A Front-End Tracking Recommendation Method and System Based on Artificial Intelligence

Abstract

The application is suitable for the technical field of Internet, and provides a front-end point embedding recommendation method and system based on artificial intelligence, which comprises the following steps: collecting historical user session data, and extracting user identification, sequence of interaction events and time stamp of event occurrence in the historical user session data; constructing a user behavior directed graph network based on the historical user session data; analyzing each directed edge in the user behavior directed graph network by using an artificial intelligence algorithm to obtain a topological key edge; and outputting a front-end interaction position corresponding to the topological key edge as a point embedding recommendation position set. The application changes the existing single-point monitoring mode into intelligent judgment from a user interaction network path, and can systematically find and monitor front-end interaction positions which are more meaningful for product business smoothness and stability.

Description

Technical Field

[0001] This invention relates to the field of Internet technology, specifically to a front-end tracking recommendation method and system based on artificial intelligence. Background Technology

[0002] Front-end event tracking refers to setting up data collection points on web pages or applications using technical means to record user actions (such as clicks and browsing). This data can be used to analyze user behavior paths and evaluate product functionality.

[0003] Currently, the mainstream data tracking methods mainly include manual code tracking and automated full tracking. Manual tracking relies on developers or product personnel to manually determine where to insert data collection code based on business experience. This method is labor-intensive and prone to omissions. Automated full tracking, on the other hand, collects data by automatically listening to all interactive elements. Although it has broad coverage, it generates massive amounts of low-value-density data, which still needs to be filtered later and is difficult to accurately capture complex and interconnected business logic.

[0004] Whether manual or automated, existing event tracking recommendation logic largely remains at the point level, focusing only on the independent exposure and click-through rates of individual pages or buttons. However, user interactions are a continuous process. Traditional methods generally overlook the connections between different interaction points during browsing. Therefore, this paper proposes an AI-based front-end event tracking recommendation method and system to address this issue. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a front-end tracking recommendation method and system based on artificial intelligence to solve the problems existing in the above-mentioned background technology.

[0006] This invention is implemented as follows: a front-end tracking recommendation method based on artificial intelligence, the method comprising the following steps:

[0007] Collect historical user session data and extract user identifiers, sequences of interaction events, and timestamps of event occurrences. The interaction events include page visits and element clicks.

[0008] A directed graph network of user behavior is constructed based on historical user session data. The directed graph network of user behavior uses interaction events as nodes and the continuous occurrence relationship between interaction events as directed edges.

[0009] The topological key edge is obtained by analyzing each directed edge in the directed graph network of user behavior using artificial intelligence algorithms. The topological key edge is a directed edge that connects different high-density interactive sub-communities.

[0010] Output the front-end interaction locations corresponding to the topological key edges as a set of recommended tracking locations.

[0011] As a further aspect of the present invention: the step of constructing a directed graph network of user behavior based on historical user session data specifically includes:

[0012] The interaction events within a single session are sorted according to the timestamp, and the session is constructed as a session behavior subgraph by defining two adjacent interaction events as a directed edge.

[0013] All session behavior subgraphs are generalized into nodes, and page URLs and element clicks are mapped to logical page identifiers and semantic component type identifiers, respectively, to improve the recognizability of the directed edges in different sessions.

[0014] Assign an initial weight to each directed edge;

[0015] Based on user identifiers, multiple session behavior subgraphs generated by the same user are merged into nodes and edges, and the directed graph network of user behavior is obtained by accumulating the weights of the same directed edges.

[0016] As a further aspect of the present invention: the step of configuring initial weights for each directed edge specifically includes:

[0017] The page hierarchy depth of each interactive event node is determined based on the site information architecture, and the hierarchy span coefficient of each directed edge is calculated. The hierarchy span coefficient is calculated based on the page hierarchy depth of the first and last nodes of the directed edge.

[0018] The path concentration coefficient is calculated based on the frequency of the occurrence of a directed edge in all session behavior subgraphs and the total number of outgoing edges from its starting node.

[0019] The comprehensive interaction weight is calculated for each directed edge based on the hierarchical span coefficient and the path concentration coefficient. The comprehensive interaction weight is used to characterize the difficulty of the interaction path corresponding to the directed edge occurring.

[0020] The calculated comprehensive interaction weights are used as the initial weights.

[0021] As a further aspect of the present invention: the step of analyzing each directed edge in the directed graph network of user behavior using artificial intelligence algorithms to obtain the topological key edge specifically includes:

[0022] The directed graph network of user behavior is input into the artificial intelligence model, and the dynamic attention coefficient is calculated for each directed edge through the self-attention mechanism;

[0023] The nodes of the directed graph network of user behavior are clustered using a community detection algorithm to divide them into multiple high-density interactive sub-communities. Each directed edge is determined to be a cross-region edge, where the first and last nodes belong to two different high-density interactive sub-communities respectively. The high-density interactive sub-communities have dense internal connections and sparse external connections.

