Website ranking method and system based on big data analysis, medium and product

Abstract

The invention provides a website ranking method and system based on big data analysis, a medium and a product, and relates to the technical field of computers.The method comprises the steps that log data sets of to-be-ranked websites accessed by multiple users in a target time period are obtained, the log data sets are divided according to preset session division conditions, and the log data sets are obtained; obtaining an access session set of each user; analyzing the access session set, determining a page access sequence of each user in each access session, and constructing a directed graph structure according to the page access sequence; extracting a user browsing path from the directed graph structure by using the access transfer relationship, and determining a page hub index of each node in the user browsing path; and performing path structure feature extraction on each access session by using the user browsing path and the page hub index to obtain the browsing behavior complexity of each access session, and performing quality analysis operation on the to-be-ranked websites by using the browsing behavior complexity to obtain a quality ranking result.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a website ranking method, system, medium, and product based on big data analysis. Background Technology

[0002] With the explosive growth of internet information and the continuous improvement of users' personalized needs, website ranking technology, as a core technology of search engines and recommendation systems, has been widely used in information retrieval and content distribution. Website ranking systems quantify the value of websites by analyzing user behavior data and generate ranked lists based on the scoring results to provide information services to users.

[0003] Among related technologies, the main approach is website ranking based on user behavior sequence modeling. This involves collecting user clickstream logs and segmenting them into sessions; inputting the URL access sequences within each session into an LSTM model to generate session vector representations; aggregating all session vectors accessing the website and combining them with explicit statistical indicators to calculate a comprehensive score; and then generating a website ranking based on this comprehensive score. This method can improve the accuracy of website ranking results by capturing the temporal characteristics of user behavior through the LSTM model.

[0004] However, the website ranking method based on user behavior sequence modeling mentioned above is problematic because the LSTM model only models user behavior data in the time dimension. Therefore, when users perform complex interactive behaviors such as complex jumps or multi-path parallel browsing within a website, it is difficult to identify the differences in service quality reflected by the complex interactive behaviors. This makes it difficult to distinguish between high-quality websites that meet user needs and low-quality websites that only have superficial traffic, resulting in a low accuracy rate of quality ranking of websites browsed by users in related technologies. Summary of the Invention

[0005] This application provides a website ranking method, system, medium, and product based on big data analysis, which is used to improve the accuracy of quality rankings for users browsing websites.

[0006] Firstly, this application provides a website ranking method based on big data analysis, applied to the aforementioned website ranking system. The method includes: acquiring a set of log data from multiple users accessing a website to be ranked within a target time period; dividing the log data set into access sessions according to preset session segmentation conditions to obtain an access session set for each user; performing temporal structure feature analysis on each user's access session set to determine the page access sequence within each access session included in each user's access session set; and constructing a directed graph structure associated with page jumps in each access session based on the page access sequence. Each node in the directed graph structure... Each node in the page access sequence corresponds one-to-one with each accessed page. The directed edges in the directed graph structure represent the access transition relationships for page jumps. The user browsing path is extracted from the directed graph structure using these access transition relationships, and the page hub index of each node in the user browsing path is determined. The page hub index represents the transit importance of the accessed page corresponding to each node in the user browsing path. The path structure features of each access session are extracted using the user browsing path and the page hub index to obtain the browsing behavior complexity of each access session. The browsing behavior complexity is then used to perform quality analysis operations on the website to be ranked, resulting in the quality ranking results of the website to be ranked.

[0007] By adopting the above technical solutions, the session partitioning of the log data set and the construction of the directed graph structure work synergistically. Session partitioning can divide continuous user behavior sequences into independent analyses, while the directed graph structure transforms the linear access sequences within each session into a network topology. This transformation allows the complex jump relationships between pages to be fully preserved. The calculation of the page hub index relies on the topological characteristics of the directed graph. By analyzing the connection relationships of nodes, important intermediate pages are identified. Compared with traditional access frequency statistics, the hub index can reflect the structural role of a page in the user's browsing path. The extraction of browsing behavior complexity comprehensively utilizes the diversity of user browsing paths and the distribution characteristics of page hub index, quantifying behavioral patterns into comparable numerical indicators. This allows quality analysis operations to evaluate website quality based on objective complexity indicators, avoiding the bias caused by simply relying on traffic data. This solves the technical problem of low accuracy in ranking the quality of websites viewed by users in related technologies, achieving the technical effect of improving the accuracy of ranking the quality of websites viewed by users.

[0008] Optionally, the path structure features of each access session are extracted using the user browsing path and page hub index to obtain the browsing behavior complexity of each access session. Specifically, this includes: traversing the directed graph structure of each access session and performing the following operations on the currently traversed directed graph structure: determining the set of strongly connected nodes in the target directed graph structure from which there is a bidirectional path between any two nodes, based on the access transition relationship; selecting high hub index nodes from the target directed graph structure that are greater than a preset hub index threshold; selecting a set of deep interaction nodes including high hub index nodes from the set of strongly connected nodes, and then adding the deep interaction nodes... The node set is labeled as a deep interaction region; the number of deep interaction regions in the target directed graph structure is determined, and the number is used as the interaction region quantity feature; the total number of nodes in the deep interaction regions is compared with the total number of nodes in the target directed graph structure to obtain the node total ratio, and the node total ratio is used as the interaction coverage feature; the interaction region quantity feature and the interaction coverage feature are determined as the session interaction depth feature; the path dispersion feature is extracted from the target directed graph structure using the user browsing path and page hub index; the feature fusion operation is performed on the session interaction depth feature, path dispersion feature, and the duration of each access session to obtain the browsing behavior complexity of each access session.

[0009] By employing the aforementioned technical solutions, the identification of strongly connected node sets and the filtering of high-hubness nodes work together. The former captures users' cyclical browsing behavior among page groups, while the latter locates the core functional pages of the website. The deep interaction area formed by their intersection accurately reflects the page groups in which users are deeply involved. The quantity and coverage features of the interaction area characterize the depth of the session from different dimensions. The quantity reflects the breadth of the distribution of deep interactions, while the coverage reflects the relative intensity of deep interactions. The combination of the two forms a complete representation of the session's interaction depth features. The extraction of path dispersion features complements the interaction depth features. The former focuses on the divergence of user browsing, while the latter focuses on the depth of cyclical access. Feature fusion integrates multi-dimensional features with duration information to generate a comprehensive browsing behavior complexity index that reflects user behavior patterns, achieving a comprehensive quantitative assessment of session quality.

[0010] Optionally, path dispersion features are extracted from the target directed graph structure using user browsing paths and page hub indexes. Specifically, this includes: identifying a first node with an in-degree of zero and a second node with an out-degree of zero from the target directed graph structure; designating the first node as the entry page node and the second node as the exit page node; determining all connected paths from the entry page node to the exit page node and determining the path length of each connected path; determining the path weight value of each connected path based on the path length and page hub index, and designating connected paths with path weight values ​​greater than a preset path weight threshold as main browsing paths, and designating all connected paths other than the main browsing paths as branch browsing paths; determining the average path length of the branch browsing paths based on the path length, and performing a length comparison analysis between the average path length and the path length of the main browsing path to obtain the path length ratio, and using the path length ratio as the path dispersion feature.

[0011] By adopting the above technical solution, the determination of entry and exit nodes provides clear start and end points for path analysis, ensuring clear boundary conditions for the traversal of all connected paths. The combination of path length and page pivotality index generates a path weight value. This weight considers not only the physical length of the path but also incorporates page importance factors, making the identification of the main browsing path more accurate. The distinction between the main browsing path and branch browsing paths reflects the primary and secondary relationships of user behavior. The main path represents the core browsing trajectory, while branch paths embody exploratory behavior. The ratio of the lengths of the two types of paths can quantify the degree of focus or divergence in user browsing. Path dispersion features, as a single numerical indicator, compress complex path structure information into a comparable scalar, facilitating subsequent fusion calculations with other features and improving the efficiency and accuracy of browsing behavior complexity calculations.

[0012] Optionally, a feature fusion operation is performed on the session interaction depth features, path dispersion features, and the duration of each access session to obtain the browsing behavior complexity of each access session. Specifically, this includes: determining the target duration interval to which the duration of each access session belongs; performing numerical interval analysis on the session interaction depth features to obtain a first value for the interaction area quantity feature and a second value for the interaction coverage feature; classifying the session interaction depth features into surface browsing level and deep exploration level based on the first and second values ​​to obtain the interaction depth classification result; performing numerical range analysis on the path dispersion features to obtain a third value for the path dispersion features; using the third value to classify the path dispersion features into linear browsing mode and dispersed browsing mode to obtain the path mode classification result; and performing a feature fusion operation on the interaction depth classification result, the path mode classification result, and the target duration interval to obtain the browsing behavior complexity of each access session.

