Knowledge graph-based vertical part e-commerce recommendation method and system

Abstract

The invention relates to the technical field of knowledge maps, in particular to a vertical part e-commerce recommendation method and system based on a knowledge map. The method comprises the following steps: acquiring test data, and preprocessing the test data to obtain preprocessed test data; constructing a part vertical knowledge graph based on the preprocessed test data; constructing a graph mining model based on the part vertical knowledge graph; determining a component knowledge graph element path of the graph mining model, embedding target type node features, and carrying out importance sampling design on nodes in the component knowledge graph; experimental comparison and analysis are carried out through a pre-collected test data set, the effectiveness of the graph mining model is verified, and the accuracy of the model can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph technology, and in particular to a vertical spare parts e-commerce recommendation method and system based on knowledge graph. Background Art

[0002] With the popularization of mobile terminals, the upgrade of the supply side and the improvement of chip computing power, the number of commodity categories and the amount of user behavior data on the consumer side have exploded exponentially. Accordingly, the recommendation system has become an indispensable part of people's lives, helping users to screen information. For example, Microsoft Research manually constructs advertising-related features and combines them with logistic regression models to explore and predict the user click-through rate of new ads, so as to determine the order of advertising and the position displayed in the search engine.

[0003] With the improvement of domestic logistics facilities, many manufacturing suppliers have gradually established their own vertical e-commerce platforms for online sales. In the past, most recommendation algorithms performed well in scenarios with large data sets, but their performance was unsatisfactory in scenarios with cold start of recommendation systems and scenarios with small data scales. The emergence of knowledge graphs provides an opportunity to solve this problem. Knowledge graphs are a knowledge base that integrates data through mathematical models with graph structures.

[0004] Currently, the recommendation system based on knowledge graph has the problem of insufficient accuracy. In order to solve this technical problem, a vertical spare parts e-commerce recommendation method and system based on knowledge graph is proposed. Summary of the invention

[0005] In order to solve the technical problems existing in the above-mentioned prior art, the present invention provides a vertical spare parts e-commerce recommendation method and system based on knowledge graph.

[0006] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, in an embodiment provided by the present invention, a vertical spare parts e-commerce recommendation method based on a knowledge graph is provided, and the method comprises the following steps:

[0008] S10, acquiring test data, and preprocessing the test data to obtain preprocessed test data;

[0009] S20, constructing a vertical knowledge graph of spare parts based on the preprocessed test data;

[0010] S30. Construct a graph mining model based on the vertical knowledge graph of spare parts;

[0011] S40, determining the spare parts knowledge graph meta-path of the graph mining model, embedding target type node features, and performing importance sampling design on the nodes in the spare parts knowledge graph;

[0012] S50. The effectiveness of the graph mining model was verified through experimental comparison and analysis of the pre-collected test data set.

[0013] As a further solution of the present invention, the test data includes: user type information, spare parts type information, assembly part type information and e-commerce platform information.

[0014] As a further solution of the present invention, the preprocessing includes filling and filtering missing values ​​of the test data and performing data standardization processing on the test data.

[0015] As a further solution of the present invention, the step S20, constructing a vertical knowledge graph of spare parts based on the preprocessed test data, includes:

[0016] S201, based on the pre-processed test data, collecting spare parts vertical platform knowledge data;

[0017] S202. Construct a spare parts vertical knowledge graph based on the spare parts vertical platform knowledge data.

[0018] As a further solution of the present invention, S202, constructing a spare parts vertical knowledge graph based on spare parts vertical platform knowledge data, including:

[0019] The knowledge data of the spare parts vertical platform is represented by various classified table structures and stored in the MySQL database. The data stored in the MySQL database is then converted through D2RQ and stored in the Neo4j graph database to obtain the spare parts vertical knowledge graph.

[0020] As a further solution of the present invention, S30, constructing a graph mining model based on the vertical knowledge graph of spare parts, including:

[0021] Build the RippleNet model as the base model;

[0022] Design meta-path rules based on preset rules;

[0023] Retrain the node representation vector of the target type node through the streaming feature embedding method;

[0024] The PageRank algorithm is used to calculate the importance value of each entity node in the spare parts vertical knowledge graph, and sampling is performed according to the node importance value during the knowledge ripple sampling process.