[0024] The attention coefficient of each directed edge and the Boolean attribute indicating whether it is a cross-region edge are combined to form a joint feature vector;

[0025] The joint feature vector is input into a lightweight edge classification neural network to obtain the probability score of each edge as a topological key edge, and directed edges with probability scores higher than a preset threshold are identified as topological key edges.

[0026] As a further aspect of the present invention: the step of outputting the front-end interaction positions corresponding to the topological key edges as a set of recommended tracking points specifically includes:

[0027] For each topological key edge, locate its first and last nodes in the directed graph network of user behavior, and extract the interaction events associated with the topological key edge to form a sequence of directly related events;

[0028] Parse the attribute information of the interaction event corresponding to each node in the directly related event sequence, and generate business description text describing the key edge of the topology based on logical relationships;

[0029] Associate and bind the front-end interaction location and business description text corresponding to the key edge of the topology to obtain the event tracking recommendation record;

[0030] All the recommended tracking points are integrated to form the recommended tracking point location set.

[0031] As a further aspect of the present invention, the method further includes:

[0032] Based on the user behavior directed graph network, the structural index of the topological key edge is obtained by quantifying the removal of a single topological key edge through simulation.

[0033] A set of failed interaction sessions is simulated by generating a virtual user behavior sequence using a random walk algorithm and blocking paths that pass through topological critical edges in the sequence.

[0034] In a directed graph network of user behavior, multiple target nodes are selected, and a decay index is obtained by calculating the difference in the proportion of sessions that reach any target node in the set of normal interaction and failed interaction sessions.

[0035] The structural index and attenuation index are fused together to obtain a comprehensive score for each topological key edge, and the recommended location set of embedding points is prioritized according to the comprehensive score.

[0036] Another objective of this invention is to provide a front-end tracking recommendation system based on artificial intelligence, the system comprising:

[0037] The session acquisition module is used to collect historical user session data and extract user identifiers, sequences of interaction events, and timestamps of event occurrences. The interaction events include page visits and element clicks.

[0038] The network modeling module is used to construct a directed graph network of user behavior based on historical user session data. The directed graph network of user behavior uses interactive events as nodes and the continuous occurrence relationship between interactive events as directed edges.

[0039] The intelligent filtering module is used to analyze each directed edge in the directed graph network of user behavior using artificial intelligence algorithms to obtain the topological key edge, which is a directed edge connecting different high-density interactive sub-communities;

[0040] The event tracking output module is used to output the front-end interaction locations corresponding to the topological key edges as a set of recommended event tracking locations.

[0041] As a further aspect of the present invention: the network modeling module includes:

[0042] The timing construction unit is used to sort the interaction events within a single session according to the timestamp, and construct the session as a session behavior subgraph by defining two adjacent interaction events as a directed edge.

[0043] The semantic generalization unit is used to perform node generalization processing on all session behavior subgraphs, mapping page URLs and element clicks to logical page identifiers and semantic component type identifiers, respectively, to improve the recognizability of the directed edges in different sessions.

[0044] The weight configuration unit is used to configure the initial weight for each directed edge;

[0045] The graph fusion unit is used to merge nodes and edges of multiple session behavior subgraphs generated by the same user based on the user identifier, and obtain the user behavior directed graph network by accumulating the weights of the same directed edges.

[0046] As a further aspect of the present invention: the intelligent screening module includes:

[0047] The attention evaluation unit is used to input the directed graph network of user behavior into the artificial intelligence model and calculate the dynamic attention coefficient for each directed edge through the self-attention mechanism.

[0048] The community identification unit is used to cluster the nodes of the directed graph network of user behavior using the community discovery algorithm, divide it into multiple high-density interactive sub-communities, and determine whether each directed edge is a cross-region edge. The cross-region edge is where the first and last nodes belong to two different high-density interactive sub-communities respectively. The high-density interactive sub-communities have dense internal connections and sparse external connections.

[0049] The feature fusion unit is used to combine the attention coefficients of each directed edge and the Boolean attribute indicating whether it is a cross-region edge to form a joint feature vector;

[0050] The key determination unit is used to input the joint feature vector into a lightweight edge classification neural network, obtain the probability score of each edge as a topological key edge, and determine the directed edges with probability scores higher than a preset threshold as topological key edges.

[0051] As a further aspect of the present invention: the system further includes a data tracking evaluation module, which includes:

[0052] The structural analysis unit is used to quantify the structural index of a single topological key edge based on the user behavior directed graph network by simulating the removal of a single topological key edge.