[0013] By adopting the above technical solutions, a two-dimensional classification system is formed by dividing the duration interval and analyzing the interaction depth features. The duration interval reflects the user's level of engagement, while the interaction depth classification reflects the quality of behavior. The combination of the two can more accurately locate the session type. The interaction depth classification result is obtained by comprehensively judging the first and second values, avoiding the one-sidedness of a single indicator and ensuring that the determination of the depth exploration level is more reliable. The pattern classification of path dispersion features and the interaction depth classification form an orthogonal dimension. The former describes the browsing breadth feature, while the latter characterizes the interaction depth feature. The combination of the two dimensions can cover more diverse user behavior patterns. The fusion operation of the three types of features organically combines the time, depth, and breadth dimensions. The interaction between different dimensional features can compensate for the limitations of a single feature, and the generated browsing behavior complexity index can more comprehensively reflect the overall quality of the session.

[0014] Optionally, a feature fusion operation is performed on the interaction depth classification result, path pattern classification result, and target duration interval to obtain the browsing behavior complexity of each access session. Specifically, this includes: when the interaction depth classification result is at the deep exploration level and the path pattern classification result is at the scattered browsing mode, multiplying the interaction coverage feature by a first preset coefficient to obtain the deep interaction contribution value; when the interaction depth classification result is at the surface browsing level and the path pattern classification result is at the linear browsing mode, multiplying the interaction area quantity feature by a second preset coefficient to obtain the surface interaction contribution value, wherein the second preset coefficient is less than the first preset coefficient. A preset coefficient is used; a duration adjustment factor is determined based on the target duration interval, wherein: when the target duration interval is a short-term trial interval, the duration adjustment factor is determined as the first adjustment value; when the target duration interval is a medium-term activation interval, the duration adjustment factor is determined as the second adjustment value; and when the target duration interval is a long-term in-depth exploration interval, the duration adjustment factor is determined as the third adjustment value. The first adjustment value is less than the second adjustment value, and the second adjustment value is less than the third adjustment value. The deep interaction contribution value or surface interaction contribution value is multiplied by the duration adjustment factor and the path dispersion feature to obtain the browsing behavior complexity.

[0015] By adopting the above technical solutions, the differentiated preset coefficient settings reflect the value distinction of different browsing modes. The design of the first preset coefficient being greater than the second preset coefficient gives higher weight to in-depth exploration behavior, which is in line with the value orientation of high-quality interaction. The separate calculation of the contribution value of in-depth interaction and the contribution value of surface interaction avoids confusion between different types of behavior and ensures that each behavior mode receives a corresponding quantitative representation according to its characteristics. The incremental setting of the duration adjustment factor reflects the positive correlation between user time investment and behavior value, and the gradient difference of the three adjustment values ​​can finely distinguish the quality contribution of sessions of different durations. The product operation multiplies the interaction contribution value, duration adjustment factor and path dispersion feature. This non-linear combination can amplify the impact of high-quality behavior while suppressing the weight of low-value behavior, so that the final browsing behavior complexity more accurately reflects the actual value of the session.

[0016] Optionally, the target duration range to which the duration of each access session belongs is determined, specifically including: obtaining a first preset duration threshold and a second preset duration threshold, wherein the first preset duration threshold is less than the second preset duration threshold; performing duration comparison analysis with the first preset duration threshold and the second preset duration threshold respectively to obtain duration comparison results; when the duration is determined to be less than the first preset duration threshold based on the duration comparison results, the access session corresponding to the duration is divided into a short-term trial interval; when the duration is determined to be greater than or equal to the first preset duration threshold and less than the second preset duration threshold based on the duration comparison results, the access session corresponding to the duration is divided into a medium-term activation interval; when the duration is determined to be greater than or equal to the second preset duration threshold based on the duration comparison results, the access session corresponding to the duration is divided into a long-term deep exploration interval.

[0017] By adopting the above technical solution, a three-segment duration division system is established by setting two preset duration thresholds. The first threshold distinguishes between quick browsing and in-depth access, while the second threshold identifies excessively long periods of in-depth exploration. The dual-threshold design provides more refined duration classification than a single threshold. The serial judgment logic of duration comparison analysis ensures that each session is uniquely assigned to a specific interval, avoiding interval overlap or omissions. The identification of short-term trial intervals can filter out sessions with quick browsing or clear objectives, the mid-term stimulation interval captures normal-depth access behavior, and the long-term in-depth exploration interval locates highly engaged user groups. The division of the three intervals is consistent with the natural distribution of user behavior patterns. This duration-based interval division provides a standardized time dimension input for subsequent feature fusion, enabling sessions of different durations to be quality-assessed within a unified framework.

[0018] Optionally, a quality analysis operation is performed on the website to be ranked based on browsing behavior complexity to obtain the quality ranking result of the website to be ranked. Specifically, this includes: traversing each access session of the website to be ranked and performing the following operations on the currently traversed access session: performing interaction depth analysis on the target access session, determining the interaction depth coefficient of the target access session when the interaction depth classification result of the target access session is the surface browsing level and the duration of the target access session is less than a preset effective duration threshold, and the target access session is the currently traversed access session; performing interaction quality analysis on the target access session, determining the interaction quality coefficient of the target access session when the browsing behavior complexity of the target access session is greater than a preset complexity threshold and the number of nodes in the directed graph structure of the target access session that have a page hub degree index greater than a preset core hub degree threshold; determining the session quality value of the target access session based on the interaction depth coefficient and the interaction quality coefficient, summing the session quality values ​​of all access sessions belonging to the same website to be ranked to obtain the total website quality score of the website to be ranked; and sorting all websites to be ranked in descending order based on the total website quality score to generate the quality ranking result.

[0019] By adopting the above technical solution, the dual evaluation mechanism of interaction depth coefficient and interaction quality coefficient can measure the value of a session from different perspectives. The former identifies low-value surface browsing, while the latter rewards high-complexity, deep interactions. The combination of the two coefficients avoids the bias of a single indicator. The session quality score is calculated by multiplying the two coefficients. This multiplicative relationship ensures that only sessions possessing both depth and quality can obtain high scores, effectively filtering out surface traffic and invalid visits. The summation of the total website quality score aggregates the contributions of all sessions, reflecting the overall user value of the website. The differentiated contributions of sessions of different quality ensure that high-quality websites can stand out. The quality ranking results generated in descending order intuitively present the quality differences between websites. The ranking based on browsing behavior complexity reflects the true value of a website better than traditional traffic ranking, providing a reliable basis for user recommendations.

[0020] In a second aspect, embodiments of this application provide a website ranking system, which includes: one or more processors and a memory; the memory is coupled to one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the website ranking system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product including instructions that, when the computer program product is run on a website ranking system, cause the website ranking system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a website ranking system, cause the website ranking system to perform the method described in the first aspect and any possible implementation thereof. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a website ranking method based on big data analysis in an embodiment of this application; Figure 2 This is a schematic diagram of the physical device structure of a website ranking system in this application embodiment. Detailed Implementation

[0024] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0025] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0026] This application provides a website ranking method based on big data analysis, see [link / reference]. Figure 1 , Figure 1 This is a flowchart illustrating a website ranking method based on big data analysis in an embodiment of this application, including the following steps: Step S101: Obtain a set of log data of multiple users accessing the website to be ranked within the target time period, and divide the log data set into access sessions according to the preset session division conditions to obtain the access session set of each user among the multiple users. Step S102: Perform temporal structure feature parsing on the access session set of each user to determine the page access sequence within each access session included in the access session set for each user, and construct a directed graph structure associated with page jumps in each access session based on the page access sequence. Each node in the directed graph structure corresponds one-to-one with each access page in the page access sequence, and the directed edges in the directed graph structure represent the access transfer relationship of page jumps. Step S103: Extract the user browsing path from the directed graph structure using the access transfer relationship, and determine the page hub index of each node in the user browsing path. The page hub index represents the transit importance of the access page corresponding to each node in the user browsing path. Step S104: Extract path structure features for each access session using user browsing path and page hub index to obtain the browsing behavior complexity of each access session, and use the browsing behavior complexity to perform quality analysis on the website to be ranked to obtain the quality ranking result of the website to be ranked.

