An e-commerce product recommendation method and system based on an artificial intelligence algorithm

Abstract

The application discloses an e-commerce product recommendation method and system based on an artificial intelligence algorithm, and relates to the technical field of product recommendation.The method comprises the following steps: a user-behavior path flow tree is constructed by adopting a multi-criterion split mechanism, and a node path set is generated; a user portrait set and a commodity portrait set are respectively sliced and reorganized, and are cross combined; a user belonging behavior cluster graph is constructed, cross commodity nodes in the cross mapping set in the candidate are projected into the anti-interleaving graph structure between the behavior clusters, and an interaction point is extracted as a candidate recommendation set; the path probability density mapping corresponding to each cross commodity node in the candidate recommendation set is utilized to perform weight regression calculation on each node of the user-behavior path flow tree path, and a commodity recommendation sequence is generated according to the maximum path passing probability.The application constructs a recommendation process with behavior chain criterion and coupling graph mapping, improves user experience, and enhances the conversion efficiency and content distribution capability of the platform.

Description

Technical Field

[0001] This invention relates to the field of product recommendation technology, and in particular to an e-commerce product recommendation method and system based on artificial intelligence algorithms. Background Technology

[0002] With the booming development of e-commerce platforms, personalized recommendation systems have gradually become a core technology for improving user satisfaction and purchase conversion rates. Early recommendation methods primarily relied on strategies such as collaborative filtering, content matching, or rating matrix modeling to identify potentially interesting products through similarity measurements. However, these methods have significant limitations when facing data sparsity, the cold start problem, and the complexity of user behavior. In recent years, the development of artificial intelligence, especially deep learning and graph neural networks, has provided new ideas for mining user behavior trajectories and modeling behavioral evolution paths. E-commerce recommendation technology has gradually shifted from "static preference modeling" to "dynamic behavior modeling." Under this trend, integrating multi-dimensional behavioral features, strengthening the interactive modeling of user profiles and product profiles, and building a more flexible and interpretable recommendation framework have become important directions for the development of current intelligent recommendation systems.

[0003] For example, CN118037401A, a knowledge graph-based agricultural e-commerce recommendation system, focuses on data collection and structured modeling in agricultural e-commerce scenarios. Its user profiling module builds basic interest models based on users' historical behavior, while the knowledge graph construction module primarily focuses on knowledge modeling around agricultural product information. The system provides auxiliary recommendation functions through multiple modules such as product quality, nutritional analysis, and market monitoring. However, this technical solution mainly relies on the static representation capabilities of knowledge graphs, resulting in relatively coarse-grained user behavior feature modeling. It only involves summarizing user behavior and analyzing basic preferences, lacking a dynamic modeling mechanism for the evolution of user behavior paths. Furthermore, this method does not introduce misaligned cross-mapping, behavior cluster coupling, or weighting mechanisms based on path probability in constructing recommendation paths between users and products. This makes it difficult for its recommendation results to fully reflect the evolutionary trends of user behavior, especially when facing multi-behavioral coupled recommendation tasks, where its generalization ability and timeliness remain insufficient.

[0004] CN117710064B describes an AI-based e-commerce product recommendation method and apparatus. This method generates a target value matrix based on user purchase records, then performs a two-stage user group matching process combining price and brand similarity, and finally generates product recommendation probabilities based on a modified model. While this method improves recommendation accuracy to some extent, its core algorithm relies heavily on purchase behavior data, neglecting the significant value of non-purchase behaviors (such as browsing, dwell time, and navigation order) in user intent recognition. Furthermore, its user profile construction method is relatively simplistic, failing to consider cross-temporal behavioral evolution patterns or introduce path probabilities or graph structures to dynamically represent user behavior. Additionally, this method determines user groups through static similarity, easily overlooking the coupling relationships between complex behaviors and potential temporal preferences. Summary of the Invention

[0005] In view of the problems of existing e-commerce recommendation methods, such as the single dimension of user behavior modeling and the lack of flexible cross-mapping and behavior cluster coupling mechanism between user profiles and product profiles, this invention is proposed.

[0006] Therefore, the problem to be solved by this invention is how to achieve dynamic modeling and intelligent recommendation of user-product interaction relationships by constructing process paths, graph structures and cross-mapping mechanisms.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, the present invention provides an e-commerce product recommendation method based on artificial intelligence algorithms, comprising: extracting multi-state feature parameters based on user historical behavior data; constructing a user-behavior path flow tree using a multi-criteria splitting mechanism to generate a set of node paths with behavioral evolution patterns; slicing and recombining the user profile set and the product profile set under multi-dimensional features, and cross-combining them through predefined misalignment mapping rules to generate a candidate cross-mapping set; constructing a behavior cluster graph of the user, projecting the cross-product nodes in the node path set and the candidate cross-mapping set onto the anti-interlaced graph structure between behavior clusters, and extracting interaction points with high transition weights and behavioral coupling as a candidate recommendation set; using the path probability density mapping corresponding to each cross-product node in the candidate recommendation set, reversely looking up the associated product-path probability mapping table, performing weight regression calculation on each node of the user-behavior path flow tree path, and generating a product recommendation sequence based on the maximum path traversal probability.