[0053] The behavior simulation unit is used to generate virtual user behavior sequences using a random walk algorithm, and to simulate a set of failed interaction sessions by blocking paths through topological key edges in the sequence.

[0054] The impact analysis unit is used to select multiple preset target nodes in the directed graph network of user behavior, and obtain the attenuation index by calculating the difference in the proportion of sessions that reach any target node in the set of normal interaction and failed interaction sessions.

[0055] The priority decision unit is used to fuse the structural index and the attenuation index to obtain a comprehensive score for each topological key edge, and to prioritize the set of recommended embedding locations based on the comprehensive score.

[0056] Compared with the prior art, the beneficial effects of the present invention are:

[0057] This invention establishes a directed graph network of user behavior by collecting historical session data. This network can represent continuous user operations, overcoming the limitations of focusing only on individual interaction points. Then, artificial intelligence algorithms are used to analyze the directed graph network, identifying key topological edges that play a connecting role in the overall user behavior flow but are obscured by low-frequency clicks. This transforms the tracking and recommendation approach from a broad-based or hotspot-focused approach to a global one, directly mapping these key topological edges to front-end tracking locations. This not only improves the automation and logical consistency of tracking configuration but also ensures that the collected data directly reflects the user's browsing process and critical points in the system. In summary, this invention transforms existing single-point monitoring methods into intelligent judgment based on user interaction network paths, systematically discovering and monitoring front-end interaction locations that are more significant for the smoothness and stability of product operations. Attached Figure Description

[0058] Figure 1 This is a flowchart of a front-end tracking recommendation method based on artificial intelligence.

[0059] Figure 2 This is a flowchart of constructing a directed graph network of user behavior in a front-end tracking recommendation method based on artificial intelligence.

[0060] Figure 3 This is a flowchart illustrating the initial weight configuration for each directed edge in an AI-based front-end tracking recommendation method.

[0061] Figure 4 This is a flowchart illustrating the analysis of each directed edge in a directed graph network of user behavior in an AI-based front-end tracking and recommendation method.

[0062] Figure 5 This is a flowchart illustrating how a front-end event tracking recommendation method based on artificial intelligence outputs the front-end interaction locations corresponding to topological key edges as a set of event tracking recommendation locations.

[0063] Figure 6 This is a flowchart illustrating the priority ranking of the set of recommended event tracking locations in an AI-based front-end event tracking recommendation method.

[0064] Figure 7 This is a schematic diagram of the structure of a front-end tracking recommendation system based on artificial intelligence.

[0065] Figure 8 This is a schematic diagram of the network modeling module in a front-end tracking recommendation system based on artificial intelligence.

[0066] Figure 9 This is a schematic diagram of the intelligent filtering module in a front-end tracking recommendation system based on artificial intelligence.

[0067] Figure 10 This is a schematic diagram of the structure of the event tracking evaluation module in an AI-based front-end event tracking recommendation system. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0069] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0070] like Figure 1 As shown in the figure, this embodiment of the invention provides a front-end tracking recommendation method based on artificial intelligence, the method including the following steps:

[0071] S100, collect historical user session data and extract user identifiers, sequences of interaction events, and timestamps of event occurrences, including page visits and element clicks;

[0072] S200, Construct a directed graph network of user behavior based on historical user session data. The directed graph network of user behavior uses interactive events as nodes and the continuous occurrence relationship between interactive events as directed edges.

[0073] S300: The topological key edge is obtained by analyzing each directed edge in the directed graph network of user behavior using artificial intelligence algorithms. The topological key edge is a directed edge that connects different high-density interactive sub-communities.

[0074] S400 outputs the front-end interaction locations corresponding to the topological key edges as a set of recommended tracking locations.

[0075] It should be noted that the sequential relationship between the interactive events refers to a directional link formed within a single user session, where two interactive events occurring consecutively (such as clicking a button on a page and then navigating to another page) are associated based on their timestamps. It is not a simple parallel arrangement of events, but rather uses directed edges—a graph theory element—to characterize discrete user actions as a continuous chain of behaviors reflecting their decision-making logic and operational habits.

[0076] In this embodiment of the invention, a directed graph network of user behavior is established by collecting historical session data. This network can represent continuous user operations, overcoming the limitation of focusing only on a single interaction point. Then, artificial intelligence algorithms are used to analyze the directed graph network, thereby identifying key topological edges that play a connecting role in the overall user behavior flow but are obscured by low-frequency clicks. This transforms the tracking and recommendation approach from a broad-based or hotspot-focused approach to a global one, directly mapping these key topological edges to front-end tracking locations. This not only improves the automation and logical consistency of tracking configuration but also ensures that the collected data directly reflects the user's browsing process and critical points in the system. In summary, this invention transforms the existing single-point monitoring method into intelligent judgment based on user interaction network paths, systematically discovering and monitoring front-end interaction locations that are more significant to the smoothness and stability of product operations.