[0025] As a further solution of the present invention, the S40, determining the spare parts knowledge graph meta-path of the graph mining model, embedding the target type node features and designing the importance sampling of the nodes in the spare parts knowledge graph, includes:

[0026] S401, determining the meta-path of the spare parts knowledge graph according to the node distribution, relationship distribution, user preference habits and calculation constraints of the spare parts knowledge graph;

[0027] S402, embedding target type node features;

[0028] S403. Use PageRank algorithm to perform importance sampling design on the nodes in the spare parts knowledge graph.

[0029] As a further solution of the present invention, S402, embedding target type node features, includes:

[0030] S4021, using the principal component analysis method to perform data preprocessing on the characteristics of the target type nodes;

[0031] S4022. The experiment performs pooling operations on the user preference vector, node vector, and node features obtained by the model and then retrains the target type node representation vector through a multi-layer perceptron model.

[0032] As a further solution of the present invention, the step S4021, performing data preprocessing on the features of the target type nodes using a principal component analysis method, comprises the following steps:

[0033] S40211, constructing a node feature data matrix X;

[0034] S40212, performing Z-mean normalization processing on the node feature matrix X;

[0035] S40213. Calculate the covariance matrix R of the matrix Q;

[0036] S40214. Calculate the eigenvalues ​​and eigenvectors of the matrix R;

[0037] S40215. Sort the eigenvalues ​​in descending order;

[0038] S40216. retain the eigenvectors corresponding to the first n largest eigenvalues;

[0039] S40217, convert the original data into the new space;

[0040] S40218. Complete the conversion of original data.

[0041] On the second aspect, in another embodiment provided by the present invention, a vertical spare parts e-commerce recommendation system based on a knowledge graph is provided, the system comprising: a data acquisition preprocessing module, a spare parts vertical knowledge graph construction module, a graph mining model construction module, a model optimization module and a model verification module.

[0042] The data acquisition preprocessing module is used to acquire test data and preprocess the test data to obtain preprocessed test data.

[0043] The spare parts vertical knowledge graph construction module is used to construct the spare parts vertical knowledge graph based on the preprocessed test data.

[0044] The graph mining model building module is used to build a graph mining model based on the spare parts vertical knowledge graph.

[0045] The model optimization module is used to determine the spare parts knowledge graph meta-path of the graph mining model, embed the target type node features, and perform importance sampling design on the nodes in the spare parts knowledge graph.

[0046] The model verification module is used to conduct experimental comparison and analysis through pre-collected test data sets to verify the effectiveness of the graph mining model.

[0047] The technical solution provided by the present invention has the following beneficial effects:

[0048] The vertical spare parts e-commerce recommendation method based on knowledge graph provided by the present invention includes: obtaining test data and preprocessing it to obtain preprocessed test data; constructing a spare parts vertical knowledge graph based on the preprocessed test data; constructing a graph mining model based on the spare parts vertical knowledge graph; determining the spare parts knowledge graph meta-path of the graph mining model, embedding target type node features and performing importance sampling design on the nodes in the spare parts knowledge graph; and verifying the effectiveness of the graph mining model through experimental comparison and analysis of pre-collected test data sets. The present invention can effectively improve the accuracy of the model.

[0049] These and other aspects of the present invention will be more clearly understood in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other embodiments can be obtained based on these drawings without paying creative work.

[0051] Figure 1 This is a flowchart of a vertical spare parts e-commerce recommendation method based on knowledge graph according to an embodiment of the present invention.

[0052] Figure 2This is a structured information graph in a spare parts vertical platform in a vertical spare parts e-commerce recommendation method based on a knowledge graph according to an embodiment of the present invention.

[0053] Figure 3 This is a semi-structured information graph in a spare parts vertical platform in a vertical spare parts e-commerce recommendation method based on a knowledge graph according to an embodiment of the present invention.

[0054] Figure 4 This is the backbone graph of the Neo4j spare parts knowledge graph in the vertical spare parts e-commerce recommendation method based on the knowledge graph in one embodiment of the present invention.

[0055] Figure 5 A roadmap for optimizing the design of a vertical spare parts e-commerce recommendation method based on a knowledge graph according to an embodiment of the present invention.