[0009] As a preferred embodiment of the e-commerce product recommendation method based on artificial intelligence algorithms described in this invention, the construction of the user-behavior path flow tree includes: extracting page dwell time parameters, browsing frequency parameters, and jump order codes for each behavior event based on user behavior log data recorded in the e-commerce platform, and constructing a multi-dimensional state vector sequence according to the time series; setting a set of transition thresholds based on the difference and change patterns of parameters in each dimension of the multi-dimensional state vector sequence, dividing the continuous state sequence into discrete state segments according to the thresholds, and labeling the corresponding state label set; transforming the state label set into a state path chain, performing frequency statistics and co-occurrence relationship mining on the node sequence, and constructing a state criterion splitting condition set containing behavior evolution relationships; and generating a path flow tree structure by sequentially splitting from the root node according to the state label sequence and the state criterion splitting condition set.

[0010] As a preferred embodiment of the e-commerce product recommendation method based on artificial intelligence algorithms described in this invention, the slicing and reorganization includes: extracting the main feature dimensions based on the user profile set and the product profile set, constructing a standardized feature matrix, wherein the standardized feature matrix includes a user feature matrix and a product feature matrix; calculating the local rate of change vector on each feature dimension and performing stability verification after slicing at fluctuation points, and dynamically slicing the user feature matrix and the product feature matrix respectively to generate a set of feature submatrices.

[0011] As a preferred embodiment of the e-commerce product recommendation method based on artificial intelligence algorithms described in this invention, the generation of the candidate cross-mapping set includes: setting an offset vector sequence based on the asymmetric distribution characteristics of product profiles in the dimensions of price and popularity, with each offset controlling the cross-combination path, to generate a misaligned mapping rule group; and performing a combination operation on the sliced ​​user and product sub-matrices according to the misaligned mapping rule group to form a candidate cross-mapping set.

[0012] As a preferred embodiment of the e-commerce product recommendation method based on artificial intelligence algorithms described in this invention, the following steps are taken: when projecting cross-product nodes onto the deinterlaced graph structure, the original product ID mapping table associated with them is recorded simultaneously to generate an associated product-path probability mapping table; based on the node path set, the jump frequency and time interval between user behavior nodes are statistically analyzed to construct a weighted directed graph structure; a cross-product node set is defined according to the candidate cross-mapping set, and the user behavior nodes involved therein are projected onto the weighted directed graph structure to construct an index matrix composed of behavior nodes and cross-product nodes.

[0013] As a preferred embodiment of the e-commerce product recommendation method based on artificial intelligence algorithms described in this invention, the generation of the candidate recommendation set includes: for each mapping unit in the index matrix, calculating the coupling weight vector between behavioral paths based on the set of shortest paths between the connecting behavioral nodes and other nodes in the graph; and jointly setting a coupling degree threshold and a transition weight threshold based on the coupling weight vector and the edge weights in the original behavioral transition graph, and filtering out cross-product node pairs that meet the conditions to form a candidate recommendation set.

[0014] As a preferred embodiment of the e-commerce product recommendation method based on artificial intelligence algorithms described in this invention, the path probability density mapping includes: based on the transition frequency of each node in the user-behavior path flow tree in the user behavior sequence, counting the number of historical behaviors jumping from the previous node to the current node, and smoothing all historical jump behaviors according to their distribution on the timeline; defining dynamic path entry detection rules based on the response time distribution of product nodes in user behavior; and mapping the above two parts to node numbers and path order to form a three-dimensional distribution structure with path node-frequency density-response time as a joint index, constituting a density mapping for behavior path sorting.

[0015] Secondly, the present invention provides an e-commerce product recommendation system based on artificial intelligence algorithms, which includes:

[0016] The multi-state feature extraction module extracts multi-state feature parameters based on users' historical behavior data, and uses a multi-criteria splitting mechanism to construct a user-behavior path flow tree, generating a set of node paths with behavioral evolution patterns.

[0017] The profile slicing and mapping module slices and reassembles the user profile set and the product profile set under multi-dimensional features, and generates a candidate cross-mapping set through predefined misalignment mapping rules;

[0018] The deinterlaced graph module constructs a graph of the user's behavior cluster, projects the cross-product nodes in the node path set and candidate cross-mapping set into the deinterlaced graph structure between behavior clusters, and extracts interaction points with high transition weights and behavior coupling as candidate recommendation sets.

[0019] The path probability regression module uses the path probability density mapping corresponding to each cross-product node in the candidate recommendation set to reverse look up the associated product-path probability mapping table, performs weight regression calculation on each node of the user-behavior path flow tree path, and generates a product recommendation sequence based on the maximum path pass probability.

[0020] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the e-commerce product recommendation method based on artificial intelligence algorithms as described in the first aspect of the present invention.

[0021] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the e-commerce product recommendation method based on artificial intelligence algorithms as described in the first aspect of the present invention.

[0022] The beneficial effects of this invention are as follows: By introducing multi-state behavior modeling and a path flow tree mechanism, this invention can comprehensively capture the behavioral evolution characteristics of users on e-commerce platforms and effectively identify the dynamic changing trends of user interests; through profile slicing and misalignment mapping strategies, it fully explores the potential matching relationships between users and products, improving the accuracy and diversity of recommendations. Combining behavioral graph structures and path probability density mapping, it can achieve quantitative evaluation and weighted regression of recommendation paths, thereby significantly improving the relevance and click-through rate of recommendation sequences.