[0077] like Figure 2 As shown in the preferred embodiment of the present invention, the step of constructing a directed graph network of user behavior based on historical user session data specifically includes:

[0078] S201, sort the interaction events within a single session according to the timestamp, and construct the session as a session behavior subgraph by defining two adjacent interaction events as a directed edge;

[0079] S202, perform node generalization processing on all session behavior subgraphs, and map page URLs and element clicks to logical page identifiers and semantic component type identifiers, respectively, to improve the identifiability of the directed edges in different sessions;

[0080] S203, Configure initial weights for each directed edge;

[0081] S204, based on the user identifier, merge the nodes and edges of multiple session behavior subgraphs generated by the same user, and obtain the user behavior directed graph network by accumulating the weights of the same directed edges.

[0082] In this embodiment of the invention, a specific example is used to illustrate the construction of a directed graph network for user behavior. For instance, user A is shopping online. This session generates three interaction events in sequence: "visiting the homepage," "browsing the product list page," and "clicking a product." Therefore, based on the timestamp order, two directed edges can be obtained (visiting the homepage—browsing the product list page and browsing the product list page—clicking a product), thus forming a session behavior subgraph for this session. Then, through generalization, the session behavior subgraph can be mapped to a form such as "homepage—list page—product click." Initial weights are configured for these two directed edges, for example, both set to 1. Suppose that user A will visit again later, but another session behavior subgraph containing the same path is generated. Of course, this session behavior subgraph does not only represent purchasing the same product, but only includes these two directed edges. During graph fusion, based on user A's user identifier, the same directed edges in the two session behavior subgraphs are accumulated. At this time, the weight becomes 2, so in the final generated global directed graph network for user behavior, this path is highlighted due to its higher accumulated weight.

[0083] like Figure 3 As shown in the preferred embodiment of the present invention, the step of configuring initial weights for each directed edge specifically includes:

[0084] S213, determine the page hierarchy depth of each interactive event node based on the site information architecture, and calculate the hierarchy span coefficient of each directed edge, wherein the hierarchy span coefficient is calculated based on the page hierarchy depth of the first and last nodes of the directed edge;

[0085] S223, calculate the path concentration coefficient based on the frequency of the occurrence of the directed edge in all session behavior subgraphs and the total number of outgoing edges from its starting node;

[0086] S233, calculate the comprehensive interaction weight for each directed edge based on the hierarchical span coefficient and the path concentration coefficient. The comprehensive interaction weight is used to characterize the difficulty of the interaction path corresponding to the directed edge occurring.

[0087] S243, the calculated comprehensive interaction weights are used as the initial weights.

[0088] Continuing with the example of User A shopping in the previous embodiment, in this embodiment of the invention, the page hierarchy depth of each interactive event node can be divided through the site information architecture. It can be concluded that the page hierarchy depth values ​​of "Homepage," "Product List Page," and "Product Details Page" increase sequentially. The hierarchy span coefficient is obtained by calculating the difference in page hierarchy depth values, which can represent a logical progression of levels. Assuming the "Product List Page" is the starting node, statistics show that users have three choices from this page: 60% of the session flows to the "Product Details Page," 30% to the "Filter Component," and 10% to the "Advertisement." Therefore, the directed edge "List Page—Details Page," due to its high frequency and relatively fixed total outflow edge count from the starting node, will have a lower calculated path concentration coefficient, indicating a highly certain and mainstream choice. Combining the lower hierarchy span coefficient and the lower path concentration coefficient, the calculated comprehensive interaction weight will also be lower, indicating that this browsing path is mainstream, i.e., the path most likely to occur during natural browsing. Suppose a user clicks from the "Product Details Page" to the "Privacy Policy." This level has a large width coefficient, and because this behavior accounts for a very small percentage, its path concentration coefficient is also high. Therefore, the calculated overall interaction weight is large, indicating that this type of user browsing behavior is unlikely to occur. Configuring the initial weight in this way allows for a more precise distinction between the essential differences between the user's main workflow and auxiliary / peripheral operations.

