized interest recommendations and platform resource allocation by weighted output of user-platform interest energy guide chain sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a module diagram of the artificial intelligence-based cross-platform user traffic diversion and growth promotion system for large companies proposed by the present invention. DETAILED DESCRIPTION

[0013] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the implementation regulations described are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0014] Reference Figure 1 , an AI-based cross-platform user traffic diversion and growth promotion system for large companies, which includes a behavior timeline alignment module, an interest energy conduction module, an interest energy tensor construction module, a user hierarchical clustering module, an ecological niche platform evaluation module, and a hierarchical guide chain generation module; Behavior Timeline Alignment Module: This module stitches user activity records across different platforms into a continuous behavior timeline, using algorithms to uniformly map interest dimensions. It uses stacked area charts to illustrate the interest migration process and annotates the attention decay nodes when user behavior jumps between platforms. Interest Energy Conduction Module: This module identifies low- and high-activity platforms based on attention decay nodes, constructs an interest energy conduction network to quantify the interest kinetic energy transfer relationship between platforms, generates a directional conduction path through path planning and energy loss calculation, and outputs the interest energy distribution. Interest Energy Tensor Construction Module: Based on the acquired interest energy distribution, it deconstructs the multi-dimensional dynamic characteristics of interest energy from the three dimensions of time, topic, and user sensitivity, and maps the multi-dimensional dynamic characteristics to the corresponding subspaces. Based on the tensor product, all subspaces are fused to generate a three-dimensional interest energy tensor field, realizing a unified quantitative expression of user interest status across platforms, time periods, and topics. User-level clustering module: Based on the three-dimensional interest energy tensor field, the niche sensitivity function is used to quantify the niche characteristics of user interest energy distribution, resulting in a multi-level clustering of users, including a high-interest energy active layer, a medium-interest energy fluctuation layer, and a low-interest energy dormant layer. Among them, users in the high-interest energy active layer have high and stable interest energy on multiple platforms, users in the medium-interest energy fluctuation layer have large fluctuations in interest energy between different platforms, and users in the low-interest energy dormant layer have low overall interest energy and are not very interested in most platforms. The competition-symbiosis game model is used to achieve dynamic migration of user levels and confidence-weighted label output. Niche Platform Assessment Module: This module defines platform interest energy niche indicators, including platform interest energy concentration (the ratio of a platform's energy share on a specific interest topic to the total energy share of all platforms on that topic), platform interest energy complementarity (the degree of difference between different platforms in interest energy topics), and platform interest energy radiation range (the ability of a platform's interest energy to attract and drive the interests of users on other platforms). It calculates the position of each platform in the interest energy niche and ranks them; at the same time, it selects the optimal platform and platform combination. Hierarchical guidance chain generation module: For users at different levels, guidance chains are constructed to optimize the interest energy distribution, and finally a weighted output of the user-platform interest energy guidance chain set is generated.