[0056] Figure 6 This is a schematic diagram of the ripple network algorithm in a vertical spare parts e-commerce recommendation method based on knowledge graph according to an embodiment of the present invention.

[0057] Figure 7 This is a flowchart of target node feature embedding in a vertical spare parts e-commerce recommendation method based on knowledge graph according to an embodiment of the present invention.

[0058] Figure 8 This is a model comparison diagram of a vertical spare parts e-commerce recommendation method based on knowledge graph according to an embodiment of the present invention.

[0059] Fig. 9 This is a structural diagram of a vertical spare parts e-commerce recommendation system based on knowledge graph according to an embodiment of the present invention.

[0060] In the figure: data collection and preprocessing module-100, spare parts vertical knowledge graph construction module-200, graph mining model construction module-300, model optimization module-400, model verification module-500. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0063] It should be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0064] Specifically, the embodiments of the present invention are further described below in conjunction with the accompanying drawings.

[0065] See also Figure 1 , Figure 1 is a flow chart of a vertical spare parts e-commerce recommendation method based on knowledge graph provided by an embodiment of the present invention, such as Figure 1 As shown, the vertical spare parts e-commerce recommendation method based on knowledge graph includes steps S10 to S50.

[0066] S10. Acquire test data and preprocess the test data to obtain preprocessed test data so as to minimize junk data entering the model.

[0067] In an embodiment of the present invention, the test data can be acquired through a spare parts vertical e-commerce platform. The test data includes: user type information, spare parts type information, assembly type information and e-commerce platform information.

[0068] In an embodiment of the present invention, the preprocessing includes filling and filtering missing values ​​of the test data and standardizing the test data. In detail, the filling and filtering of missing values ​​in the data: for missing data that may affect the recommendation accuracy of the model and reduce the user experience, the average value is used for filling; the data of some delisted products and some blacklisted users are directly deleted to prevent these noise data from reducing the accuracy of the recommendation model. Data standardization: After the data standard deviation is standardized, the speed of model convergence is accelerated to save model training time and improve the accuracy of the model to a certain extent.

[0069] S20. Construct a vertical knowledge graph of spare parts based on the preprocessed test data.

[0070] In an embodiment of the present invention, a seven-step method is applied to complete the construction of the spare parts ontology, and seven major categories of first-level ontologies are defined, namely: assembly knowledge, user knowledge, spare parts knowledge, logistics platform knowledge, material supplier knowledge, retail platform knowledge and manufacturer knowledge.

[0071] In an embodiment of the present invention, the step S20 of constructing a vertical knowledge graph of spare parts based on preprocessed test data includes:

[0072] S201, based on the pre-processed test data, collecting spare parts vertical platform knowledge data;

[0073] It should be noted that the preprocessed test data can be divided into unstructured data, structured data and semi-structured data according to its degree of structure. Unstructured data is generally composed of text. Structured data is generally stored in a relational database. Semi-structured data is between structured and unstructured data, has a non-strict model, and is commonly found in XML and HTML format data. The data used to construct the knowledge graph comes from the e-commerce website data of the partner company, such as Figure 2 As shown, it belongs to structured data; Figure 4-5 As shown, it belongs to semi-structured data. The spare parts knowledge of this website is generated with reference to the standards of vertical fields at home and abroad to ensure the reliability and availability of components.

[0074] S202. Construct a spare parts vertical knowledge graph based on the spare parts vertical platform knowledge data.

[0075] Specifically, the knowledge data of the spare parts vertical platform is represented by various classified table structures and stored in the MySQL database. The data stored in the MySQL database is then converted by D2RQ and stored in the Neo4j graph database to obtain the spare parts vertical knowledge graph. Exemplarily, the spare parts knowledge graph may contain 3721 entities, including 2862 spare parts entities. After extraction, there are 3 user type entities, 2862 work entities, 4999 user entities, and 87944 relationship triples. Figure 4 It is the backbone graph of the spare parts knowledge graph in Neo4j, where the red entity is the part category entity, the yellow entity is the part entity, the green entity is the work category entity, the blue entity is the first-level ontology entity, the gray entity is the user category entity, and the pink entity is the work entity.