[0089] like Figure 4 As shown, in a preferred embodiment of the present invention, the step of analyzing each directed edge in the directed graph network of user behavior using artificial intelligence algorithms to obtain the topological key edge specifically includes:

[0090] S301, input the directed graph network of user behavior into the artificial intelligence model, and calculate the dynamic attention coefficient for each directed edge through the self-attention mechanism;

[0091] S302, use the community detection algorithm to cluster the nodes of the directed graph network of user behavior, divide it into multiple high-density interaction sub-communities, and determine whether each directed edge is a cross-region edge. The cross-region edge is where the first and last nodes belong to two different high-density interaction sub-communities respectively. The high-density interaction sub-communities have dense internal connections and sparse external connections.

[0092] S303, combine the attention coefficient of each directed edge and the Boolean attribute of whether it is a cross-region edge to form a joint feature vector;

[0093] S304. Input the joint feature vector into a lightweight edge classification neural network to obtain the probability score of each edge as a topological key edge, and determine the directed edges with probability scores higher than a preset threshold as topological key edges.

[0094] In this embodiment of the invention, a directed graph network of user behavior is input into an artificial intelligence model based on a self-attention mechanism (as shown in the figure attention network). This model automatically calculates a dynamic attention coefficient for each directed edge in the network through information transmission and weighted aggregation between nodes. The attention coefficient is not preset, but is learned by the model based on the context structure of the edge, and can represent the potential influence and importance of a directed edge in the global information propagation. Then, a community detection algorithm (such as the Louvain algorithm) is used to perform unsupervised clustering on the nodes of the same directed graph network of user behavior, automatically dividing multiple high-density interactive sub-communities. For example, there can be a browsing and exploration community mainly composed of "homepage", "product list page", and "filter component" nodes; a purchase decision community mainly composed of "product details page", "add to cart button", and "cart page" nodes; and a payment completion community composed of "payment page" and "payment success page" nodes. Based on this, the topological key edge is determined, which of course requires combining a joint feature vector composed of the attention coefficient of the directed edge and the Boolean attribute of whether it is a cross-region edge. Specifically, for example, a directed edge from the "Product Details Page" (belonging to the purchase decision community) to the "Shopping Cart Page" (also belonging to the purchase decision community) that leads to the "View Shopping Cart" option, although its attention coefficient is high (due to its high-frequency operation), it is not a cross-region edge, and its joint feature vector has a false cross-region Boolean attribute. Therefore, it has a relatively low probability of being classified as a topological key edge by the classifier. However, a directed edge from the "Product Details Page" (purchase decision community) to the "Payment Page" (payment completion community), representing immediate purchase, not only receives a high attention coefficient but is also a cross-region edge connecting the two functional modules of "Purchase Decision" and "Payment Completion." Its joint feature vector possesses both a high attention value and a true cross-region Boolean attribute, so the edge classification neural network outputs a high probability score for it, classifying it as a crucial topological key edge.

[0095] like Figure 5 As shown, in a preferred embodiment of the present invention, the step of outputting the front-end interaction positions corresponding to the topological key edges as a set of recommended tracking points specifically includes:

[0096] S401, For each topological key edge, locate its head and tail nodes in the directed graph network of user behavior, and extract the interaction events associated with the topological key edge to form a sequence of directly related events;

[0097] S402, parse the attribute information of the interaction event corresponding to each node in the directly related event sequence, and generate a business description text describing the key edge of the topology based on the logical relationship;

[0098] S403, associate and bind the front-end interaction location and business description text corresponding to the key edge of the topology to obtain the event recommendation record;

[0099] S404, Integrate all the event tracking recommendation records to form the event tracking recommendation location set.

[0100] In this embodiment of the invention, the previous embodiment is still used, and a topological key edge has been determined as: the "Buy Now" button click interaction from the "Product Details Page" (Node A) to the "Payment Page" (Node B). Then, nodes A and B are located in the network, and their directly related event sequence is extracted, for example, in the form of ("Add to Cart" button on the product details page) — ("Buy Now" button on the topological key edge) — ("Select Payment Method" component on the payment page). Then, business description text is generated. The above sequence is analyzed, and its node attribute information is "Add to Cart Operation", "Buy Now Operation", and "Select Payment Method Operation". Based on this attribute information, business description text such as "This tracking point is used to monitor the key conversion behavior of users directly initiating payment by clicking the "Buy Now" button on the product details page. This behavior directly connects the product confirmation and payment process" can be generated. Then, by querying the code mapping table, the button that logically represents "Buy Now" is mapped to a specific location in the front-end project. This location information is bound to the aforementioned business description text to form a recommended tracking record. All similar records are then integrated to output a list containing all front-end code locations where tracking needs to be implemented and their business explanations, which is the final set of recommended tracking locations.

[0101] like Figure 6 As shown, in a preferred embodiment of the present invention, the AI-based front-end tracking recommendation method further includes:

[0102] S501, Based on the user behavior directed graph network, the structural index of the topological key edge is obtained by quantifying the removal of a single topological key edge through simulation.