[0027] In the above embodiment, assume an e-commerce platform needs to perform quality ranking on thousands of merchant websites to optimize its recommendation algorithm and advertising strategy. Probe programs are deployed at various network nodes to capture user HTTP / HTTPS request data in real time. The probe programs parse network packets, extracting key fields such as user MAC address, VLAN identifier, protocol type, source IP address and port, destination IP address and port, request timestamp, accessed domain name, HTTP request method, response status code, uplink bytes, and downlink bytes. The collected raw logs first undergo a data cleaning process to remove duplicate records, repair missing fields, unify the time format to UTC standard time, and map IP addresses to corresponding domain name information. The target time period is set to the most recent 7 days, i.e., all access records within a 168-hour time window counting backwards from the current time. A time interval-based segmentation strategy is used for session segmentation; when two adjacent access records of the same user are detected with a time interval exceeding 30 minutes, they are divided into different sessions. For example, if user U1 visits page A at 9:00 AM, page B at 9:05 AM, page C at 9:10 AM, then visits page A again at 9:50 AM, and page D at 10:00 AM, the first three visits will be classified as session S1, and the latter two visits as session S2. In this way, the access logs of all users within 7 days are divided into independent session units, with each user forming a set containing multiple sessions. During the directed graph structure construction phase, taking user U1's session S1 as an example, its page access sequence is [Homepage A → Product List B → Product Details C → Homepage A → Shopping Cart D → Product Details C]. This sequence is converted into a directed graph structure containing 4 nodes, corresponding to the four pages: Homepage A, Product List B, Product Details C, and Shopping Cart D. Directed edges represent the navigation relationships between pages. For example, there is a directed edge from homepage A to product list B, a directed edge from product list B to product details C, a directed edge from product details C back to homepage A, a directed edge from homepage A to shopping cart D, and a directed edge from shopping cart D back to product details C. Each node in the graph structure has a unique identifier, and the hash value of the URL ensures consistent identification of the same page across different sessions.

[0028] In the above embodiment, the pivotality index is calculated using an improved PageRank algorithm combined with a betweenness centrality measure. The pivotality H(v) of node v is calculated as H(v) = α × PR(v) + β × BC(v), where PR(v) is the PageRank value of node v, BC(v) is the betweenness centrality of node v, and α and β are weight coefficients, typically set to α = 0.6 and β = 0.4. The PageRank value is obtained through iterative calculation: PR(v) = (1-d) / N + d × Σ(PR(u) / L(u)), where d is the damping coefficient (0.85), N is the total number of nodes in the graph, and L(u) is the out-degree of node u. The betweenness centrality BC(v) calculates the participation of node v in all shortest paths: BC(v) = Σ(σ(s,t|v) / σ(s,t)), where σ(s,t) is the number of shortest paths from node s to node t, and σ(s,t|v) is the number of shortest paths passing through node v. In the example above, homepage A, due to being accessed multiple times and connecting to multiple pages, has a hub index of 0.42; product details C, as a page pointed to by both the shopping cart and the product list, has a hub index of 0.35; and product list B and shopping cart D have hub indices of 0.15 and 0.08, respectively. The extraction of browsing behavior complexity comprehensively considers multiple dimensions such as path diversity, access depth, and hub node distribution. First, all independent paths in the session are identified, and the standard deviation of path length is calculated as a path diversity index; the proportion of pages with an access depth exceeding 3 levels is counted as a depth access index; and the proportion of high hub nodes (hub index greater than 0.3) in the total number of nodes is calculated as a core page coverage index. These indices are weighted and fused to form the final browsing behavior complexity score, with the formula C = w1 × path diversity + w2 × depth access ratio + w3 × core page coverage, where w1 = 0.3, w2 = 0.4, and w3 = 0.3. The complexity score is normalized and mapped to the [0,1] interval, where 0 represents extremely simple linear browsing and 1 represents extremely complex mesh browsing. In the quality analysis phase, the complexity scores of all sessions for each website are aggregated, and a weighted average is calculated as the website's quality score. The weighting is designed considering session duration and user activity, assigning higher weight to sessions with longer, more in-depth browsing. For example, e-commerce website W1 has 1000 sessions with an average complexity of 0.75 and a quality score of 82; news website W2 has 2000 sessions with an average complexity of 0.45 and a quality score of 58; corporate website W3 has 500 sessions with an average complexity of 0.30 and a quality score of 35; social networking website W4 has 1500 sessions with an average complexity of 0.68 and a quality score of 74; and video website W5 has 800 sessions with an average complexity of 0.55 and a quality score of 61. The final ranking results are: W1 (82 points), W4 (74 points), W5 (61 points), W2 (58 points), and W3 (35 points).

[0029] Through the above steps, the session partitioning of the log data set and the construction of the directed graph structure work synergistically. Session partitioning can divide continuous user behavior sequences into independent analyses, while the directed graph structure transforms the linear access sequences within each session into a network topology. This transformation allows the complex jump relationships between pages to be fully preserved. The calculation of the page hub index relies on the topological characteristics of the directed graph. By analyzing the connection relationships of nodes, important intermediate pages are identified. Compared with traditional access frequency statistics, the hub index can reflect the structural role of a page in the user's browsing path. The extraction of browsing behavior complexity comprehensively utilizes the diversity of user browsing paths and the distribution characteristics of page hub, quantifying behavioral patterns into comparable numerical indicators. This allows quality analysis operations to evaluate website quality based on objective complexity indicators, avoiding the bias caused by simply relying on traffic data. This solves the technical problem of low accuracy in ranking the quality of websites viewed by users in related technologies, achieving the technical effect of improving the accuracy of ranking the quality of websites viewed by users.

[0030] The entity executing the above steps can be a system, such as a website ranking system, or a device, or a controller or processor in a device or system, or a standalone controller or processor, or other processing devices or processing units with similar processing functions, but is not limited to these.

[0031] In an optional embodiment, path structure features are extracted for each access session using the user browsing path and page hub index to obtain the browsing behavior complexity of each access session. Specifically, this includes: traversing the directed graph structure of each access session and performing the following operations on the currently traversed directed graph structure: determining a set of strongly connected nodes from the target directed graph structure where there is a bidirectional path between any two nodes based on the access transition relationship; selecting high hub index nodes from the target directed graph structure where the page hub index is greater than a preset hub index threshold; selecting a set of deep interactive nodes including high hub index nodes from the set of strongly connected nodes, and then... The set of interactive nodes is labeled as a deep interactive region; the number of deep interactive regions in the target directed graph structure is determined, and this number is used as the interaction region quantity feature; the total number of nodes in the deep interactive regions is compared with the total number of nodes in the target directed graph structure to obtain the node ratio, and this node ratio is used as the interaction coverage feature; the interaction region quantity feature and the interaction coverage feature are determined as the session interaction depth feature; the path dispersion feature is extracted from the target directed graph structure using the user browsing path and page hub index; a feature fusion operation is performed on the session interaction depth feature, the path dispersion feature, and the duration of each access session to obtain the browsing behavior complexity of each access session.

[0032] In the above embodiment, we assume a deep analysis of user sessions on a large portal website. Consider a typical user session, whose directed graph structure contains a set of nodes {A, B, C, D, E, F, G, H}, representing the homepage (A), news list (B), news details (C), comment page (D), user center (E), search page (F), recommendation list (G), and advertising page (H), respectively. The edge relationships of this directed graph are as follows: A points to B and F, B points to C and A, C points to D and B, D points to C and E, E points to A and D, F points to G and A, G points to C and F, and H points unidirectionally to A. The Tarjan algorithm is used for strongly connected component identification. This algorithm traverses the graph structure using a depth-first search, maintaining the discovery timestamp and the earliest traceable timestamp for each node. During the traversal, the first strongly connected component was identified, containing nodes {A, B, C, D, E}. These five nodes have a complete bidirectional connected path; for example, starting from A, one can reach C via B, then E via D, and finally return to A from E, forming a closed loop. Simultaneously, one can also return from C via the path D, E, A, and B. The second strongly connected component contains nodes {F, G}, which are mutually reachable. Node H, having only outgoing edges and no incoming edges, constitutes an independent weakly connected component. By analyzing the importance distribution of pages in historical data, a preset hub degree threshold of 0.65 was determined. This threshold was determined based on statistical analysis of 10,000 historical sessions, revealing a distinct bimodal distribution with a natural dividing point at 0.65. Pages with a value higher than this typically function as navigation hubs or content aggregation points. In this example, based on the combined calculations of PageRank and betweenness centrality, the hub indexes for each node are as follows: Homepage A = 0.78, News List B = 0.72, News Details C = 0.68, Comments Page D = 0.55, User Center E = 0.71, Search Page F = 0.62, Recommendation List G = 0.58, and Ad Page H = 0.15. Based on a threshold of 0.65, nodes with high hub indexes include A, B, C, and E.