[0076] S30. Build a graph mining model (RippleNet) based on the vertical knowledge graph of spare parts.

[0077] In an embodiment of the present invention, the graph mining model is a graph user interest ripple propagation recommendation model that can embed entity node attribute features, focus on sampling to alleviate neighbor node explosion noise, and enhance interpretability of ripple paths.

[0078] In an embodiment of the present invention, S30, constructing a graph mining model (RippleNet) based on a vertical knowledge graph of spare parts, including:

[0079] S301. Build the RippleNet model as the basic model;

[0080] S302, designing meta-path rules according to preset rules; the preset rules can be designed according to the needs of the enterprise, and the meta-path design rules are as follows: Figure 5 As shown, user-product-user: user 1 bought the product, and another user 2 also bought the product; part-assembly-part: part 1 can be assembled with the assembly, and another part 2 can also be assembled with the assembly.

[0081] S303, retraining the node representation vector of the target type node by using a streaming feature embedding method;

[0082] S304. Use the PageRank algorithm to calculate the importance value of each entity node in the spare parts vertical knowledge graph, and perform sampling according to the node importance value during the knowledge ripple sampling process.

[0083] It should be noted that the ripple network model can generate multiple starting points on the knowledge graph and generate propagation ripples through the user's historical click records, and finally use these ripples to superimpose and generate the user's potential interest distribution for the product nodes in the knowledge graph, and use this to predict the user's click probability on the product nodes.

[0084] There are two main types of ripple networks. The first one is to converge from the outer n layers of the ripple starting point to the starting point to capture the user's preference distribution; the second one is to spread the ripple n layers outward from the starting point to capture the user's preference distribution. The typical model is RippleNet. Figure 6 The main workflow of the ripple network algorithm in this invention is shown as follows:

[0085] The model loss function is defined as shown in Formula 1, and expanded as shown in Formula 2, where

[0086]

[0087] Input: interaction matrix Y, spare parts vertical e-commerce knowledge graph G.

[0088] Output: prediction function Its specific form is shown in Formula 3.

[0089]

[0090] (1) Initialize model parameters and randomly initialize all node vectors in the knowledge graph using one-hot encoding;

[0091] (2) Calculate the ripple set of each user using Formula 4, Formula 5, and Formula 6;

[0092]

[0093] (3) Start training iteration;

[0094] (4) Sample small batches of positive and negative interaction samples from the interaction matrix Y;

[0095] (5) Sample a small batch of positive and negative knowledge triples from the spare parts vertical e-commerce knowledge graph G;

[0096] (6) Perform user preference propagation and calculate the user’s vector representation u using Formula 7, Formula 8, and Formula 9.

[0097]

[0098] (7) Calculate the gradient through small batch error back propagation and

[0099] (8) Update the parameters by gradient descent of the learning rate η and

[0100] (9) If the number of iterations has not been exceeded, return to step 3;

[0101] (10) Return the prediction function

[0102] Table 1 Ripple network algorithm variable definition

[0103]

[0104]

[0105] S40. Determine the spare parts knowledge graph meta-path of the graph mining model, embed the target type node features, and perform importance sampling design on the nodes in the spare parts knowledge graph.

[0106] The S40, determining the meta-path of the spare parts knowledge graph of the graph mining model, embedding the target type node features and designing importance sampling of the nodes in the spare parts knowledge graph, includes:

[0107] S401. Determine the meta-path of the spare parts knowledge graph based on the node distribution, relationship distribution, user preference habits, and calculation limitations of the spare parts knowledge graph.

[0108] Specifically, several meta-paths were designed based on the characteristics of the knowledge graph of the spare parts vertical field. Different meta-paths were embedded in the original graph user preference ripple mining model for experiments. The final result analysis verified the effectiveness of the meta-path in improving the model and selecting the optimal meta-path. Furthermore, by analyzing the node distribution, relationship distribution, user preference habits of the spare parts knowledge graph in the study and considering the calculation limitations, six different meta-path node type rules were finally determined for experiments, namely, node paths with only work types, node paths with only behavior types, node paths with both work and behavior types, node paths with both work and parts types, node paths with both behavior and parts types, and node paths with both behavior and work and social types.