[0103] S502 uses a random walk algorithm to generate a sequence of virtual user behaviors and simulates a set of failed interactions by blocking paths through the topological critical edges in the sequence.

[0104] S503 selects multiple preset target nodes in the directed graph network of user behavior, and obtains the decay index by calculating the difference in the proportion of sessions that reach any target node in the set of normal interaction and failed interaction sessions.

[0105] S504, the structural index and attenuation index are fused and calculated to obtain a comprehensive score for each topological key edge, and the recommended location set of embedding points is prioritized according to the comprehensive score.

[0106] In this embodiment of the invention, based on a constructed global user behavior directed graph network, each topological key edge is sequentially simulated for removal (making it ineffective), and the degree of degradation of the overall network topology is observed and calculated. A structural vulnerability index for each edge is obtained through quantitative indicators (such as the growth rate of the network's average shortest path or the decrease in the network's global efficiency). The higher the index, the more important the topological key edge is in network connectivity. Then, a massive number of virtual user behavior sequences simulating normal users browsing from various entry pages are generated using a random walk algorithm. By explicitly blocking paths through specific topological key edges in this algorithm, an abnormal session set corresponding to the edge's failure scenario can be generated, thereby simulating user behavior trends when the interaction point is unusable. Multiple business target nodes representing core conversion goals (such as "payment success page" and "registration completion page") are preset. The proportion of sessions that can successfully reach any preset business target node in both the normal virtual session set and the aforementioned abnormal session set is counted, and the difference between the two is calculated as a business attenuation rate index. This business attenuation rate index reflects the potential impact of the failure of the topological key edge on the final business conversion rate. Finally, the structural vulnerability index and business decay rate index of each topological key edge are standardized and fused together (such as by weighted average) to obtain the comprehensive utility score of each topological key edge. Based on this, the set of recommended event tracking locations can be prioritized, and event tracking locations with higher scores can be implemented first.

[0107] like Figure 7 As shown in the figure, this embodiment of the invention also provides a front-end tracking recommendation system based on artificial intelligence, the system comprising:

[0108] The session acquisition module 100 is used to collect historical user session data and extract user identifiers, sequences of interaction events, and timestamps of event occurrences. The interaction events include page access and element clicks.

[0109] The network modeling module 200 is used to construct a directed graph network of user behavior based on historical user session data. The directed graph network of user behavior uses interactive events as nodes and the continuous occurrence relationship between interactive events as directed edges.

[0110] The intelligent filtering module 300 is used to analyze each directed edge in the directed graph network of user behavior using artificial intelligence algorithms to obtain the topological key edge, which is a directed edge connecting different high-density interactive sub-communities.

[0111] The tracking point output module 400 is used to output the front-end interaction positions corresponding to the topological key edges as a set of recommended tracking point positions.

[0112] like Figure 8 As shown, in a preferred embodiment of the present invention, the network modeling module 200 includes:

[0113] The timing construction unit 201 is used to sort the interaction events within a single session according to the timestamp, and construct the session as a session behavior subgraph by defining two adjacent interaction events as a directed edge.

[0114] The semantic generalization unit 202 is used to perform node generalization processing on all session behavior subgraphs, mapping page URLs and element clicks to logical page identifiers and semantic component type identifiers, respectively, to improve the identifiability of the directed edges in different sessions.

[0115] Weight configuration unit 203 is used to configure initial weights for each directed edge;

[0116] Graph fusion unit 204 is used to merge nodes and edges of multiple session behavior subgraphs generated by the same user based on user identifier, and obtain the user behavior directed graph network by accumulating the weights of the same directed edges.

[0117] like Figure 9 As shown, in a preferred embodiment of the present invention, the intelligent screening module 300 includes:

[0118] The attention evaluation unit 301 is used to input the user behavior directed graph network into the artificial intelligence model and calculate the dynamic attention coefficient for each directed edge through the self-attention mechanism.

[0119] The community identification unit 302 is used to cluster the nodes of the directed graph network of user behavior using the community discovery algorithm, divide them into multiple high-density interactive sub-communities, and determine whether each directed edge is a cross-region edge. The cross-region edge is where the first and last nodes belong to two different high-density interactive sub-communities respectively. The high-density interactive sub-communities have dense internal connections and sparse external connections.

[0120] The feature fusion unit 303 is used to combine the attention coefficients of each directed edge and the Boolean attribute of whether it is a cross-region edge to form a joint feature vector;

[0121] The key determination unit 304 is used to input the joint feature vector into a lightweight edge classification neural network to obtain the probability score of each edge as a topological key edge, and to determine the directed edge with the probability score higher than a preset threshold as a topological key edge.