[0033] In the above embodiments, during the marking of deep interaction regions, each strongly connected component is checked to see if it contains high hub-degree nodes. The first strongly connected component {A,B,C,D,E} contains four high hub-degree nodes (A,B,C,E), so the entire component is marked as a deep interaction region. Although node D has a hub-degree of only 0.55, below the threshold, it is also included in the deep interaction region because it forms a strong connection with other high hub-degree nodes, indicating that the user is deeply browsing and repeatedly visiting this area. The second strongly connected component {F,G} forms a closed loop, but the hub-degree of both nodes does not reach the threshold, so it is not marked as a deep interaction region. This marking strategy ensures that only page groups that truly reflect deep user engagement are identified as deep interaction regions. The calculation of the interaction region quantity feature directly counts the number of marked deep interaction regions. There is only one deep interaction region (containing nodes A,B,C,D,E), so the interaction region quantity feature value is 1. In contrast, another session may identify 3 independent deep interaction regions, with a quantity feature value of 3, reflecting the user's deep exploration behavior among multiple functional modules. For example, some simple browsing sessions may have no deep interaction areas at all, with a quantity feature value of 0, indicating that the user only acquires surface-level information. The interaction coverage feature is obtained by calculating the ratio of the number of nodes in the deep interaction area to the total number of nodes. A deep interaction area contains 5 nodes, and the total number of nodes is 8, so the coverage is calculated as 5 / 8 = 0.625. This ratio reflects the proportion of deep interaction behavior in the entire session. As a comparison, a focused session has 7 deep interaction nodes, a total of 9 nodes, and a coverage of 0.778, indicating that the user has browsed almost all pages in depth; while another exploratory session has only 3 deep interaction nodes, a total of 12 nodes, and a coverage of 0.25, indicating that although the user browsed many pages, they only engaged in deep interaction on a few pages.

[0034] In the above embodiments, the extraction of path dispersion features requires identifying all paths from entry nodes to exit nodes. First, nodes with an in-degree of zero are identified as potential entry points. Although node H has an in-degree of zero, it points to A, and A is the primary access entry point; therefore, A is taken as the actual entry node. Out-degree analysis shows that although nodes D and E have outgoing edges, E is identified as the primary exit node based on the session end marker. All possible paths from A to E include: path 1 is A→B→C→D→E, with a length of 4 and a total hub degree of 3.44; path 2 is A→F→G→C→D→E, with a length of 5 and a total hub degree of 3.26; path 3 is A→B→C→B→A→E (containing a cycle), with an effective length of 5 and a total hub degree of 3.67. Path dispersion is calculated by combining the standard deviation of these path lengths (0.816) and the hub degree-weighted coefficient of variation (0.058), resulting in a final dispersion feature value of 0.437, indicating a moderate level of browsing path diversity. The feature fusion operation comprehensively calculates the session interaction depth feature, path dispersion feature, and session duration. The interaction area quantity feature is 1, the interaction coverage feature is 0.625, and the session interaction depth feature vector formed by the two is [1, 0.625]. The path dispersion feature is 0.437, and the session duration is 42 minutes. The fusion algorithm first segments the duration, with 42 minutes falling into the medium duration range (30-60 minutes), and assigns a duration coefficient of 0.8. Then, the weighted fusion value is calculated: Browsing behavior complexity = 0.3 × (1 / 3) + 0.4 × 0.625 + 0.2 × 0.437 + 0.1 × 0.8 = 0.1 + 0.25 + 0.0874 + 0.08 = 0.5174. Here, 1 / 3 is the normalization of the interaction area quantity (assuming a maximum value of 3), and the weight coefficients are pre-set according to the importance of the features. The final browsing behavior complexity of 0.5174 indicates that the session exhibits a moderate to high level of complexity, with the user engaging in both deep interaction and maintaining a certain breadth of exploration.

[0035] In an optional embodiment, path dispersion features are extracted from the target directed graph structure using user browsing paths and page hub indexes. Specifically, this includes: identifying a first node with an in-degree of zero and a second node with an out-degree of zero from the target directed graph structure; designating the first node as an entry page node and the second node as an exit page node; determining all connected paths from the entry page node to the exit page node and determining the path length of each connected path; determining the path weight value of each connected path based on the path length and page hub index, and designating connected paths with path weight values ​​greater than a preset path weight threshold as main browsing paths, and designating all connected paths other than the main browsing paths as branch browsing paths; determining the average path length of the branch browsing paths based on the path length, and performing a length comparison analysis between the average path length and the path length of the main browsing path to obtain a path length ratio, and using the path length ratio as the path dispersion feature.

[0036] In the above embodiment, we assume we are analyzing a user learning session on an online education platform. The directed graph structure of this session contains 9 nodes: course homepage P, course directory M, video playback page V, practice question page E, answer explanation page S, discussion forum F, personal notes N, knowledge point summary K, and course completion page C. By counting the number of inbound and outbound edges for each node, we generate a node degree statistics table: node P has an in-degree of 0 and an out-degree of 3; node M has an in-degree of 2 and an out-degree of 4; node V has an in-degree of 3 and an out-degree of 2; node E has an in-degree of 2 and an out-degree of 2; node S has an in-degree of 1 and an out-degree of 1; node F has an in-degree of 2 and an out-degree of 1; node N has an in-degree of 1 and an out-degree of 1; node K has an in-degree of 2 and an out-degree of 1; and node C has an in-degree of 3 and an out-degree of 0. Based on degree statistics, node P with an in-degree of zero was selected as the entry page node, reflecting the user's actual learning scenario starting from the course homepage; node C with an out-degree of zero was selected as the exit page node, representing the user's endpoint upon completion of the course. A depth-first search algorithm was used to traverse all possible paths from entry node P to exit node C. The algorithm maintains a path stack, recursively exploring each adjacent node starting from P until reaching C or being unable to proceed further. During the traversal, the algorithm records access markers to avoid loops and saves the complete path sequence. After a complete traversal, the following 5 connected paths were identified: Path 1 is P→M→V→E→S→C, indicating the user is learning according to the standard process; Path 2 is P→M→V→K→C, indicating the user is quickly browsing core content; Path 3 is P→M→E→S→K→C, indicating the user is focusing on completing exercises and reviewing summaries; Path 4 is P→F→K→C, indicating the user is mainly learning through discussion and summarization; Path 5 is P→M→V→N→K→C, indicating the user is taking notes while watching videos. Each path represents a different learning pattern and behavioral characteristic.

[0037] In the above embodiments, the path length is calculated using the number of edges traversed as the metric. Path 1 contains 5 edges and has a length of 5; Path 2 contains 4 edges and has a length of 4; Path 3 contains 5 edges and has a length of 5; Path 4 contains 3 edges and has a length of 3; Path 5 contains 5 edges and has a length of 5. This edge-based length definition intuitively reflects the number of times a user navigates within the website. A longer path indicates that the user has experienced more page transitions, which may imply deeper content exploration or a more complex task completion process. The path weight value is calculated by comprehensively considering both the path length and the hub degree of the nodes in the path. The calculation formula is: Path weight = Path length × (Sum of hub degrees of all nodes in the path / Number of nodes in the path). The hub degree of each node has been calculated using the aforementioned method: P = 0.45, M = 0.82, V = 0.75, E = 0.68, S = 0.52, F = 0.48, N = 0.35, K = 0.71, and C = 0.30. For path 1, the average hub degree is (0.45+0.82+0.75+0.68+0.52+0.30) / 6=0.587, and the path weight is 5×0.587=2.935. For path 2, the average hub degree is (0.45+0.82+0.75+0.71+0.30) / 5=0.606, and the weight is 4×0.606=2.424. For path 3, the average hub degree is 0.596, and the weight is 2.980. For path 4, the average hub degree is 0.485, and the weight is 1.455. For path 5, the average hub degree is 0.576, and the weight is 2.880.