[0109] S402: embedding target type node features.

[0110] In the embodiment of the present invention, S402, embedding target type node features, includes:

[0111] S4021, using the principal component analysis method to perform data preprocessing on the characteristics of the target type nodes;

[0112] Specifically, Figure 7 As shown in the figure, the experiment first uses the principal component analysis method to preprocess the features of the target type nodes to reduce the dimension and noise of the input features and improve the data quality and model training speed.

[0113] In the embodiment of the present invention, the step S4021, performing data preprocessing on the features of the target type nodes using a principal component analysis method, comprises the following steps:

[0114] S40211. Construct node feature data matrix X:

[0115] The node feature data set contains N features and M node feature samples, so that an N*M original data set matrix X can be constructed.

[0116] S40212. Perform Z-mean normalization on the node feature matrix X:

[0117] There are N features in the node features, and there are M target nodes in total. The jth feature of the i-th node is set to x ij ,in and j is the sample mean and sample standard deviation of the jth node; each value is converted into a standardized value using formula 10 And transform the node feature matrix X into matrix Q.

[0118]

[0119] S40213. Calculate the covariance matrix R of the matrix Q:

[0120] Find the covariance matrix R of M*M from the matrix Q. The covariance matrix is ​​defined as shown in formula 11. In formula 12, r ii =1, r ij =r ji , r ij is the correlation coefficient between the i-th feature and the j-th feature.

[0121] R=(r ij ) M*M (11)

[0122]

[0123] S40214. Calculate the eigenvalues ​​and eigenvectors of matrix R:

[0124] The eigenvector λ1≥λ2≥...≥λ of the matrix R is calculated by formula 11 m ≥0, and the corresponding eigenvector u1≥u2≥...≥u m ≥0, where D is the diagonal matrix of eigenvalues ​​of Q.

[0125] V -1 RV=D (13)

[0126] S40215. Sort the eigenvalues ​​in order from large to small.

[0127] S40216. Retain the eigenvectors corresponding to the first n largest eigenvalues.

[0128] S40217. Convert the original data into the new space.

[0129] Get the n*m ​​data matrix Y after dimensionality reduction T =X T *W, where the m*m matrix W is XX T The eigenvector matrix of .

[0130] S40218. Complete the conversion of original data.

[0131] S4022. The experiment performs pooling operations on the user preference vector, node vector, and node features obtained by the model and then retrains the target type node representation vector through a multi-layer perceptron model.

[0132] It should be noted that the node vector is obtained through linear mapping or multi-layer perceptron neural network model. Linear mapping is a special mapping method that can maintain vector addition and scalar multiplication between two vector spaces. In the user preference propagation recommendation model based on the knowledge graph, the linear mapping method is used to update the vector representation of the target product node. By aggregating the user's preference propagation on the knowledge graph as the user's preference representation, and using the original vector representation of the product node as input for linear mapping to the new vector space, the vector representation of the product node on the knowledge graph is finally updated. The knowledge graph node vector obtained by this method integrates the user's historical behavior information and has a better degree of distinction.

[0133] The workflow of the multi-layer perceptron neural network model includes:

[0134] S40221. Initialize all connection weight values ​​v in the multilayer perceptron network in and φ nj and threshold θ j , and the learning rate η.

[0135] S40222. Extract a number of data from the total data set D as the input data set.

[0136] Where d is the dimension of the x vector and l is the dimension of the y vector.

[0137] S40223. Provide input data to the neuron.

[0138] S40224. Calculate the input data set and propagate forward layer by layer.

[0139] S40225. Take the output layer as the final node representation vector representation.

[0140] S40226. Utilize the user preference ripple network mining model to calculate the result error.

[0141] S40227, the error E m Back propagates to the neurons in the hidden layer of the multilayer perceptron, and updates the connection weights v according to the error in and φ nj and the neuron threshold θ j .

[0142] S40228: If the termination condition is reached, stop the algorithm, otherwise jump to step S40222 for loop iteration.

[0143] S403. Use PageRank algorithm to perform importance sampling design on the nodes in the spare parts knowledge graph.