[0023] Compared to traditional recommendation algorithms that rely solely on static features or collaborative filtering, this invention achieves deep integration at the structural modeling and temporal response levels, resulting in stronger user preference recognition and real-time recommendation adaptability.

[0024] In summary, this invention achieves an intelligent upgrade of personalized recommendations by constructing a recommendation process with behavioral chain criteria and coupled graph mapping, which not only improves user experience but also enhances the platform's conversion efficiency and content distribution capabilities. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating an e-commerce product recommendation method based on artificial intelligence algorithms;

[0027] Figure 2 This is a structural diagram of an e-commerce product recommendation system based on artificial intelligence algorithms. Detailed Implementation

[0028] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0029] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0030] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0031] As mentioned in the background section, early recommendation methods primarily relied on strategies such as collaborative filtering, content matching, or rating matrix modeling to identify potentially interesting products through similarity metrics. However, these methods have significant limitations when facing data sparsity, the cold start problem, and the complexity of user behavior. In recent years, the development of artificial intelligence, especially deep learning and graph neural networks, has provided new ideas for mining user behavior trajectories and modeling behavioral evolution paths. E-commerce recommendation technology has gradually shifted from "static preference modeling" to "dynamic behavior modeling." Under this trend, integrating multi-dimensional behavioral features, strengthening the interactive modeling of user profiles and product profiles, and building a more flexible and interpretable recommendation framework have become important directions for the development of current intelligent recommendation systems.

[0032] Figure 1 This is a flowchart of an e-commerce product recommendation method based on an artificial intelligence algorithm according to an embodiment of the present invention. Figure 1 As shown, the e-commerce product recommendation method based on artificial intelligence algorithms includes:

[0033] S1: Extract multi-state feature parameters based on user historical behavior data, construct a user-behavior path flow tree using a multi-criteria splitting mechanism, and generate a set of node paths with behavioral evolution patterns.

[0034] In this embodiment of the invention, the generation of the node path set includes the following steps:

[0035] a. Based on the user behavior log data pre-recorded in the e-commerce platform, extract the page dwell time parameter, browsing frequency parameter and jump order code for each behavior event, and construct a multi-dimensional state vector sequence according to the time series to represent the behavior evolution trajectory.

[0036] Specifically, each user interaction while browsing products is transformed into a composite vector containing multiple feature states. Each dimension represents browsing frequency (the number of times a user accesses a product or page per unit of time), page dwell time (the average time a user spends on a product page), and jump sequence encoding (the sequential number of the path the user traverses before and after entering the current page). Through this process, the original discrete behavior logs can be converted into a multidimensional state vector sequence. Each vector sequence is arranged continuously on the time axis, thus reflecting the dynamic behavioral trajectory of the user throughout the e-commerce platform.

[0037] b. Based on the difference and change pattern of parameters in each dimension of the multidimensional state vector sequence, set a set of transition thresholds, divide the continuous state sequence into discrete state segments according to the thresholds, and label the corresponding state label set.

[0038] In practice, based on the changing trends of each parameter in the multidimensional state vector, differential features are calculated and compared with a preset set of transition thresholds to determine whether the behavioral state has changed significantly.

[0039] The set of transfer thresholds is obtained based on statistics of global user behavior samples on e-commerce platforms, and includes critical criteria for situations such as sudden changes in frequency, significant fluctuations in dwell time, and changes in jump paths.

[0040] Once the change in one or more dimensions of a continuous state vector exceeds the corresponding threshold, a behavioral state transition is considered to have occurred. Based on this, the continuous vector sequence is divided into multiple discrete state segments. Each state label is appended with an associated splitting condition identifier, recording the threshold type that triggered the state transition, providing a basis for condition matching for subsequent process tree splitting.

[0041] c. Transform the set of state labels into a chain of state paths, perform frequency statistics and co-occurrence relationship mining on the node sequences, and construct a set of state criterion splitting conditions that includes behavioral evolution relationships.

[0042] For example, firstly, state labels are connected sequentially to form a path chain based on the time sequence of user behavior, and the frequency of occurrence of each label combination is counted to establish a frequency matrix to characterize the evolution probability between different states.

[0043] Meanwhile, to further improve the ability of the path structure to express behavior patterns, state pairs and path sub-segments that co-occur frequently among user groups are mined, which includes: for paths with length ≥ I steps (where I is set according to actual conditions), even if the occurrence frequency is lower than the average value, if the end node of the path is a purchase behavior, mark it as a key conversion path; maintain an independent long path storage pool, when the conventional condition set cannot cover all conditions, extract the top 10% of paths from the pool according to the conversion rate ranking and add them to the split condition set; the state split condition set for process tree construction is extracted accordingly.

[0044] Two-layer condition screening is adopted, wherein, the main condition set is: the transition frequency of a transition pair is greater than the mean transition frequency plus 3 times the standard deviation; the auxiliary condition set is: after clustering by DBSCAN, cluster centers with a silhouette coefficient greater than H are selected to be used as the structural basis for node splitting in the process tree; wherein, H is a set silhouette coefficient threshold, which can be set according to actual conditions.