[0038] In the above embodiment, the preset path weight threshold is set to 2.5. This threshold can be determined based on statistical analysis of a large amount of historical session data. Paths with a weight higher than 2.5 usually contain the core functional pages of the website, and user behavior on these paths better reflects their main access purpose. Based on this threshold, path 1 (weight 2.935), path 3 (weight 2.980), and path 5 (weight 2.880) are identified as the main browsing paths, all of which involve the platform's core learning functions. Path 2 (weight 2.424) and path 4 (weight 1.455) are classified as branch browsing paths, representing users' auxiliary or exploratory behaviors. The calculation of the path dispersion feature first determines the average length of the branch paths. The lengths of the two branch paths are 4 and 3, respectively, with an average length of 3.5. There are three main browsing paths with lengths of 5, 5, and 5, respectively. The most representative path 1 is selected as the main path representative, with a length of 5. The path length ratio is calculated as 5 / 3.5 = 1.43, and this ratio is used as the path dispersion feature. A value greater than 1 indicates that the main path is longer than the branch paths, meaning users invest more depth in browsing core functions. A ratio of 1.43 represents moderate dispersion, neither overly concentrated on a single path nor excessively dispersed, reflecting that users explore appropriately while maintaining the main learning focus. In contrast, a ratio close to 1 indicates a highly dispersed browsing pattern, while a ratio greater than 2 indicates a highly concentrated browsing pattern. This feature effectively quantifies the focus of user browsing behavior, providing an important behavioral dimension indicator for subsequent website quality assessment.

[0039] In an optional embodiment, a feature fusion operation is performed on the session interaction depth features, path dispersion features, and the duration of each access session to obtain the browsing behavior complexity of each access session. Specifically, this includes: determining the target duration interval to which the duration of each access session belongs; performing numerical interval analysis on the session interaction depth features to obtain a first value for the interaction area quantity feature and a second value for the interaction coverage feature; classifying the session interaction depth features into surface browsing level and deep exploration level based on the first and second values ​​to obtain the interaction depth classification result; performing numerical range analysis on the path dispersion features to obtain a third value for the path dispersion features; using the third value to classify the path dispersion features into linear browsing mode and dispersed browsing mode to obtain the path mode classification result; and performing a feature fusion operation on the interaction depth classification result, the path mode classification result, and the target duration interval to obtain the browsing behavior complexity of each access session.

[0040] In the above embodiment, it is assumed that a comprehensive analysis of user shopping sessions on an e-commerce platform is performed. Based on statistical analysis of millions of user behavior data, the criteria for dividing the duration intervals were determined. The first preset duration threshold was set at 15 minutes, and the second preset duration threshold was set at 45 minutes. These two thresholds were determined based on in-depth mining of user behavior patterns. Data analysis showed that 78% of sessions within 15 minutes were quick browsing or direct purchase behavior with a clear goal; sessions between 15 and 45 minutes showed obvious comparative selection characteristics; and sessions exceeding 45 minutes showed characteristics of in-depth exploration and iterative decision-making. Based on this distribution pattern, the session duration was divided into three intervals: a short-term trial interval (less than 15 minutes), a medium-term activation interval (15-45 minutes), and a long-term in-depth exploration interval (greater than 45 minutes). Consider three typical user session examples. The first session lasted 12 minutes, during which the user quickly browsed the product list and placed an order, representing a short-term exploratory phase. The second session lasted 35 minutes, during which the user compared multiple products, viewed reviews and details, representing a mid-term stimulating phase. The third session lasted 72 minutes, during which the user not only browsed products but also viewed store information, after-sales policies, and related recommendations, representing a long-term in-depth exploration phase. This division of duration provides an important time dimension reference for subsequent feature fusion. For the numerical analysis of session interaction depth features, two key indicators were extracted. Taking the second session as an example, the first value of the interaction area quantity feature was 2, indicating the identification of 2 deep interaction areas; the second value of the interaction coverage feature was 0.42, indicating that deep interaction nodes accounted for 42% of the total number of nodes. Cluster analysis of historical data revealed a clear bimodal distribution of interaction depth, with a natural boundary at a coverage of 0.35. Therefore, a comprehensive judgment rule was established: when the interaction coverage is greater than or equal to 0.35 and the number of interaction areas is greater than or equal to 2, it is judged as a deep exploration level; otherwise, it is judged as a surface browsing level. According to this rule, the second session's interaction coverage of 0.42 exceeds the threshold of 0.35, and it has two interaction areas, so it is classified as a deep exploration level.

[0041] In the above embodiments, the logic for determining the interaction depth classification fully considers the synergistic effect of two dimensions. High coverage but a small number of interactive areas may indicate that the user is looping within a single functional module, lacking breadth of exploration; multiple interactive areas but low coverage may indicate that the user has tried multiple functions but only superficially. Only when both conditions are met simultaneously can it be determined that the user has engaged in true in-depth exploration. For example, a session with a coverage of 0.38 but only one interactive area is still considered surface browsing; another session with three interactive areas but only 0.28 coverage is also considered surface browsing. The analysis of path dispersion characteristics uses a similar method. The third value of the path dispersion characteristic for the second session is 1.65, which is calculated using the aforementioned ratio of main path to branch path length. Based on the distribution analysis of a large amount of historical data, 1.5 is determined as the boundary value between linear and dispersed browsing. When the path dispersion is less than 1.5, it indicates that the user's browsing path is relatively concentrated, and they tend to operate along the main functional lines, which is classified as a linear browsing mode. When the path dispersion is greater than or equal to 1.5, it indicates that the user switches between multiple paths to explore, exhibiting divergent browsing characteristics, which is classified as a dispersed browsing mode. The dispersion of the second session, 1.65, exceeds the threshold and is therefore judged as a dispersed browsing mode.

[0042] In the above embodiment, the feature fusion operation comprehensively calculates the classification results of the three dimensions. For the second session, its interaction depth classification result is deep exploration level, its path pattern classification result is scattered browsing mode, and its target duration interval is mid-term activation interval. A multi-dimensional weighted fusion strategy is adopted. First, a base score is assigned to each classification result: 30 points for deep exploration level (10 points for surface browsing), 25 points for scattered browsing mode (15 points for linear browsing), and 20 points for mid-term activation interval (10 points for short-term exploration and 30 points for long-term deep exploration). Then, adjustments are made according to the synergistic effect of different combinations. When deep exploration and scattered browsing occur simultaneously, it indicates that the user has conducted a comprehensive and in-depth exploration, and an additional 15 points are added; when the duration interval matches the interaction depth well (such as mid-term duration corresponding to deep exploration), another 10 points are added. Therefore, the browsing behavior complexity of the second session is calculated as follows: 30 (deep exploration) + 25 (dispersed browsing) + 20 (mid-term stimulation) + 15 (deep and dispersed collaboration) + 10 (duration matching) = 100 points. This is then normalized to the 0-100 range, resulting in a final score of 78 points. This score indicates that the session exhibits high behavioral complexity, suggesting that the user engaged in deep and diverse browsing exploration. Such sessions typically correspond to high-value potential customer behaviors.

[0043] In an optional embodiment, a feature fusion operation is performed on the interaction depth classification result, the path pattern classification result, and the target duration interval to obtain the browsing behavior complexity of each access session. Specifically, this includes: when the interaction depth classification result is at the deep exploration level and the path pattern classification result is at the scattered browsing mode, multiplying the interaction coverage feature by a first preset coefficient to obtain the deep interaction contribution value; when the interaction depth classification result is at the surface browsing level and the path pattern classification result is at the linear browsing mode, multiplying the interaction area quantity feature by a second preset coefficient to obtain the surface interaction contribution value. The number is less than the first preset coefficient; the duration adjustment factor is determined according to the target duration interval, wherein: when the target duration interval is a short-term trial interval, the duration adjustment factor is determined as the first adjustment value; when the target duration interval is a medium-term activation interval, the duration adjustment factor is determined as the second adjustment value; when the target duration interval is a long-term in-depth exploration interval, the duration adjustment factor is determined as the third adjustment value, the first adjustment value is less than the second adjustment value, and the second adjustment value is less than the third adjustment value; the deep interaction contribution value or surface interaction contribution value is multiplied by the duration adjustment factor and the path dispersion feature to obtain the browsing behavior complexity.