[0015] It should be further explained that in the specific implementation process, the user's operation records on different platforms are spliced into a continuous behavior timeline, and an algorithm is used to uniformly map the interest dimensions. A stacked area chart is used to show the interest migration process, and the attention decay nodes when the user behavior jumps between platforms are marked. The process is as follows: Collect user operation records from various platforms used by users (such as mobile applications, web browsers, desktop software, etc.); these records should include but are not limited to click operations (such as clicking buttons, links, etc.), dwell time (such as the length of time spent on a page or application), and search behavior (such as search keywords entered, search result clicks, etc.); based on the timestamp of each operation record, splice the user's operation records on various platforms into a continuous behavior timeline in chronological order, and mark each operation record with its type (such as click, dwell, search, etc.) on the behavior timeline; extract user behavior features from the user's behavior timeline, such as operation frequency, operation type distribution, dwell time distribution, etc. These features will serve as the basis for behavioral fingerprints; use a behavioral fingerprint alignment algorithm to map user behavior features on different platforms to the same interest dimension; it can be understood that the behavioral fingerprint alignment algorithm extracts user behavior features (such as operation frequency, type distribution, etc.) from multi-platform operation records and maps them to a unified interest dimension space to eliminate platform differences and achieve standardized alignment of user interest features across platforms; define a set of interest dimensions (such as entertainment, shopping, learning, etc.) and map user behavior features to the defined interest dimensions; Based on the aligned behavioral timeline, analyze the user's behavioral jumps between different platforms: record the user's operational behavior when jumping from one platform to another, as well as the changes in interest dimensions before and after the jump; for each time point (or time period) on the continuous behavioral timeline, count the distribution of users' interest dimensions on different platforms (for example, entertainment, shopping, learning, etc.); calculate the user activity of each interest dimension at each time point (or time period) (this activity indicator is obtained based on the extracted user behavior features); draw a stacked area chart, where the X-axis represents the behavioral timeline, corresponding to a specific time point or time period; the Y-axis represents the user activity of each interest dimension at the corresponding time point (or time period); a stacking area corresponds to a stacking area for each interest dimension, and the areas of different interest dimensions are stacked in sequence in the Y-axis direction to form a total area chart; mark the attention decay nodes when users jump between platforms on the stacked area chart: attention decay nodes are manifested as a sudden decrease in the user's activity in a certain interest dimension on a certain platform, or a discontinuous change in the user's activity when jumping from one interest dimension to another; by marking these nodes, a user's interest changes and attention distribution when jumping between different platforms are intuitively displayed.