[0144] It should be noted that the prior art enhances the model's ability to extract graph node information by deeply mining node information on meta-paths, but the model algorithm uses random sampling to select the next connection node each time in this step. When the number of adjacent nodes of a node exceeds a certain number, the random sampling method will miss more important nodes in the network. The present invention uses the PageRank algorithm to calculate the importance of nodes in the spare parts knowledge graph instead of random sampling as a reference for model sampling.

[0145] The PageRank algorithm is a link analysis algorithm. In fact, the PageRank algorithm can also be used on directed graphs to calculate the importance of nodes in the spare parts knowledge graph and other problems.

[0146] The basic idea of ​​the PageRank algorithm is to define a random walk model on a directed graph, namely a first-order Markov chain, which describes the behavior of a random walker randomly visiting each node along the directed graph. Under certain conditions, the probability of visiting each node in the extreme case converges to a stable distribution. At this time, the stable probability value of each node is its PageRank value, which indicates the importance of the node. PageRank is defined recursively, and the calculation of PageRank can be performed through an iterative algorithm.

[0147] The step S403, performing importance sampling design on nodes in the spare parts knowledge graph by using the PageRank algorithm, includes:

[0148] S4031. Input the second constructed spare parts knowledge graph directed graph G, which has 3721 nodes.

[0149] S4032, let t=0, and select the initial vector x0.

[0150] S4033. Use the directed graph G to calculate the transfer matrix A:

[0151]

[0152] S4034, iterate and normalize the result vector:

[0153] y t+1 =Ax t (15)

[0154]

[0155] S4035, when || x t+1 -x t When ||<ε, let R=x t , stop the iteration.

[0156] S4036. Otherwise, set t=t+1 and execute step S4034.

[0157] S4037. Normalize R so that it represents a probability distribution.

[0158] S4038. Obtain the importance of all nodes in the spare parts knowledge graph.

[0159] S50. The effectiveness of the graph mining model was verified through experimental comparison and analysis of the pre-collected test data set.

[0160] Specifically, we used the redesigned model, the original graph ripple recommendation model, the logistic regression model, the Xgboost model, the factorization machine model, and the deep width model to conduct experimental comparisons to verify the effectiveness of the proposed solution. The results are as follows: Figure 8 As shown in Table 2, the accuracy (ACC) and the area under the receiver operating characteristic curve (AUC) are used as model evaluation indicators in the experiment. By comparing the results in Table 2, the following conclusions can be drawn:

[0161] (1) The two evaluation indicators of the redesigned model are the best among the six different models, with an accuracy of 0.787 and an AUC of 0.863.

[0162] (2) After the present invention reconstructs and improves the RippleNet model, the new model improves the accuracy of the original model by 6.5% over 0.722 and the AUC index by 8.1% over 0.782.

[0163] (3) The original RippleNet outperforms traditional machine learning algorithms on this dataset, proving that the accuracy of the model can be improved by adding additional information from the knowledge graph.

[0164] (4) Experiments have shown that by optimizing the ripple node sampler, designing ripple relationship screening, and embedding target type node features, the model's generalization ability and accuracy can be greatly improved.

[0165] Table 2 Comparison of prediction results of different click rate prediction models

[0166]

[0167] It should be understood that, although described in a certain order, these steps are not necessarily performed in sequence in the above order. Unless there is clear explanation in this article, the execution of these steps does not have strict order restriction, and these steps can be performed in other orders. Moreover, a part of the steps of the present embodiment may include a plurality of steps or a plurality of stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a part of the steps or stages in other steps or other steps.

[0168] In one embodiment, see Fig. 9 As shown, in an embodiment of the present invention, a vertical spare parts e-commerce recommendation system based on a knowledge graph is also provided, which includes a data acquisition preprocessing module 100, a spare parts vertical knowledge graph construction module 200, a graph mining model construction module 300, a model optimization module 400 and a model verification module 500.

[0169] The data acquisition and preprocessing module 100 is used to acquire test data and preprocess the test data to obtain preprocessed test data.

[0170] The spare parts vertical knowledge graph construction module 200 is used to construct a spare parts vertical knowledge graph based on preprocessed test data.

[0171] The graph mining model building module 300 is used to build a graph mining model based on the spare parts vertical knowledge graph.