[0044] In the above embodiment, it is assumed that through in-depth analysis of 100,000 user sessions, the quantitative relationship of the contribution of different browsing modes to website quality was determined. The conversion rate and dwell value of users engaging in deep exploration are significantly higher than those engaging in surface browsing; the former's average contribution is 1.875 times that of the latter. Based on this finding, a first preset coefficient of 1.8 is set to amplify the weight of deep interaction behavior; a second preset coefficient of 0.96 is set to moderately account for the contribution of surface browsing. The first preset coefficient of 1.8 is significantly greater than the second preset coefficient of 0.96. This differentiated setting ensures that deep interaction behavior receives a higher weight in quality assessment, consistent with the distribution pattern of actual business value. Consider a deep exploration session, whose interaction depth classification result is deep exploration level and path pattern classification result is scattered browsing mode. The interaction coverage feature value of this session is 0.52, indicating that more than half of the pages participated in deep interaction. The calculation process for the deep interaction contribution value is: 0.52 multiplied by the first preset coefficient of 1.8, resulting in 0.936. This value reflects the intensity of a user's in-depth exploration within the website; a higher contribution value indicates a stronger user interest and engagement with the website content. This calculation method transforms the original coverage metric into a business-meaning contribution measure. For surface-level browsing sessions, the interaction depth classification result is the surface-level browsing level, and the path pattern classification result is the linear browsing pattern. Assuming the session identifies two interaction areas, the surface interaction contribution value is calculated as 2 multiplied by the second preset coefficient 0.96, resulting in 1.92. Although the session has multiple interaction areas, due to its overall surface-level browsing characteristics, its contribution value is adjusted using a smaller coefficient. This approach avoids the one-sidedness of judging quality solely by the number of interaction areas and more accurately reflects the user's actual level of engagement.

[0045] In the above embodiments, the duration adjustment factor is set based on the principle that the value of user behavior increases with time investment. The first adjustment value is set to 0.7, corresponding to the short-term trial period, representing a value discount for quick browsing behavior; the second adjustment value is 1.0, corresponding to the mid-term activation period, serving as a standard reference value; and the third adjustment value is 1.3, corresponding to the long-term in-depth exploration period, reflecting the value premium for long-term in-depth participation. These three adjustment values ​​show an increasing relationship, with 0.7 less than 1.0 less than 1.3, accurately reflecting the positive correlation between user time investment and behavioral value. Statistical data shows that the conversion probability of users engaging in long-term in-depth exploration is 1.86 times that of users engaging in short-term trial, and the ratio of the adjustment factor 1.3 / 0.7 ≈ 1.86 precisely matches this actual pattern. The final calculation of browsing behavior complexity uses a three-factor product formula: complexity equals the interaction contribution value multiplied by the duration adjustment factor, and then multiplied by the path dispersion characteristic. For the aforementioned deep exploration session, assuming it's in a long-term deep exploration range with a path dispersion feature of 1.75, the browsing behavior complexity is calculated as: 0.936 multiplied by 1.3 multiplied by 1.75, resulting in 2.1294. This higher complexity value indicates that the session exhibits deep, persistent, and diverse high-quality browsing characteristics. In contrast, the surface browsing session, assuming it's in a short-term exploration range with a path dispersion feature of 1.2, has a complexity calculated as: 1.92 multiplied by 0.7 multiplied by 1.2, resulting in 1.6128. The difference in complexity between the two sessions reflects the value distinction between different browsing modes.

[0046] In an optional embodiment, determining the target duration interval to which the duration of each access session belongs specifically includes: obtaining a first preset duration threshold and a second preset duration threshold, wherein the first preset duration threshold is less than the second preset duration threshold; performing duration comparison analysis with the first preset duration threshold and the second preset duration threshold respectively to obtain duration comparison results; when the duration is determined to be less than the first preset duration threshold based on the duration comparison results, the access session corresponding to the duration is divided into a short-term trial interval; when the duration is determined to be greater than or equal to the first preset duration threshold and less than the second preset duration threshold based on the duration comparison results, the access session corresponding to the duration is divided into a medium-term activation interval; when the duration is determined to be greater than or equal to the second preset duration threshold based on the duration comparison results, the access session corresponding to the duration is divided into a long-term deep exploration interval.

[0047] In the above embodiment, it is assumed that the standard for setting the duration threshold was determined through statistical analysis of massive user behavior data. A duration distribution analysis of 80 million user sessions on a video website over the past year revealed a clear three-segment distribution characteristic. Data shows two natural dividing points at 20 minutes and 90 minutes, corresponding to significant shifts in user behavior patterns. Based on this finding and adjustments made in conjunction with practical experience, a first preset duration threshold of 20 minutes and a second preset duration threshold of 90 minutes were set. These two thresholds are located near the 33rd and 67th percentiles of the user behavior distribution, effectively dividing the user group into three groups with different behavioral characteristics. The duration comparison analysis employs a dual-judgment logic. First, the duration of the session is compared with the first preset duration threshold of 20 minutes. If it is less than 20 minutes, it is directly determined as a short-term trial period; if it is greater than or equal to 20 minutes, it is further compared with the second preset duration threshold of 90 minutes. Less than 90 minutes is determined as a medium-term activation period, and greater than or equal to 90 minutes is determined as a long-term deep exploration period. This serial comparison method ensures that each session can be accurately categorized into a unique duration range. During the comparison process, duration calculations are performed with millisecond-level precision, avoiding ambiguity in boundary value determination. Consider a user session lasting 12 minutes. Comparing 12 minutes with a first preset duration threshold of 20 minutes, 12 is less than 20, therefore this session is directly classified as a short-term exploratory range. These sessions typically exhibit clear exploratory characteristics: users visit an average of 8 pages, the page navigation speed is relatively fast, and the average time spent on each page is only 90 seconds. In-depth analysis reveals that 65% of users in the short-term exploratory range are first-time visitors to the website, primarily performing quick content scanning and feature familiarization. Another 25% of users come with a specific goal, directly navigating to the target content to complete a specific task before leaving. This behavioral pattern determines that short-term exploratory range sessions have a small area of ​​deep interaction, averaging only 0.3.

[0048] In the above embodiments, for sessions lasting 58 minutes, a complete dual comparison process is executed. First, 58 minutes exceeds the first preset duration threshold of 20 minutes, thus not belonging to the short-term trial range; further comparison shows that 58 minutes is less than the second preset duration threshold of 90 minutes, therefore it is classified as a mid-term activation range. Sessions in the mid-term activation range exhibit characteristics of both exploration and focus. During this time period, users visit an average of 35 pages, forming 2.1 deep interaction areas, with an interaction coverage of 0.42. Typical behavioral patterns include: multiple filtering and comparisons on the video list page, selecting content and watching a 20-30 minute video in its entirety, interspersed with viewing comments, related recommendations, and other auxiliary information. This behavior indicates that the user has moved beyond the initial understanding stage and begun to deeply experience the core functions of the website. Sessions lasting 135 minutes exhibit typical characteristics of long-term deep exploration. In the judgment process, 135 minutes is first greater than 20 minutes, and further comparison reveals that it is also greater than 90 minutes, therefore it is classified as a long-term deep exploration range. Users in these sessions exhibited extremely high engagement, visiting an average of 68 pages and forming 4.3 deep interaction zones, resulting in an interaction coverage of 0.71. Users not only watched multiple complete videos but also actively participated in deeper functions such as commenting, sharing, and managing their personal profiles. Notably, 82% of sessions in the long-term deep exploration zone included at least one instance of returning to the website via an external link, indicating that users used this website as a central node for sustained activity. The criteria for identifying deep interaction zones were more stringent in this zone, requiring at least 5 nodes within strongly connected components and an average dwell time exceeding 3 minutes to ensure that truly valuable deep interaction behaviors were identified.