[0016] It should be further explained that in the specific implementation process, low-activity and high-activity platforms are identified based on attention decay nodes, and an interest energy conduction network is constructed to quantify the interest kinetic energy transfer relationship between platforms. Through path planning and energy loss calculation, a directional conduction path is generated, and the interest energy distribution is output as follows: The process of constructing the interest energy conduction network is as follows: each platform is abstracted as a node in the graph; the connection relationship between platforms is regarded as the edge in the graph; based on the user's switching frequency and interest similarity between different platforms, the connection strength between nodes is calculated and used as the weight of the edge in the graph: the switching frequency is statistically calculated based on the user's platform switching record. The more frequent the user switches, the stronger the correlation between platforms at the user behavior level; the interest similarity uses the cosine similarity algorithm to calculate the similarity between the user's interest feature vectors on different platforms; finally, the connection strength between nodes is obtained by taking the weighted sum of the switching frequency and interest similarity. The larger the connection strength value, the closer the relationship between the nodes. The interest energy conduction coefficient is introduced and calculated using regression analysis. This coefficient is a key parameter for measuring the efficiency of interest energy conduction between different platforms. It is closely related to the correlation between platforms and the difficulty of migrating user interests. The specific process is as follows: Q1. We quantify the correlation between platforms by analyzing user overlap between them. This is done by counting the proportion of users who are active on both platforms to the total number of users on one platform. A higher ratio indicates a stronger correlation between the user groups on the two platforms. We also quantify the difficulty of user interest migration by evaluating the stability of user interests. This is done by analyzing the long-term changes in the distribution of users' interests across platforms. If user interests fluctuate little over time, they are more stable and the migration difficulty is relatively low. Q2. We use the quantified indicators of inter-platform correlation and user interest transfer difficulty as independent variables and the energy conduction coefficient as the dependent variable. We use the multivariate linear regression analysis method to establish a regression equation and obtain the interest energy conduction coefficient: , where represents the interest energy conduction coefficient between node i and node j, is the regression coefficient, They are the inter-platform correlation index and the user interest migration difficulty index. is the error term; Multiply the connection strength by the interest energy conduction coefficient to obtain the weight w of the edge in the adjusted graph; Based on the constructed interest energy conduction network, the specific steps of using the Dijkstra algorithm to generate the interest energy conduction path from the high-activity platform to the low-activity platform are as follows: S1. Suppose there are platforms A, B, C, D, and E, where platform A is a highly active platform, and platforms D and E are low - active platforms. Then the node set V = {A, B, C, D, E}; S2. Set the highly active platform as the starting node j. In this example, j = A; S3. Create an array Dist whose length is equal to the number of platform nodes |V|; Dist[i] represents the shortest path length from the starting node j to other nodes i (here the path length is defined as the number of platforms through which the interest energy is conducted from the starting platform to the target platform); Initially, Dist[j] = 0, indicating that the distance from the starting node to itself is 0; For other nodes i ≠ j, Dist[i] = +∞, indicating that no path from the starting node to these nodes has been found yet; S4. Create an array Prev to record the predecessor nodes of each node on the shortest path; Initially, all elements are empty, that is, Prev[i] = null, i ∈ V; S5. Create a set J to store the nodes for which the shortest paths have been determined; Initially, J = ; S6. From all the nodes that have not been added to the set J, select the node u with the smallest Dist value and add it to the set J. For example, in the initial state, J = , and among all the nodes, Dist[A] = 0 is the smallest, so add node A to the set J, that is, J = {A}; S7. For all adjacent nodes u' of node u, calculate the path length Dist[u] + w(u, u') from the starting node j through node u to the adjacent node u' of u. If Dist[u] + w(u, u') < dist[u'], then update dist[u'] = dist[u] + w(u, u'), and set the predecessor node of the adjacent node u' of u as Prev[u'] = u; Specifically: Suppose the adjacent nodes of node A are B and C, and w(A, B) = 2, w(A, C) = 5. At this time, Dist[A] = 0. Calculate Dist[B] = Dist[A] + w(A, B) = 0 + 2 = 2, Dist[C] = Dist[A] + w(A, C) = 0 + 5 = 5; Since initially Dist[B] = +∞ and Dist[C] = +∞, update Dist[B] = 2, Prev[B] = A; update Dist[C] = 5, Prev[C] = A; S8. Repeat the above steps of selecting the current node and updating the distances of adjacent nodes until the set J contains all nodes, that is, the shortest paths from the starting node j to all other nodes have been found; Specifically: among the nodes B, C, D, and E that are not added to set J, Dist[B]=2 is the smallest, and node B is added to set J, that is, J={A,B}; update the distance of node B's adjacent node (assuming it is D), if w(B,D)=3, then Dist[D]=Dist[B]+w(B,D)=2+3=5; because Dist[D]=+∞ at the beginning, update Dist[D]=5, Prev[D]=B; continue to select the current node, assuming that node C is selected next to join set J, that is, J={A,B,C}; update the distance of node C's adjacent node (assuming it is E), if w(C,E)=4, then Dist[E]=Dist[C]+w(C,E)=5+4=9; because Dist[E]=+∞ at the beginning, update Dist[E]=9, Prev[E]=C; repeat this process until all nodes are added to set J; S9. Assuming that the low-activity platforms are D and E, generate interest energy conduction paths from the high-activity platforms A to D and A to E respectively. Here, the generation of the interest energy conduction path from A to D is taken as an example; S10. Set the low-activity platform as the target node; starting from the target node D, trace back to the starting node A according to the predecessor node array Prev; because Prev[D]=B and Prev[B]=A, the conduction path is A→B→D; similarly, if an energy conduction path of interest is generated from A to E, according to Prev[E]=C and Prev[C]=A, the energy conduction path of interest is A→C→E; During the conduction process, the energy of the interest dimension of the high-activity platform is directed to the low-activity platform according to the interest energy conduction coefficient; an interest energy receiving monitoring point is set on each low-activity platform to record the received interest energy information in real time, where the interest energy information refers to the resources or influence of the platform that can influence the user's interest tendency, such as the energy contained in the hot topics and high-quality content on the platform that can stimulate the user's interest; at the same time, combined with the length of the interest energy conduction path, the interest energy loss algorithm is used to calculate the final interest energy arrival amount, that is, the amount of interest energy actually reaching the low-activity platform (the interest energy conduction path length refers to the number of platforms that the interest energy passes through when it is transmitted from the starting platform to the target platform. The longer the path, the greater the loss of interest energy that may occur during the conduction process; the path length is used as a quantitative indicator, for example, a path length of 1 indicates direct conduction from one platform to another adjacent platform, and a path length of 2 indicates conduction through an intermediate platform, etc.; the expression of the interest energy loss algorithm is: , where is the amount of interest energy reached, is the initial conduction amount of energy of interest, is the conduction path length of the energy of interest, is the loss factor ( By performing regression analysis on the arrival amount of energy of interest under different path lengths, we can fit The best estimate of the energy consumption is used to measure the impact of factors such as user activity, content quality, and platform scale on energy loss. The interest energy information received by each low-activity platform is integrated to draw an interest energy distribution map, thereby obtaining the energy distribution of each platform in the interest energy conduction network.