[0045] The above operation not only avoids the problem of path fragmentation caused by random behaviors, but also enhances the sensitivity and expression ability of behavior modeling to group laws.

[0046] According to the state label sequence and the state criterion split condition set, a path process tree structure is generated by sequential splitting from the root node, and each terminal node path is classified into a user behavior path set, where each path represents a type of behavior evolution pathway.

[0047] In the process tree construction step, the present invention adopts an iterative splitting strategy based on the state label sequence and the state split condition set, and generates a complete user behavior path process tree structure from top to bottom starting from the root node.

[0048] Specifically, taking all state path chains as the initial input, perform the first state splitting operation on the root node, and judge whether the sub-path division condition is satisfied according to the aforementioned state splitting condition; if the condition is satisfied, generate two or more child nodes, each child node represents a state evolution direction; thereafter, repeat the above splitting process for each child node respectively, the termination condition satisfies any of the following: the current node depth ≥ I, the number of paths included in the node < max(0.1%, co-occurrence frequency threshold of similar users / 10), no gain is produced after 3 consecutive splits, and then the construction is completed.

[0049] In the final process tree, each terminal node represents an independent user behavior evolution path, and all such paths are uniformly classified into the user behavior path set.

[0050] It is worth emphasizing that the path flow tree construction method of this invention, compared with the traditional user clustering-behavior inductive recommendation modeling approach, can more accurately depict the behavioral evolution trajectory of individual users on e-commerce platforms, especially demonstrating excellent expressive ability for non-linear behavioral paths such as shallow visits-deep dwell times and skipping browsing-target clicks. By introducing multi-dimensional state features and splitting condition sets, not only is the discriminative power of path classification improved, but the adaptability of behavioral modeling to temporal heterogeneity and multi-modal features is also enhanced, thereby constructing a more targeted and dynamically evolving user behavior profile.

[0051] S2: Slice and recombine the user profile set and product profile set under multi-dimensional features respectively, and perform cross-combination through predefined misalignment mapping rules to generate a candidate cross-mapping set.

[0052] S2.1: Based on the user profile set U and the product profile set G, extract the main feature dimensions and construct a standardized feature matrix, which includes the user feature matrix and the product feature matrix.

[0053] In the technical solution of this invention, the selection of main feature dimensions includes, but is not limited to, interest category, consumption level, activity cycle (for user profile), price range, product popularity, and category tags (for product profile).

[0054] In the actual construction process, the Z-score normalization method is used to normalize the original features, unify the dimensions, and eliminate feature offset problems caused by inconsistent dimensions. After construction, two normalized feature matrices are formed: the user feature matrix and the... and product feature matrix Where m and n represent the number of users and products, respectively, and d represents the uniform number of feature dimensions. This standardization method ensures the comparability and mathematical consistency of subsequent slicing and misalignment mapping operations.

[0055] S2.2: Calculate the local rate of change vector for each feature dimension and perform stability verification after slicing at fluctuation points. Dynamically slice the user feature matrix and the product feature matrix respectively to generate a set of feature submatrices.

[0056] The local rate of change vector is obtained by performing a sliding window difference on the eigenvalues ​​in the corresponding dimension, and the local fluctuation point is marked by the peak value of the absolute value of the first derivative.

[0057] In the specific operation of this invention, the stability check performed after slicing the fluctuation points includes: if the Jaccard similarity between adjacent slices is greater than the similarity threshold, then the slices are merged; ensuring that the misalignment mapping offset is less than or equal to the length of the merged slice. After this processing, a set of user sub-feature matrices and a set of product sub-feature matrices can be obtained. The purpose is to improve the ability to characterize the implicit local associations between users and products and avoid the detail occlusion problem caused by overall vector operations.

[0058] S2.3: Based on the asymmetric distribution characteristics of product profiles in the dimensions of price and popularity, set an offset vector sequence, with each offset controlling the cross-combination path to generate a misalignment mapping rule group.

[0059] It should be noted that each offset δ i This represents the step size of the product submatrix along the feature dimension axis when the user submatrix and product submatrix are combined. For example, when δ i When = +1, it means that the user submatrix U is... j With the commodity submatrix G j+1 Combine them, rather than combining them according to their corresponding positions.

[0060] The theoretical basis of this offset combination design is that there is a temporal misalignment or trend deviation between user interest hotspots and product popularity highs in the feature space. By constructing offset relationships, potential cross-interest matching units can be effectively identified, thereby expanding the potential interaction space between users and products.

[0061] Subsequently, based on the aforementioned offset vector sequence, a misalignment mapping rule set R is generated, where each rule r i By quintuple (U j , δ i It consists of mapping type and path fingerprint, where path fingerprint = Hash(user submatrix features + product submatrix features).

[0062] The combination method and offset relationship between the user submatrix and the product submatrix are clearly defined.

[0063] Guided by this set of rules, all combinable paths are traversed throughout the entire quantum space, enabling multi-path, multi-step, and multi-dimensional cross-combination operations, thereby constructing a set of cross-matching units with distinctive features and structural jump characteristics.

[0064] S2.4: Based on the misalignment mapping rule, perform a combination operation on the sliced ​​user and product submatrices to form a candidate cross-mapping set.