[0049] In the above embodiments, statistical analysis of the three intervals verified the rationality of the division. The short-term exploratory interval had an average of 0.3 interactive areas per session, with an average coverage of 0.15, reflecting the characteristics of rapid browsing; the mid-term stimulating interval had an average of 2.1 interactive areas, with a coverage of 0.42, reflecting moderately in-depth exploration; and the long-term deep exploration interval had an average of 4.3 interactive areas, with a coverage of 0.71, demonstrating deep engagement. More noteworthy is the difference in user behavior paths. In the short-term exploratory interval, user access paths were relatively simple and direct, with an average of only 2.3 independent paths per session and few intersections between paths; the mid-term stimulating interval had an average of 5.8 paths per session, showing a clear path interweaving phenomenon, indicating that users were comparing and choosing between different functional modules; the long-term deep exploration interval had 9.6 paths, forming a complex network structure, with users repeatedly switching between multiple content sections, reflecting comprehensive and in-depth exploration behavior. These differences in behavioral patterns directly affect the distribution of page importance. Long-term deep access naturally increases the access frequency and transit role of core pages, correspondingly increasing their hub index. This natural evolution based on actual behavior patterns fully demonstrates that time-based interval division can effectively distinguish user behavior patterns at different depths, providing reliable time-dimensional features for subsequent quality assessment.

[0050] In an optional embodiment, a quality analysis operation is performed on the website to be ranked using browsing behavior complexity to obtain the quality ranking result of the website to be ranked. Specifically, this includes: traversing each access session of the website to be ranked, and performing the following operations on the currently traversed access session: performing interaction depth analysis on the target access session, determining the interaction depth coefficient of the target access session when the interaction depth classification result of the target access session is surface browsing level and the duration of the target access session is less than a preset effective duration threshold, and the target access session is the currently traversed access session; performing interaction quality analysis on the target access session, determining the interaction quality coefficient of the target access session when the browsing behavior complexity of the target access session is greater than a preset complexity threshold and the number of nodes in the directed graph structure of the target access session that have a page hub degree index greater than a preset core hub degree threshold; determining the session quality value of the target access session based on the interaction depth coefficient and the interaction quality coefficient, summing the session quality values ​​of all access sessions belonging to the same website to be ranked to obtain the total website quality score of the website to be ranked; and sorting all websites to be ranked in descending order based on the total website quality score to generate the quality ranking result.

[0051] In the above embodiment, it is assumed that a comprehensive evaluation of the website quality of a certain information platform is being conducted. The core of interaction depth analysis lies in identifying low-value sessions and assigning them corresponding discount coefficients. A preset effective duration threshold is set to 8 minutes. This threshold can be determined based on statistical analysis of user behavior. Studies have found that 87% of sessions shorter than 8 minutes are invalid browsing or accidental clicks. When the interaction depth classification result of a target access session is detected as surface browsing, and the duration is only 6 minutes, less than the effective duration threshold of 8 minutes, the low-value session judgment mechanism is triggered. At this time, the interaction depth coefficient of the session is determined to be 0.25. This low coefficient reflects the limited contribution of the session to the website quality. The judgment process first checks the interaction depth classification flag. If it is surface browsing, it enters the duration judgment branch, comparing the actual duration with the threshold. If the condition is met, the corresponding depth coefficient is output. For sessions with an interaction depth of deep exploration or a duration exceeding the threshold, a standard depth coefficient of 1.0 is assigned to ensure that valuable sessions are fully represented. Interaction quality analysis focuses on the identification and weighting of high-value sessions. A preset complexity threshold of 65 points and a preset core hub degree threshold of 0.75 are set. When the browsing behavior complexity of a target access session reaches 72 points, exceeding the threshold of 65 points, it indicates that the user has engaged in diverse and in-depth browsing. Simultaneously, a core page node with a hub degree of 0.82 is detected in the session's directed graph structure, exceeding the core hub degree threshold of 0.75, indicating that the user has accessed a key functional page of the website. When both conditions are met, the interaction quality coefficient of the session is determined to be 0.85. This high quality coefficient reflects the value of in-depth and complex browsing behavior. The judgment process employs a dual verification mechanism: first, sessions that meet the complexity threshold are selected; then, all nodes in their directed graph structure are traversed to check for any nodes exceeding the core hub degree threshold. Only when both conditions are met is a high quality coefficient assigned. For ordinary sessions that do not meet the conditions, the quality coefficient is set to a baseline value of 0.5.

[0052] In the above embodiments, the session quality score is calculated by multiplying the interaction depth coefficient and the interaction quality coefficient. The calculation process is illustrated using five typical sessions as examples: the first session is a deep exploration session with an interaction depth coefficient of 1.0, an interaction quality coefficient of 0.85, and a quality value of 0.85; the second session is a surface-level quick browsing session with a depth coefficient of 0.25, a quality coefficient of 0.5, and a quality value of 0.125; the third session is a medium-depth exploration session with a depth coefficient of 1.0, a quality coefficient of 0.5, and a quality value of 0.5; the fourth session is a long session but lacking depth with a depth coefficient of 1.0, a quality coefficient of 0.3, and a quality value of 0.3; and the fifth session is a short-term, efficient session with a depth coefficient of 0.25, a quality coefficient of 0.85, and a quality value of 0.2125. This calculation method ensures that only sessions possessing both depth and quality can obtain high scores. The calculation of the website's total quality score requires aggregating the quality values ​​of all sessions on the website. For example, a news website generated 1200 access sessions during the evaluation period. Each session was iterated over, and its quality score was calculated and accumulated. 285 sessions were classified as high-quality (deep), with an average quality score of 0.82, contributing 233.7 points; 420 sessions were classified as medium-quality (medium), with an average quality score of 0.48, contributing 201.6 points; and 495 sessions were classified as low-quality (low), with an average quality score of 0.15, contributing 74.25 points. The final quality score for the website was 509.55. This accumulation method fully considers the differences in contribution from sessions of different quality levels, avoiding the biases caused by simple counting. After calculating the total quality score for all websites to be ranked, a quicksort algorithm was used to generate the final ranking. In a test set containing 500 websites, the top ten results were as follows: 1st place: Technology news website (892 points); 2nd place: Financial portal website (856 points); 3rd place: Education and learning platform (823 points); 4th place: Video entertainment website (781 points); 5th place: E-commerce shopping platform (759 points); 6th place: Social forum website (724 points); 7th place: News website (698 points); 8th place: Lifestyle service platform (652 points); 9th place: Game and entertainment website (621 points); 10th place: Enterprise information website (598 points). The sorting process employed a divide-and-conquer strategy, selecting the median as the benchmark value to recursively divide and sort the website set. The time complexity was O(nlogn), ensuring efficient processing of large-scale website sets.

[0053] It should also be noted that the examples of actual values ​​for the above-mentioned parameters are merely exemplary embodiments, and the examples of actual values ​​for the above-mentioned parameters are not limited to the examples mentioned above.

[0054] In this application's embodiments, the session partitioning of the log data set and the construction of the directed graph structure work synergistically. Session partitioning can divide continuous user behavior sequences into independent analyses, while the directed graph structure transforms the linear access sequences within each session into a mesh topology. This transformation allows the complex jump relationships between pages to be fully preserved. The calculation of the page hub index relies on the topological characteristics of the directed graph. By analyzing the connection relationships of nodes, important intermediate pages are identified. Compared with traditional access frequency statistics, the hub index can reflect the structural role of a page in the user's browsing path. The extraction of browsing behavior complexity comprehensively utilizes the diversity of user browsing paths and the distribution characteristics of page hub, quantifying behavioral patterns into comparable numerical indicators. This allows quality analysis operations to evaluate website quality based on objective complexity indicators, avoiding the bias caused by simply relying on traffic data.

[0055] It should be noted that the embodiments described above are only some embodiments of this application, and not all embodiments. The present application will be described in detail below with reference to specific embodiments.

[0056] The website ranking system in the embodiments of this invention is described below from the perspective of hardware processing. (See attached document.) Figure 2 , Figure 2 This is a schematic diagram of the physical device structure of a website ranking system in this application embodiment.

[0057] It should be noted that, Figure 2 The structure of the website ranking system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0058] like Figure 2 As shown, the website ranking system includes a Central Processing Unit (CPU) 201, which can perform various appropriate actions and processes according to a program stored in Read-Only Memory (ROM) 202 or a program loaded from storage portion 208 into Random Access Memory (RAM) 203, such as performing the methods described in the above embodiments. The RAM 203 also stores... It contains various programs and data required for system operation. The CPU 201, ROM 202, and RAM 203 are interconnected via bus 204. Input / output (I / O) interface 205 is also connected to bus 204.

[0059] The following components are connected to I / O interface 205: input section 206 including audio input devices, push-button switches, etc.; output section 207 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 208 including a hard disk, etc.; and communication section 209 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 209 performs communication processing via a network such as the Internet. Drive 210 is also connected to I / O interface 205 as needed. Removable media 211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 210 as needed so that computer programs read from them can be installed into storage section 208 as needed.