[0017] It should be further explained that in the specific implementation process, based on the obtained interest energy distribution, the multi-dimensional dynamic characteristics of interest energy are deconstructed from the three dimensions of time, topic, and user sensitivity, and the multi-dimensional dynamic characteristics are mapped to the corresponding subspaces; the process of fusing all subspaces based on the tensor product to generate a three-dimensional interest energy tensor field is as follows: The time dimension (short-term, medium-term, and long-term interest energy trends), the interest energy theme dimension (the energy proportion of different interest themes on each platform), and the user individual difference dimension (user sensitivity to different types of interest energy) are introduced. Short-term, medium-term, and long-term interest energy are mapped into three orthogonal subspaces of the tensor field, where each element is defined as follows: For the time dimension, the user interest energy is compressed into three channels according to the time granularity: short-term (e.g., 1 hour), medium-term (e.g., 1 day), and long-term (e.g., 1 week). The value of each channel represents the elastic weight of the interest energy in that period, generating an interest energy tensor with time elastic weights. For the topic dimension, the topic gravity algorithm is used to calculate the quantum entanglement entropy to obtain the interest energy ratio of each topic of interest, that is, , where Indicates the proportion of user's topic interest energy. The larger the value, the higher the user's attention to the topic. Represents the probability distribution of user behavior on the topic; For the user dimension, a sensitivity fingerprint matrix is constructed. The distance between the user's historical interest energy curve and the platform average curve is calculated through dynamic time warping to generate the user's sensitivity gradient to different topics: , where Indicates the user's sensitivity gradient to the topic, which is used to reflect the sensitivity of the user's interest energy over time. Represents the user's historical interest energy time series, represents the mean sequence of all user interest energy time series, The dynamic time warping distance (DTW) is used to measure the similarity between a user's interest curve and the group average curve. The sensitivity gradient is constrained to the range [0, 1] by normalizing the DTW distance (dividing it by the maximum DTW value). A larger value indicates a more sensitive user to changes in interest in the topic. The three subspaces are fused through tensor product to output a three-dimensional energy tensor field of interest , where each element in the three-dimensional energy tensor of interest is Indicates the energy density of the user's interest in the topic during a certain period of time.

[0018] It should be further explained that in the specific implementation process, based on the three-dimensional interest energy tensor field, the niche sensitivity function is used to quantify the niche characteristics of the user interest energy distribution, and the multi-level clustering of users is obtained. The competition-symbiosis game model is used to realize the dynamic migration of user levels and the confidence-weighted label output process as follows: The niche distance metric is obtained by calculating : , where is the clustering weight factor, and the interest energy entropy of the three-dimensional interest energy tensor field is obtained by Automatically adjust the clustering weight factor. It can be understood that the interest energy entropy is obtained by quantifying the unevenness and uncertainty of the user interest energy distribution in the three-dimensional interest energy tensor field (such as the difference in interest energy distribution at different times, platforms, and topics), and is used to measure the complexity of the interest energy distribution; are the interest energy gradients of user 1 and user 2 respectively (reflecting the short-term fluctuation intensity of interest energy), and , where t is the time variable, indicating the rate of change of interest energy over time; KL divergence captures the differences in user interest distribution; The topic probability distributions of user 1 and user 2 respectively; MaxPool is used to extract interest energy mutation features and enhance sensitivity to mutation behavior; When the interest energy entropy exceeds the preset dynamic migration threshold, fuzzy boundary clustering is triggered, allowing users to dynamically migrate between different levels. The dynamic migration threshold is used to quantify the degree of chaos in the user's interest energy distribution in the three-dimensional interest energy tensor field (measured by the interest energy entropy). When the entropy exceeds this threshold, it indicates that the user's interest characteristics tend to be fuzzy, triggering the fuzzy boundary clustering mechanism to allow users to dynamically migrate between different levels, thereby adapting to the dynamic changes in the interest energy distribution. Organize and output user hierarchical labels ;in, They are high interest energy active layer, medium interest energy fluctuation layer and low interest energy dormant layer. Each user layer label carries a dynamic confidence weight. :Calculation of the profit function of the competition-symbiosis game model , where ESS is the profit value of the evolutionary stable strategy, Layer labels for other users and distinguish them from the current user's layer labels , exp is the exponential function, which is used to map the ESS return into a probability distribution; It is understandable that the niche sensitivity function outputs the niche distance matrix between users as the input of non-equilibrium clustering; dynamic migration is achieved by real-time updating of clustering weight factors and dynamic confidence weights. For example, when the user's interest energy fluctuation exceeds the threshold, its label carries the dynamic confidence weight. Increase.