[0065] Guided by the aforementioned misalignment mapping rule set, the user submatrix and the product submatrix are combined pairwise to form a candidate cross-mapping set C, where each candidate mapping C... i For one The cross-combination unit. During the combination process, feature concatenation and linked encoding are used to fuse the two matrices into a unified representation structure, while recording their offset labels and path metadata.

[0066] S3: Construct a behavior cluster graph of the user, project the cross-product nodes in the node path set and candidate cross-mapping set into the anti-interlaced graph structure between behavior clusters, and extract the interaction points with high transfer weight and behavior coupling as the candidate recommendation set.

[0067] When projecting cross-product nodes onto the deinterlaced graph structure, the original product ID mapping table associated with them is recorded simultaneously, and an associated product-path probability mapping table is generated to ensure a one-to-many association between cross-nodes and product entities.

[0068] It should be noted that cross-product nodes are composite nodes generated by the user feature submatrix and the product feature submatrix through a misalignment mapping rule; product nodes are original entity nodes containing only product IDs and inherent attributes.

[0069] First, based on the set of node paths, the frequency and time interval of jumps between user behavior nodes are statistically analyzed to construct a weighted directed graph structure, where the edge weights are defined by the transition frequency vector and the time delay factor.

[0070] Specifically, a weighted directed graph structure is constructed based on the state sequence and path frequency data extracted from the node path set to represent the jump relationship between user behavior nodes.

[0071] In the graph construction process, each state label with significant behavioral characteristics is used as a node element in the graph. A directed edge is constructed for every two states with a continuous transition relationship to reflect the migration direction of the behavioral path. When defining the edge weight, the transition frequency vector and the time delay factor are used as dual weight components.

[0072] The transition frequency vector quantifies the relative frequency of the behavioral transition among sample users, reflecting the dominance of the transition. The time delay factor measures the tightness and response rate of the transition by calculating the average dwell time between adjacent state tags. The weighted calculation of these two factors allows the graph to not only express the structural relationships of behaviors but also reflect the execution rhythm and response pattern of user behaviors, helping to identify the main transition flows in the behavioral path. Traditional methods, which only use the transition frequency vector or the time delay factor for path construction, cannot accurately characterize the actual difference in importance between high-frequency but slow behaviors and low-frequency but fast behaviors. This invention, through fusion modeling, forms a weighted structure that is more sensitive to actual path selection.

[0073] Secondly, a set of cross-product nodes is defined based on the candidate cross-mapping set, and the user behavior nodes involved are projected into a weighted directed graph structure to construct an index matrix composed of behavior nodes and cross-product nodes. This matrix is ​​used to represent the anti-interlaced positional relationship of product mapping points in the behavior graph.

[0074] During the mapping process, firstly, based on the unique identifier of each cross-product or behavior node in the candidate cross-product node set, it is searched to see if the behavior node already exists in the currently constructed graph structure. If it exists, the corresponding relationship is directly established. If it does not exist, it is inserted into the graph structure according to the new behavior node insertion judgment rule, and an edge connection is established between it and the high-frequency jump node. The initial frequency and time factor of the connection edge are calculated to improve the boundary information of the graph structure.

[0075] The rules for inserting new behavior nodes include: there is at least one indirect association (within two hops) in the current user's historical behavior; the co-occurrence frequency of similar users is greater than 3 times the average level of the group; if the end of the path is a purchase behavior, even if the co-occurrence frequency is low, the weight is increased by half and the S1 termination condition is exempted (the path terminated by S1 is temporarily stored, and if S3 triggers the insertion rule and there is a matching path in the buffer, it is reused first); the initial value of the new edge weight is min(the average edge weight of the current graph, 0.1).

[0076] Subsequently, an index matrix is ​​constructed between behavior nodes and cross-product nodes. This matrix expresses the anti-interlacing mapping relationship between nodes in the behavior graph in a two-dimensional form, that is, the behavior anchor point corresponding to each cross-product node in different path sequences and its frequency of occurrence and structural position.

[0077] Finally, for each mapping unit in the index matrix, based on the set of shortest paths between the connecting behavior node and other behavior nodes in the graph, the coupling weight vector between behavior paths is calculated. This weight is obtained by weighting the path length, jump frequency, and common product density. Based on the coupling weight vector and the edge weights in the original behavior transition graph, the coupling degree threshold and the transition weight threshold are jointly set to screen the cross-product node pairs that meet the conditions, forming a candidate recommendation set, and recording their path sequence and coupling background parameters.

[0078] In this embodiment of the invention, coupling analysis is based on the set of all shortest paths formed in the graph between the cross-product node and the original behavior node, and the coupling weight vector between the behavior paths is calculated comprehensively.

[0079] This weighted vector incorporates three core factors to form a weighting mechanism: first, path length, where shorter paths typically imply a higher probability of direct behavioral jumps, thus assigning a higher coupling tendency; second, jump frequency, which is the number of times the shortest path has actually been triggered in historical user behavior, reflecting the path's reliability; and third, common product density, representing the degree of overlap between products or pages accessed by behavioral nodes along the path within the user group, with higher density indicating stronger behavioral coupling. In constructing this weighted coupling model, weight parameters are set for different factors, ensuring that the sum of these weight parameters is 1, allowing for flexible adjustment of the contribution of different factors to the final coupling weight in the specific implementation.