[0060] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 209, and / or installed from removable medium 211. When the computer program is executed by central processing unit (CPU) 201, it performs the various functions defined in the present invention.

[0061] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0062] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0063] Specifically, the website ranking system in this embodiment includes a processor and a memory. The memory stores a computer program, and when the computer program is executed by the processor, it implements the website ranking method based on big data analysis provided in the above embodiment.

[0064] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the website ranking system described in the above embodiments; or it may exist independently and not assembled into the website ranking system. The storage medium carries one or more computer programs that, when executed by a processor of the website ranking system, cause the website ranking system to implement the big data analysis-based website ranking method provided in the above embodiments.

[0065] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0066] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for ranking websites based on big data analysis, characterized in that, The method comprises the following steps: acquiring a log data set of a plurality of users accessing a website to be ranked within a target time period, and performing access session division on the log data set according to a preset session division condition to obtain an access session set of each user in the plurality of users; performing time sequence structure feature analysis on the access session set of each user, determining a page access sequence in each access session included in the access session set of each user, and constructing a directed graph structure associated with page jumps of each access session according to the page access sequence, each node in the directed graph structure corresponding to each access page in the page access sequence, and a directed edge in the directed graph structure being an access transfer relationship of the page jump; extracting a user browsing path from the directed graph structure by using the access transfer relationship, and determining a page pivot degree index of each node in the user browsing path, the page pivot degree index representing a transit importance degree of the access page corresponding to each node in the user browsing path; performing path structure feature extraction on each access session by using the user browsing path and the page pivot degree index to obtain a browsing behavior complexity of each access session, and performing a quality analysis operation on the website to be ranked by using the browsing behavior complexity to obtain a quality ranking result of the website to be ranked.

2. The method of claim 1, wherein, The path structure feature extraction on each access session by using the user browsing path and the page pivot degree index to obtain the browsing behavior complexity of each access session specifically comprises: traversing the directed graph structure of each access session, and performing the following operations on the currently traversed directed graph structure: determining a strongly connected node set between any two nodes existing a bidirectional connected path in the target directed graph structure according to the access transfer relationship, the target directed graph structure being the currently traversed directed graph structure; filtering out a high pivot degree node with a page pivot degree index greater than a preset pivot degree threshold from the target directed graph structure; filtering out a deep interaction node set including the high pivot degree node from the strongly connected node set, and marking the deep interaction node set as a deep interaction area; determining a number of the deep interaction areas in the target directed graph structure, and taking the number as an interaction area number feature; performing total number comparison analysis on a total number of nodes in the deep interaction area and a total number of nodes in the target directed graph structure to obtain a node total number ratio, and taking the node total number ratio as an interaction coverage degree feature; determining the interaction area number feature and the interaction coverage degree feature as a session interaction depth feature; extracting a path dispersion degree feature from the target directed graph structure by using the user browsing path and the page pivot degree index; performing feature fusion operations on the session interaction depth feature, the path dispersion degree feature, and a duration of each access session to obtain the browsing behavior complexity of each access session.

3. The method of claim 2, wherein, The path dispersion degree feature is extracted from the target directed graph structure by using the user browsing path and the page hub degree indicator, and specifically includes the following steps: determining a first node with an in-degree of zero from the target directed graph structure, and determining a second node with an out-degree of zero from the target directed graph structure; taking the first node as an entry page node and the second node as an exit page node; determining all connected paths from the entry page node to the exit page node, and determining the path length of each connected path in the all connected paths; determining the path weight value of each connected path according to the path length and the page hub degree indicator, and determining a main browsing path as the connected path with a path weight value greater than a preset path weight threshold, and determining a branch browsing path as the connected path other than the main browsing path in the all connected paths; determining the average path length of the branch browsing path according to the path length, and performing length comparison analysis on the average path length and the path length of the main browsing path to obtain a path length ratio value, and taking the path length ratio value as the path dispersion degree feature.

4. The method of claim 2, wherein, The session interaction depth feature, the path dispersion degree feature, and the duration of each access session are subjected to feature fusion operation to obtain the browsing behavior complexity of each access session, and specifically includes the following steps: determining a target time interval to which the duration of each access session belongs; performing numerical interval analysis on the session interaction depth feature to obtain a first value of the interaction region number feature and a second value of the interaction coverage degree feature; dividing the session interaction depth feature into a surface browsing level and a deep exploration level according to the first value and the second value to obtain an interaction depth classification result; performing numerical range analysis on the path dispersion degree feature to obtain a third value of the path dispersion degree feature; dividing the path dispersion degree feature into a linear browsing mode and a dispersed browsing mode by using the third value to obtain a path mode classification result; performing feature fusion operation on the interaction depth classification result, the path mode classification result, and the target time interval to obtain the browsing behavior complexity of each access session.

5. The method of claim 4, wherein, The session interaction depth feature, the path dispersion degree feature, and the duration of each access session are subjected to feature fusion operation to obtain the browsing behavior complexity of each access session, and specifically includes the following steps: when the interaction depth classification result is the deep exploration level and the path mode classification result is the dispersed browsing mode, multiplying the interaction coverage degree feature by a first preset coefficient to obtain a deep interaction contribution value; when the interaction depth classification result is the surface browsing level and the path mode classification result is the linear browsing mode, multiplying the interaction region number feature by a second preset coefficient to obtain a surface interaction contribution value, and the second preset coefficient is less than the first preset coefficient; determining a time length adjustment factor according to the target time interval, wherein: When the target time length interval is a short-term exploration interval, the time length adjustment factor is determined as a first adjustment value; when the target time length interval is a medium-term stimulation interval, the time length adjustment factor is determined as a second adjustment value; and when the target time length interval is a long-term deep exploration interval, the time length adjustment factor is determined as a third adjustment value, the first adjustment value being smaller than the second adjustment value, and the second adjustment value being smaller than the third adjustment value; The depth interaction contribution value or the surface interaction contribution value is multiplied by the time length adjustment factor and the path dispersion feature to obtain the browsing behavior complexity.

6. The method of claim 5, wherein, The determination of the target time length interval to which the duration of each access session belongs specifically includes: A first preset time length threshold and a second preset time length threshold are obtained, the first preset time length threshold being smaller than the second preset time length threshold; The duration is compared with the first preset time length threshold and the second preset time length threshold respectively to obtain a time length comparison result; When the duration is determined to be smaller than the first preset time length threshold according to the time length comparison result, the access session corresponding to the duration is divided into the short-term exploration interval; When the duration is determined to be greater than or equal to the first preset time length threshold and smaller than the second preset time length threshold according to the time length comparison result, the access session corresponding to the duration is divided into the medium-term stimulation interval; When the duration is determined to be greater than or equal to the second preset time length threshold according to the time length comparison result, the access session corresponding to the duration is divided into the long-term deep exploration interval.

7. The method of claim 4, wherein, The quality analysis operation on the to-be-ranked websites is performed by using the browsing behavior complexity to obtain a quality ranking result of the to-be-ranked websites, and specifically includes: Each access session of the to-be-ranked websites is traversed, and the following operations are performed on the currently traversed access session: When the interaction depth classification result of a target access session is the surface browsing level and the duration of the target access session is smaller than a preset effective time length threshold, an interaction depth coefficient of the target access session is determined, the target access session being the currently traversed access session; When the browsing behavior complexity of the target access session is greater than a preset complexity threshold and a page hub degree index of a node existing in the directed graph structure of the target access session is greater than a preset core hub degree threshold, an interaction quality coefficient of the target access session is determined; A session quality value of the target access session is determined according to the interaction depth coefficient and the interaction quality coefficient, and session quality values of all access sessions belonging to the same to-be-ranked website are accumulated and summed to obtain a website quality total score of the to-be-ranked website; All to-be-ranked websites are arranged in descending order according to the website quality total score to generate the quality ranking result.

8. A website ranking system, characterized by, The website ranking system comprises one or more processors and a memory; the memory is coupled with the one or more processors, the memory is configured to store computer program code, the computer program code comprises computer instructions, the one or more processors invoke the computer instructions to enable the website ranking system to perform the method in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instructions run on the website ranking system, enable the website ranking system to perform the method in any one of claims 1-7.

10. A computer program product, characterised in that, When the computer program product runs on the website ranking system, enable the website ranking system to perform the method in any one of claims 1-7.

Images