[0019] It should be further explained that in the specific implementation process, the platform interest energy niche indicator is defined, the position of each platform in the interest energy niche is calculated, and the platform is ranked; at the same time, the process of screening out the optimal platform and platform combination is as follows: The calculation of each platform's position in the interest energy niche includes the calculation of platform interest energy concentration, platform interest energy complementarity, and platform interest energy radiation range. For platform interest energy concentration, the ratio of each platform's energy share on each interest topic to the total energy share of all platforms on that topic is calculated. The higher the ratio, the more concentrated the platform's niche on that interest topic. For platform interest energy complementarity, the complementarity between platforms is quantified by analyzing the degree of difference between different platforms on interest energy topics. The greater the difference, the stronger the complementarity. For platform interest energy radiation range, the ability of each platform's interest energy to attract and drive the interests of users on other platforms is evaluated. The wider the radiation range, the greater the influence of the platform: Concentration of interest energy on the platform : , where A collection of topics of interest, For a comprehensive overview of all topics of interest, Indicates the interest energy weight of the topic on the platform; Complementarity of platform interest and energy : , where is the tensor product of the interest energy of platform 1 and platform 2, which captures the nonlinear correlation of cross-platform topic distribution through tensor operations. are the L2 norm of the energy tensor of interest on platform 1 and the L2 norm of the energy tensor of interest on platform 2, respectively, which are used to measure the tensor sparsity and energy intensity; Energy radiation range of platform interest : , where For all users, is the activation function, is the cosine similarity, is the radiation range weight, is the platform interest energy tensor, is the user's interest energy tensor; Construct a non-cooperative game payoff matrix and solve the optimal strategy combination (i.e., interest theme promotion combination) between platforms to maximize the niche index: consider the platform as an intelligent agent in the game, define the strategy space as the interest theme promotion combination, and set the payoff function as the weighted sum of each niche index. Through iterative solution of evolutionary stable strategy, screen out the Nash equilibrium strategy, that is, the platform combination whose payoff cannot be improved by unilaterally changing the strategy; after the evolutionary stable strategy converges, rank the platforms according to the niche index value; output the platform niche ranking results. , and identify complementary platform pairs , that is, a platform combination with strong niche complementarity; where n represents the number of platform niche rankings, All are platform indexes, where complementary platform pairs satisfy Greater than the preset complementary synergy threshold; it can be understood that the complementary synergy threshold is used to quantify the strength of niche complementarity between platforms. When the niche complementarity index value of the combination of two platforms exceeds this threshold, it is determined that they have significant synergistic effects and can be identified as a complementary platform pair for optimizing the combination of interest topic promotion strategies.