[0080] After the coupling weight vector is calculated, the transition weight information of the edges in the original graph is combined with the dual threshold judgment mechanism to screen the cross-product node pairs and construct the recommended candidate set.

[0081] This mechanism sets a coupling degree threshold and a transfer weight threshold, and periodically calculates the correlation coefficient between the coupling degree and the transfer weight. That is, if there is a strong positive correlation, the coupling degree threshold is lowered, such as to 20%, and the transfer weight threshold is increased, such as to 15%. The values ​​here are only for reference and should be set according to actual operation. If there is a negative correlation, the transfer weight threshold is frozen and only the coupling degree threshold is used for filtering.

[0082] This joint screening structure can effectively avoid the bias problem of a single indicator dominating the recommendation results, and improve the rationality of the behavioral correlation of recommendation nodes and the structural stability.

[0083] Through the above steps, this invention achieves in-depth mining of potential recommendation nodes in user behavior paths. Particularly in graph construction and cross-product node linkage, by combining time weighting and structural co-occurrence considerations, the recommendation results take into account both the true semantics of behavior and the stability of group behavior.

[0084] S4: Utilize the path probability density mapping corresponding to each cross-product node in the candidate recommendation set to reverse-look up the associated product-path probability mapping table, perform weight regression calculation on each node of the user-behavior path flow tree path, and generate a product recommendation sequence based on the maximum path pass probability.

[0085] First, the reason for employing path probability density mapping instead of a simple node scoring mechanism is that user behavior in complex process paths exhibits significant temporal, dynamic, and context-dependent characteristics. A single scoring model cannot effectively reflect the transition patterns and temporal response features between nodes. Therefore, this invention proposes a probability density mapping based on a behavior path process tree, aiming to quantify the fluctuations in user behavior probability along a specific path and their impact on recommendation performance.

[0086] S4.1: For each cross-product node in the candidate recommendation set, based on the historical traversal frequency of the behavior node in the associated path set within the process tree path, parse its associated original product set and construct a path probability density mapping. This path probability density mapping is obtained by jointly fitting the node transition frequency kernel density estimation and the product behavior response time distribution. The specific steps are as follows:

[0087] 1. Based on the transition frequency of each node in the user behavior path flow tree in the user behavior sequence, the number of historical behaviors from the previous node to the current node is counted, and all historical jump behaviors are smoothed according to their distribution on the timeline.

[0088] It should be noted that this invention introduces time axis density estimation, which enables the expression of the activity differences of jumping behavior in different time periods, thereby improving the sensitivity to changes in behavior frequency.

[0089] To achieve a continuous expression of the above node transition relationships, this invention uses a kernel density estimation method to smooth the node transition frequency.

[0090] 2. Define dynamic path entry detection rules based on the response time distribution of product nodes in user behavior.

[0091] Specifically, when the number of entry paths to a node (i.e., the number of arrows pointing to it from other nodes) exceeds twice the number of exit paths (the number of arrows pointing to other nodes), it is automatically marked as a candidate entry node; the response time is calculated starting from the most recently visited candidate entry node; if there are multiple candidate entry nodes in the path, the entry node closest to the current node is selected.

[0092] 3. The above two parts are mapped to node numbers and path order to form a three-dimensional distribution structure with "path node-frequency density-response time" as the joint index. Specifically, path node-frequency density-response time, where path node is the numbered node in the process tree, frequency density reflects its transfer activity level, and response time dimension reflects the time characteristics of behavior reach, forming a density mapping that can be used for behavior path sorting.

[0093] S4.2: Taking each node on each path as input, and based on its matching degree in the probability density mapping of different paths, calculate the node affiliation weight vector through weighted regression. This vector reflects the contribution of each node to the corresponding product path in the candidate recommendation set.

[0094] Compared to traditional simple weighted methods based on click-through rate, dwell time, or TF-IDF calculation, this invention captures the coupling relationship between structural context and behavioral dynamics by matching the density of nodes within the path, thus possessing a higher ability to interpret behavior.

[0095] Specifically, this invention sequentially inputs the joint feature value of each node in the path probability density mapping for each node in each path, namely the kernel density frequency value and response time feature vector of the current node in a certain product path, and calculates the matching score in the product path.

[0096] The matching score calculation includes: first, determining the probability density value of a node appearing in the target product path; second, calculating the actual time interval between the node and its predecessor, and simultaneously, finding the average dwell time and standard deviation of the behavior jump on the target product path from the graph. If the difference between the actual time interval and the average dwell time is less than the standard deviation, it is considered that the response time is consistent and is assigned a value of 1; if the difference is between the standard deviation and twice the standard deviation, it is assigned a value of 0.5; if it exceeds twice the standard deviation, it is assigned a value of 0; and the above two scores are weighted and summed to obtain the score.

[0097] The score reflects the actual influence of the node in the current path on triggering the product. To ensure the stability and comparability of the calculation results, this invention normalizes the matching scores of each path through standardization.

[0098] Furthermore, to achieve node-level path attribution determination, this invention uses the matching score of each node under multiple paths as the independent variable and the historical product trigger tag as the dependent variable. The attribution weight vector of each node is obtained by least squares regression. This vector represents the degree of attribution of the node under different product paths, that is, whether the node belongs to the core behavior facilitating node of a certain recommendation path.