[0122] like Figure 10 As shown, in a preferred embodiment of the present invention, the AI-based front-end event tracking recommendation system further includes an event tracking evaluation module 500, which includes:

[0123] The structural analysis unit 501 is used to quantify the structural index of the topological key edge based on the user behavior directed graph network by simulating the removal of a single topological key edge.

[0124] The behavior simulation unit 502 is used to generate a virtual user behavior sequence using a random walk algorithm, and to simulate a set of failed interactions by blocking the path through the topological key edge in the sequence.

[0125] The impact analysis unit 503 is used to select multiple preset target nodes in the directed graph network of user behavior, and obtain the attenuation index by calculating the difference in the proportion of sessions that reach any target node in the set of normal interaction and failed interaction sessions.

[0126] The priority decision unit 504 is used to fuse the structural index and the attenuation index to obtain a comprehensive score for each topological key edge, and to prioritize the set of recommended embedding locations based on the comprehensive score.

[0127] The above description only details the preferred embodiments of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0128] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0129] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the disclosure in the specification and embodiments. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

Claims

1. A front-end tracking recommendation method based on artificial intelligence, characterized in that, The method includes the following steps: Collect historical user session data and extract user identifiers, sequences of interaction events, and timestamps of event occurrences. The interaction events include page visits and element clicks. A directed graph network of user behavior is constructed based on historical user session data. The directed graph network of user behavior uses interaction events as nodes and the continuous occurrence relationship between interaction events as directed edges. The topological key edge is obtained by analyzing each directed edge in the directed graph network of user behavior using artificial intelligence algorithms. The topological key edge is a directed edge that connects different high-density interactive sub-communities. Output the front-end interaction locations corresponding to the topological key edges as a set of recommended tracking locations; The steps for configuring initial weights for each directed edge include: The page hierarchy depth of each interactive event node is determined based on the site information architecture, and the hierarchy span coefficient of each directed edge is calculated. The hierarchy span coefficient is calculated based on the page hierarchy depth of the first and last nodes of the directed edge. The path concentration coefficient is calculated based on the frequency of the occurrence of a directed edge in all session behavior subgraphs and the total number of outgoing edges from its starting node. The comprehensive interaction weight is calculated for each directed edge based on the hierarchical span coefficient and the path concentration coefficient. The comprehensive interaction weight is used to characterize the difficulty of the interaction path corresponding to the directed edge occurring. The calculated comprehensive interaction weights are used as the initial weights; The step of using artificial intelligence algorithms to analyze each directed edge in a directed graph network of user behavior to obtain the topological key edge specifically includes: The directed graph network of user behavior is input into the artificial intelligence model, and the dynamic attention coefficient is calculated for each directed edge through the self-attention mechanism; The nodes of the directed graph network of user behavior are clustered using a community detection algorithm to divide them into multiple high-density interactive sub-communities. Each directed edge is determined to be a cross-region edge, where the first and last nodes belong to two different high-density interactive sub-communities respectively. The high-density interactive sub-communities have dense internal connections and sparse external connections. The attention coefficient of each directed edge and the Boolean attribute indicating whether it is a cross-region edge are combined to form a joint feature vector; The joint feature vector is input into a lightweight edge classification neural network to obtain the probability score of each edge as a topological key edge, and directed edges with probability scores higher than a preset threshold are identified as topological key edges.

2. The AI-based front-end tracking recommendation method according to claim 1, characterized in that, The steps for constructing a directed graph network of user behavior based on historical user session data specifically include: The interaction events within a single session are sorted according to the timestamp, and the session is constructed as a session behavior subgraph by defining two adjacent interaction events as a directed edge. All session behavior subgraphs are generalized into nodes, and page URLs and element clicks are mapped to logical page identifiers and semantic component type identifiers, respectively, to improve the recognizability of the directed edges in different sessions. Based on user identifiers, multiple session behavior subgraphs generated by the same user are merged into nodes and edges, and the directed graph network of user behavior is obtained by accumulating the weights of the same directed edges.

3. The front-end tracking recommendation method based on artificial intelligence according to claim 1, characterized in that, The step of outputting the front-end interaction locations corresponding to the topological key edges as a set of recommended tracking locations specifically includes: For each topological key edge, locate its first and last nodes in the directed graph network of user behavior, and extract the interaction events associated with the topological key edge to form a sequence of directly related events; Parse the attribute information of the interaction event corresponding to each node in the directly related event sequence, and generate business description text describing the key edge of the topology based on logical relationships; Associate and bind the front-end interaction location and business description text corresponding to the key edge of the topology to obtain the event tracking recommendation record; All the recommended tracking points are integrated to form the recommended tracking point location set.