[0020] It should be further explained that, in the specific implementation process, for different levels of users, guidance chains are constructed to optimize the interest energy distribution. The final weighted output process of the user-platform interest energy guidance chain set is as follows: Z1. Targeting users in the high-interest energy active layer: Analyze the high-interest energy topics of users in the high-interest energy active layer on different platforms, identify the resonance points of interest energy between platforms, and generate corresponding guidance chains to recommend platforms with strong ecological niche complementarity to users; specifically: Based on the detection of the resonance peak of interest in the tensor field, the phase coherence points of the tensor field are found between the complementary pairs of platforms, that is, , generating a phase-coherent guiding chain : , where Expressing support for the platform The energy tensor of interest is Fourier transformed; Re is used to take the real part of the complex number, that is, the real part of the resonant frequency of the energy of interest; To match the resonant frequency of the energy of interest; Z2. For users in the medium interest energy fluctuation layer: Monitor the interest energy changes of users in the medium interest energy fluctuation layer on different platforms. When the user's interest energy on a platform decreases, guide them to transfer to a platform with increasing interest energy. Combined with the analysis results of the platform's interest energy radiation range, generate an interest energy migration guidance chain for the user and recommend platforms with high interest energy radiation capabilities. Specifically: A pulse neural navigator is set up to encode the user interest energy drop event into a pulse signal. When the user interest energy is lower than the preset trigger threshold, the topological shortcut search within the platform radiation range is triggered; the weight of the guide chain node is , where For the target platform, is the interest energy tensor of the target platform. It can be understood that the trigger threshold is used to quantify the degree of interest energy attenuation of users in the middle interest energy fluctuation layer on the current platform. When the user's interest energy is lower than this threshold, it is determined that their interest energy is no longer sufficient to maintain the current platform's activity, triggering a topological shortcut search based on the platform's radiation range to guide users to migrate to platforms with rising interest energy. Z3. For users in the dormant layer with low interest energy: select platforms with high interest energy concentration and potential correlation with users' historical interests from the platform interest energy niche. Based on users' historical behavior data and interest energy distribution, build an interest energy activation guidance chain for users; specifically: Constructing an interest embryo cultivation mechanism: Screening dormant topic seeds that match the user's historical sensitivity fingerprint from the platform's ecological niche, and generating interest energy activation guide chains through a diffusion model , where Dormant topic seeds screened from platform niches (e.g., low-frequency but high-potential interest topics in users' historical behavior); Z4, the guide chain of different layers through confidence weight Weighted, output user-platform interest energy guidance chain, and obtain the final set of each interest energy guidance chain : , where each energy-directed chain of interest contains a dynamic attenuation factor ; m represents the number of energy-directed chains of interest.

[0021] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0022] It should be understood that determining B based on A does not mean determining B based solely on A. B can also be determined based on A and / or other information.

[0023] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0024] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An AI-based cross-platform user traffic generation and growth promotion system for large companies, featuring: It includes a behavior timeline alignment module, an interest energy conduction module, an interest energy tensor construction module, a user hierarchical clustering module, an ecological niche platform evaluation module, and a hierarchical guide chain generation module; Behavior Timeline Alignment Module: This module stitches user activity records across different platforms into a continuous behavior timeline, using algorithms to uniformly map interest dimensions. It uses stacked area charts to illustrate the interest migration process and annotates the attention decay nodes when user behavior jumps between platforms. Interest Energy Conduction Module: This module identifies low- and high-activity platforms based on attention decay nodes, constructs an interest energy conduction network to quantify the interest kinetic energy transfer relationship between platforms, generates a directional conduction path through path planning and energy loss calculation, and outputs the interest energy distribution. Interest Energy Tensor Construction Module: Based on the acquired interest energy distribution, it deconstructs the multi-dimensional dynamic characteristics of interest energy from the three dimensions of time, topic, and user sensitivity, and maps the multi-dimensional dynamic characteristics to the corresponding subspaces. It then fuses all subspaces based on the tensor product to generate a three-dimensional interest energy tensor field. User-level clustering module: Based on the three-dimensional interest energy tensor field, the niche sensitivity function is used to quantify the niche characteristics of user interest energy distribution, and multi-level clustering of users is obtained. The competition-symbiosis game model is used to achieve dynamic migration of user levels and confidence-weighted label output. Niche platform evaluation module: defines the platform interest energy niche indicators, calculates the position of each platform in the interest energy niche, and ranks them; at the same time, screens out the optimal platform and platform combination; Hierarchical guidance chain generation module: For users at different levels, guidance chains are constructed to optimize the interest energy distribution, and finally a weighted output of the user-platform interest energy guidance chain set is generated.