[0099] As can be seen, unlike traditional label classification methods, this method can reflect the cross-contribution of a node in multiple recommendation paths and has a stronger ability to represent path structure.

[0100] S4.3: Based on the regressed path node weight vectors and the node sequence relationship of the path structure in the process tree, calculate the probability of passage for each path, sort the path set according to the probability of passage, and extract the top N paths as the candidate set of recommended paths, where N is a constant.

[0101] Specifically, the calculation process for the probability of passage is as follows:

[0102] Record the historical number of transitions between every two adjacent nodes in the path sequence, and calculate the transition probability of each step (number of times the next node appears / total number of transitions of the current node); for necessary nodes (such as login nodes), force a minimum transition probability (such as not less than 20%) to avoid blocking the critical path due to accidental low weight; multiply the transition probabilities of all adjacent node pairs in the path in turn to obtain the passability value of the complete path.

[0103] After calculating the probability of passage, this invention sorts all candidate path sets according to the probability of passage and selects the top N paths with the highest probability of passage as the recommended path candidate set.

[0104] This operation enables an effective transition from node-level weight modeling to path-level accessibility determination, and is one of the key steps in this invention to achieve multi-path normalization selection.

[0105] S4.4: Aggregate the cross-product nodes associated with each path in the candidate recommended path set according to their corresponding path passability probability, parse them into the original product set through the product-path probability mapping table, sort them according to the aggregation weight, form the final product recommendation sequence, and simultaneously retain the associated path index and regression weight parameters for subsequent iterative optimization.

[0106] Specifically, the cosine similarity between paths is calculated, and paths with a similarity of <0.3 are considered conflicting paths (0.3 is a set threshold, which needs to be set according to the actual situation during operation); if they are non-conflicting paths, the maximum probability value is taken; if they are conflicting paths, the Borda counting method is used to sort them, and finally the overall recommendation probability index of the product is formed.

[0107] Unlike existing technologies that rank products solely based on click count or average rating, this invention's recommendation mechanism centers on the path, integrating structural, behavioral, and temporal features to generate a more logically evolving recommendation sequence. After ranking, the final product recommendation sequence is output to the user, while the path index and node regression weight vectors are retained as foundational data for the next round of recommendation training and model iteration optimization.

[0108] As can be seen, this invention conducts a series of regression analyses and sorting and filtering operations around the probability density characteristics of path nodes, aiming to realize a dynamic product recommendation and sorting process driven by historical behavior trajectories. This not only improves the personalization accuracy of the recommendation results, but also enhances the adaptability to the process path topology.

[0109] Furthermore, such as Figure 2 As shown, this embodiment also provides an e-commerce product recommendation system based on artificial intelligence algorithms, including:

[0110] The multi-state feature extraction module 100 extracts multi-state feature parameters based on user historical behavior data, and uses a multi-criteria splitting mechanism to construct a user-behavior path flow tree, generating a set of node paths with behavior evolution rules.

[0111] The profile slicing and mapping module 200 slices and recombines the user profile set and the product profile set under multi-dimensional features, and generates a candidate cross-mapping set through predefined misalignment mapping rules;

[0112] The deinterlaced graph module 300 constructs a graph of the user's behavior cluster, projects the cross-product nodes in the node path set and candidate cross-mapping set to the deinterlaced graph structure between behavior clusters, and extracts interaction points with high transition weights and behavior coupling as candidate recommendation sets.

[0113] The path probability regression module 400 uses the path probability density mapping corresponding to each cross-product node in the candidate recommendation set to perform weight regression calculation on each node of the user-behavior path flow tree path, and generates a product recommendation sequence based on the maximum path pass probability.

[0114] This embodiment also provides a computer device applicable to e-commerce product recommendation methods based on artificial intelligence algorithms, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the e-commerce product recommendation method based on artificial intelligence algorithms as proposed in the above embodiment.

[0115] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0116] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the e-commerce product recommendation method based on artificial intelligence algorithms as proposed in the above embodiments.

[0117] In summary, this invention, by introducing multi-state behavioral modeling and a path flow tree mechanism, can comprehensively capture the behavioral evolution characteristics of users on e-commerce platforms and effectively identify the dynamic trends of user interests. Through profile slicing and misalignment mapping strategies, it fully explores the potential matching relationships between users and products, improving the accuracy and diversity of recommendations. Combining behavioral graph structure and path probability density mapping enables quantitative evaluation and weighted regression of recommendation paths, thereby significantly improving the relevance and click-through rate of recommendation sequences. Compared to traditional recommendation algorithms that rely solely on static features or collaborative filtering, this invention achieves deep integration at the structural modeling and temporal response levels, possessing stronger user preference recognition capabilities and real-time recommendation adaptability. In conclusion, by constructing a recommendation process with behavioral chain criteria and coupled graph mapping, this invention achieves an intelligent upgrade of personalized recommendations, not only improving user experience but also enhancing the platform's conversion efficiency and content distribution capabilities.