4. The front-end tracking recommendation method based on artificial intelligence according to claim 1, characterized in that, The method further includes: Based on the user behavior directed graph network, the structural index of the topological key edge is obtained by quantifying the removal of a single topological key edge through simulation. A set of failed interaction sessions is simulated by generating a virtual user behavior sequence using a random walk algorithm and blocking paths that pass through topological critical edges in the sequence. In a directed graph network of user behavior, multiple target nodes are selected, and a decay index is obtained by calculating the difference in the proportion of sessions that reach any target node in the set of normal interaction and failed interaction sessions. The structural index and attenuation index are fused together to obtain a comprehensive score for each topological key edge, and the recommended location set of embedding points is prioritized according to the comprehensive score.

5. A front-end tracking recommendation system based on artificial intelligence, characterized in that, The system includes: The session acquisition module is used to collect historical user session data and extract user identifiers, sequences of interaction events, and timestamps of event occurrences. The interaction events include page visits and element clicks. The network modeling module is used to construct a directed graph network of user behavior based on historical user session data. The directed graph network of user behavior uses interactive events as nodes and the continuous occurrence relationship between interactive events as directed edges. The intelligent filtering module is used to analyze each directed edge in the directed graph network of user behavior using artificial intelligence algorithms to obtain the topological key edge, which is a directed edge connecting different high-density interactive sub-communities; The event tracking output module is used to output the front-end interaction locations corresponding to the topological key edges as a set of recommended event tracking locations; The intelligent filtering module includes: The attention evaluation unit is used to input the directed graph network of user behavior into the artificial intelligence model and calculate the dynamic attention coefficient for each directed edge through the self-attention mechanism. The community identification unit is used to cluster the nodes of the directed graph network of user behavior using the community discovery algorithm, divide it into multiple high-density interactive sub-communities, and determine whether each directed edge is a cross-region edge. The cross-region edge is where the first and last nodes belong to two different high-density interactive sub-communities respectively. The high-density interactive sub-communities have dense internal connections and sparse external connections. The feature fusion unit is used to combine the attention coefficients of each directed edge and the Boolean attribute indicating whether it is a cross-region edge to form a joint feature vector; The key determination unit is used to input the joint feature vector into a lightweight edge classification neural network, obtain the probability score of each edge as a topological key edge, and determine the directed edge with the probability score higher than a preset threshold as a topological key edge. The steps for configuring initial weights for each directed edge include: The page hierarchy depth of each interactive event node is determined based on the site information architecture, and the hierarchy span coefficient of each directed edge is calculated. The hierarchy span coefficient is calculated based on the page hierarchy depth of the first and last nodes of the directed edge. The path concentration coefficient is calculated based on the frequency of the occurrence of a directed edge in all session behavior subgraphs and the total number of outgoing edges from its starting node. The comprehensive interaction weight is calculated for each directed edge based on the hierarchical span coefficient and the path concentration coefficient. The comprehensive interaction weight is used to characterize the difficulty of the interaction path corresponding to the directed edge occurring. The calculated comprehensive interaction weights are used as the initial weights.

6. The AI-based front-end tracking recommendation system according to claim 5, characterized in that, The network modeling module includes: The timing construction unit is used to sort the interaction events within a single session according to the timestamp, and construct the session as a session behavior subgraph by defining two adjacent interaction events as a directed edge. The semantic generalization unit is used to perform node generalization processing on all session behavior subgraphs, mapping page URLs and element clicks to logical page identifiers and semantic component type identifiers, respectively, to improve the recognizability of the directed edges in different sessions. The graph fusion unit is used to merge nodes and edges of multiple session behavior subgraphs generated by the same user based on the user identifier, and obtain the user behavior directed graph network by accumulating the weights of the same directed edges.

7. The AI-based front-end tracking recommendation system according to claim 5, characterized in that, The system also includes a data tracking evaluation module, which includes: The structural analysis unit is used to quantify the structural index of a single topological key edge based on the user behavior directed graph network by simulating the removal of a single topological key edge. The behavior simulation unit is used to generate virtual user behavior sequences using a random walk algorithm, and to simulate a set of failed interaction sessions by blocking paths through topological key edges in the sequence. The impact analysis unit is used to select multiple preset target nodes in the directed graph network of user behavior, and obtain the attenuation index by calculating the difference in the proportion of sessions that reach any target node in the set of normal interaction and failed interaction sessions. The priority decision unit is used to fuse the structural index and the attenuation index to obtain a comprehensive score for each topological key edge, and to prioritize the set of recommended embedding locations based on the comprehensive score.

Images