2. The AI-based cross-platform user traffic diversion and growth promotion system for large companies according to claim 1 is characterized by: The process involves stitching user activity records from different platforms into a continuous timeline, applying algorithms to uniformly map interest dimensions, using stacked area charts to illustrate interest migration, and noting the attention decay nodes when users jump between platforms. Collect user operation records from various platforms used by the user; based on the timestamp of each operation record, splice the user operation records on each platform into a continuous behavior timeline in chronological order, and mark the type of each operation record on the behavior timeline; extract user behavior features from the user behavior timeline, these features will serve as the basis for behavioral fingerprints; use the behavioral fingerprint alignment algorithm to map user behavior features on different platforms to the same interest dimension; define a set of interest dimensions, and map user behavior features to the defined interest dimensions; Based on the aligned behavior timeline, analyze the user behavior jumps between different platforms: record the user's operation behavior when jumping from one platform to another, as well as the changes in interest dimensions before and after the jump; for each time point on the continuous behavior timeline, count the distribution of users' interest dimensions on different platforms; calculate the user activity of each interest dimension at each time point; draw a stacked area chart, with the X-axis representing the behavior timeline; the Y-axis representing the user activity of each interest dimension at the corresponding time point; a stacking area corresponds to a stacking area for each interest dimension, and the areas of different interest dimensions are stacked in sequence in the Y-axis direction to form a total area chart; mark the attention attenuation nodes when user behavior jumps between platforms on the stacked area chart.

3. The AI-based cross-platform user traffic diversion and growth promotion system for large companies according to claim 1 is characterized by: Based on attention decay nodes, low-activity and high-activity platforms are identified, and an interest energy conduction network is constructed to quantify the interest kinetic energy transfer relationship between platforms; The process of generating a directional conduction path through path planning and energy loss calculation and outputting the energy distribution of interest includes: The process of constructing the interest energy conduction network is as follows: each platform is abstracted as a node in the graph; the connection relationship between platforms is regarded as the edge in the graph; the connection strength between nodes is calculated based on the user's switching frequency and interest similarity between different platforms, and used as the weight of the edge in the graph; the interest energy conduction coefficient is introduced, and the connection strength is multiplied by the interest energy conduction coefficient to obtain the adjusted edge weight in the graph; Based on the constructed interest energy conduction network, the Dijkstra algorithm is used to generate the interest energy conduction path from the high-activity platform to the low-activity platform; During the conduction process, the energy of the interest dimension of the high-activity platform is directed to the low-activity platform according to the interest energy conduction coefficient; an interest energy receiving monitoring point is set on each low-activity platform to record the received interest energy information in real time; at the same time, combined with the length of the interest energy conduction path, the interest energy loss algorithm is used to calculate the final interest energy arrival amount, that is, the actual amount of interest energy reaching the low-activity platform; the interest energy information received by each low-activity platform is integrated to draw an interest energy distribution map, thereby obtaining the energy distribution of each platform in the interest energy conduction network.

4. The AI-based cross-platform user traffic diversion and growth promotion system for large companies according to claim 1 is characterized by: Based on the obtained interest energy distribution, the multi-dimensional dynamic characteristics of interest energy are deconstructed from three dimensions: time, topic, and user sensitivity, and the multi-dimensional dynamic characteristics are mapped to the corresponding subspace; The process of fusing all subspaces based on tensor products to generate a three-dimensional energy tensor field of interest includes: The gated loop compression unit is used to decompose the user interest energy sequence into time granularity, generate three elastic weight channels for short-term, medium-term and long-term, and quantify the temporal dynamic characteristics of interest energy; the quantum entanglement entropy is calculated based on the topic gravity algorithm to quantify the energy distribution and competitive relationship of each interest topic on different platforms; the user interest sensitivity fingerprint matrix is constructed through dynamic time regularization to quantify the user's sensitivity gradient to different topics; the three subspaces of time elastic weight, topic energy proportion and user sensitivity gradient are fused through tensor product to generate a three-dimensional interest energy tensor field.