[0118] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An e-commerce product recommendation method based on an artificial intelligence algorithm, characterized by: include: Based on the extraction of multi-state feature parameters from user historical behavior data, a multi-criteria splitting mechanism is used to construct a user-behavior path flow tree, generating a set of node paths with behavioral evolution patterns; The user profile set and the product profile set are sliced ​​and recombined under multi-dimensional features, and then cross-combined through predefined misalignment mapping rules to generate a candidate cross-mapping set; Construct a behavior cluster graph of the user, project the cross-product nodes in the node path set and candidate cross-mapping set into the anti-interlaced graph structure between behavior clusters, and extract the interaction points with high transition weight and behavior coupling as the candidate recommendation set; By utilizing the path probability density mapping corresponding to each cross-product node in the candidate recommendation set, the associated product-path probability mapping table is reversed, and weight regression calculation is performed on each node of the user-behavior path flow tree path. Based on the maximum path pass probability, a product recommendation sequence is generated. The construction of the user-behavior path flow tree includes: extracting page dwell time parameters, browsing frequency parameters, and jump order codes for each behavior event based on user behavior log data recorded in the e-commerce platform, and constructing a multi-dimensional state vector sequence according to the time series; setting a set of transition thresholds based on the difference and change patterns of parameters in each dimension of the multi-dimensional state vector sequence, dividing the continuous state sequence into discrete state segments according to the thresholds, and labeling the corresponding state label set; transforming the state label set into a state path chain, performing frequency statistics and co-occurrence relationship mining on the node sequence, and constructing a state criterion splitting condition set containing behavior evolution relationships; and generating a path flow tree structure by sequentially splitting from the root node according to the state label sequence and the state criterion splitting condition set. When projecting cross-product nodes onto the de-interlaced graph structure, the original product ID mapping table associated with them is recorded simultaneously, generating an associated product-path probability mapping table. Based on the node path set, the jump frequency and time interval between user behavior nodes are statistically analyzed to construct a weighted directed graph structure. A cross-product node set is defined according to the candidate cross-mapping set, and the user behavior nodes involved are projected onto the weighted directed graph structure to construct an index matrix composed of behavior nodes and cross-product nodes. The index matrix is ​​used to represent the de-interlaced positional relationship of product mapping points in the behavior graph.

2. The e-commerce product recommendation method based on artificial intelligence algorithm according to claim 1, characterized in that: The slice recombination includes: Based on user profile sets and product profile sets, the main feature dimensions are extracted, and a standardized feature matrix is ​​constructed. The standardized feature matrix includes a user feature matrix and a product feature matrix. Calculate the local rate of change vector for each feature dimension and perform stability verification after slicing at fluctuation points. Dynamically slice the user feature matrix and the product feature matrix respectively to generate a set of feature submatrices.

3. The method of claim 2, wherein: The generation of the candidate cross-mapping set includes: Based on the asymmetric distribution characteristics of product profiles in the dimensions of price and popularity, an offset vector sequence is set, and each offset controls the cross-combination path to generate a misalignment mapping rule group. Based on the misalignment mapping rule, the user and product submatrices after being sliced ​​are combined to form a candidate cross-mapping set.

4. The method of claim 3, wherein: The generation of the candidate recommendation set includes: For each mapping unit in the index matrix, based on the set of shortest paths between the connecting behavior node and other nodes in the graph, the coupling weight vector between behavior paths is calculated; based on the coupling weight vector and the edge weights in the original behavior transition graph, the coupling degree threshold and the transition weight threshold are jointly set, and the cross-product node pairs that meet the conditions are selected to form a candidate recommendation set.

5. The e-commerce product recommendation method based on artificial intelligence algorithms as described in claim 1, characterized in that: The mapping of the path probability density includes: Based on the transition frequency of each node in the user behavior path flow tree in the user behavior sequence, the number of historical behaviors from the previous node to the current node is counted, and all historical jump behaviors are smoothed according to their distribution on the timeline. Based on the response time distribution of cross-product nodes in user behavior, dynamic path entry detection rules are defined. By mapping the above two parts to node numbers and path order, a three-dimensional distribution structure with path node-frequency density-response time as a joint index is formed, which constitutes a density mapping for behavior path sorting.

6. An e-commerce product recommendation system based on artificial intelligence algorithms, based on the e-commerce product recommendation method based on artificial intelligence algorithms according to any one of claims 1 to 5, characterized in that: Also includes: The multi-state feature extraction module extracts multi-state feature parameters based on users' historical behavior data, and uses a multi-criteria splitting mechanism to construct a user-behavior path flow tree, generating a set of node paths with behavioral evolution patterns. The profile slicing and mapping module slices and reassembles the user profile set and the product profile set under multi-dimensional features, and generates a candidate cross-mapping set through predefined misalignment mapping rules; The deinterlaced graph module constructs a graph of the user's behavior cluster, projects the cross-product nodes in the node path set and candidate cross-mapping set into the deinterlaced graph structure between behavior clusters, and extracts interaction points with high transition weights and behavior coupling as candidate recommendation sets. The path probability regression module uses the path probability density mapping corresponding to each cross-product node in the candidate recommendation set to reverse look up the associated product-path probability mapping table, performs weight regression calculation on each node of the user-behavior path flow tree path, and generates a product recommendation sequence based on the maximum path pass probability.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the e-commerce product recommendation method based on artificial intelligence algorithms as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the e-commerce product recommendation method based on artificial intelligence algorithms as described in any one of claims 1 to 5.

Images