5. The AI-based cross-platform user traffic diversion and growth promotion system for large companies according to claim 1 is characterized by: Based on the three-dimensional interest energy tensor field, the niche sensitivity function is used to quantify the niche characteristics of user interest energy distribution, and the multi-level clustering of users is obtained; The specific process of implementing dynamic migration of user levels and confidence-weighted label output using the competition-symbiosis game model includes: The niche distance metric is obtained by calculation, and the clustering weight factor is automatically adjusted using the interest energy entropy of the three-dimensional interest energy tensor field; When the interest energy entropy value exceeds the preset dynamic migration threshold, fuzzy boundary clustering is triggered, allowing users to dynamically migrate between different levels; Organize and output user stratification labels, namely high-interest energy active layer labels, medium-interest energy fluctuation layer labels, and low-interest energy dormant layer labels; among them, each user stratification label carries a dynamic confidence weight and is calculated through the payoff function of the competition-symbiosis game model.

6. The AI-based cross-platform user traffic diversion and growth promotion system for large companies according to claim 1 is characterized by: Define the platform interest energy niche indicator, calculate the position of each platform in the interest energy niche, and rank them; at the same time, the process of screening the optimal platform and platform combination includes: The calculation of each platform's position in the interest energy niche includes the calculation of platform interest energy concentration, platform interest energy complementarity, and platform interest energy radiation range. For platform interest energy concentration, the ratio of each platform's energy share on each interest topic to the total energy share of all platforms on that topic is calculated. For platform interest energy complementarity, the degree of difference between different platforms in interest energy topics is analyzed to quantify the complementarity between platforms. For platform interest energy radiation range, the ability of each platform's interest energy to attract and drive the interests of users on other platforms is evaluated. Construct a non-cooperative game payoff matrix and solve the optimal strategy combination between platforms to maximize the niche index: regard the platform as an intelligent agent in the game, define the strategy space as a combination of interest-themed promotions, set the payoff function as the weighted sum of each niche index, and iterate through the evolutionary stable strategy to screen out the Nash equilibrium strategy; after the evolutionary stable strategy converges, sort the platforms according to the niche index value; output the platform niche ranking results, and identify complementary platform pairs, that is, platform combinations with strong niche complementarity.

7. The AI-based cross-platform user traffic diversion and growth promotion system for large companies according to claim 1 is characterized by: For different levels of users, guide chains are constructed to optimize the distribution of interest energy. The process of finally outputting a weighted set of user-platform interest energy guide chains includes: Z1. Targeting users in the high-interest energy active layer: Analyze the high-interest energy topics of these users on different platforms, determine the resonance points of interest energy between platforms, and generate corresponding guidance chains. Specifically, based on the interest resonance peak detection of the tensor field, find the phase coherence points of the tensor field between complementary pairs of platforms to generate phase coherence guidance chains. Z2. For users in the medium interest energy fluctuation layer: Monitor the changes in interest energy of users in the medium interest energy fluctuation layer on different platforms. When the user's interest energy on one platform decreases, guide them to transfer to a platform with increasing interest energy. Combined with the analysis results of the platform's interest energy radiation range, generate an interest energy migration guidance chain for the user. Specifically, set up a pulse neural navigator to encode the user's interest energy decrease event into a pulse signal. When the user's interest energy falls below the preset trigger threshold, trigger a topological shortcut search within the platform's radiation range. Z3. For users in the dormant layer with low interest energy: select platforms with high interest energy concentration and potential correlation with users' historical interests from the platform's interest energy niche. Based on users' historical behavior data and interest energy distribution, build an interest energy activation guidance chain for users. Specifically, build an interest embryo cultivation mechanism, select dormant topic seeds that match users' historical sensitivity fingerprints from the platform's niche, and generate an interest energy activation guidance chain through a diffusion model. Z4. The guidance chains at different layers are weighted by confidence weights, and the user-platform interest energy guidance chain is output to obtain the final set of each interest energy guidance chain.

Images