[0172] The model optimization module 400 is used to determine the spare parts knowledge graph meta-path of the graph mining model, embed the target type node features and perform importance sampling design on the nodes in the spare parts knowledge graph.

[0173] The model verification module 500 is used to perform experimental comparison and analysis through a pre-collected test data set to verify the effectiveness of the graph mining model.

[0174] It should be understood that, as used herein, the singular form "a" or "an" is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, "and / or" refers to any and all possible combinations of one or more of the items listed in association. The serial numbers of the embodiments disclosed in the above embodiments of the present invention are for description only and do not represent the advantages and disadvantages of the embodiments.

[0175] A person skilled in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the disclosure of the embodiments of the present invention (including the claims) is limited to these examples; under the concept of the embodiments of the present invention, the technical features in the above embodiments or different embodiments can also be combined, and there are many other changes in different aspects of the embodiments of the present invention as above, which are not provided in detail for the sake of simplicity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the protection scope of the embodiments of the present invention.

Claims

1. A vertical spare parts e-commerce recommendation method based on knowledge graph, characterized in that: The method includes: S10, acquiring test data, and preprocessing the test data to obtain preprocessed test data; S20, constructing a vertical knowledge graph of spare parts based on the preprocessed test data; S30. Construct a graph mining model based on the vertical knowledge graph of spare parts; S40, determining the spare parts knowledge graph meta-path of the graph mining model, embedding target type node features, and performing importance sampling design on the nodes in the spare parts knowledge graph; S50. The effectiveness of the graph mining model was verified through experimental comparison and analysis of the pre-collected test data set.

2. The vertical spare parts e-commerce recommendation method based on knowledge graph according to claim 1, characterized in that: The test data includes: user type information, spare parts type information, assembly part type information and e-commerce platform information.

3. The vertical spare parts e-commerce recommendation method based on knowledge graph as claimed in claim 2, characterized in that: The preprocessing includes filling and filtering missing values ​​of the test data and performing data standardization on the test data.

4. The vertical spare parts e-commerce recommendation method based on knowledge graph according to claim 1, characterized in that: The step S20, constructing a vertical knowledge graph of spare parts based on the preprocessed test data, includes: S201, based on the pre-processed test data, collecting spare parts vertical platform knowledge data; S202. Construct a spare parts vertical knowledge graph based on the spare parts vertical platform knowledge data.

5. The vertical spare parts e-commerce recommendation method based on knowledge graph as claimed in claim 4, characterized in that: S202. Constructing a spare parts vertical knowledge graph based on spare parts vertical platform knowledge data, including: The knowledge data of the spare parts vertical platform is represented by various classified table structures and stored in the MySQL database. The data stored in the MySQL database is then converted through D2RQ and stored in the Neo4j graph database to obtain the spare parts vertical knowledge graph.

6. The vertical spare parts e-commerce recommendation method based on knowledge graph according to claim 1, characterized in that: The step S30, constructing a graph mining model based on the vertical knowledge graph of spare parts, includes: Build the RippleNet model as the base model; Design meta-path rules based on preset rules; Retrain the node representation vector of the target type node through the streaming feature embedding method; The PageRank algorithm is used to calculate the importance value of each entity node in the spare parts vertical knowledge graph, and sampling is performed according to the node importance value during the knowledge ripple sampling process.

7. The vertical spare parts e-commerce recommendation method based on knowledge graph according to claim 1, characterized in that: The S40, determining the meta-path of the spare parts knowledge graph of the graph mining model, embedding the target type node features and designing importance sampling of the nodes in the spare parts knowledge graph, includes: S401, determining the meta-path of the spare parts knowledge graph according to the node distribution, relationship distribution, user preference habits and calculation constraints of the spare parts knowledge graph; S402, embedding target type node features; S403. Use PageRank algorithm to perform importance sampling design on the nodes in the spare parts knowledge graph.

8. The vertical spare parts e-commerce recommendation method based on knowledge graph according to claim 7 is characterized in that: S402, embedding target type node features, including: S4021, using the principal component analysis method to perform data preprocessing on the characteristics of the target type nodes; S4022. The experiment performs pooling operations on the user preference vector, node vector, and node features obtained by the model and then retrains the target type node representation vector through a multi-layer perceptron model.

9. The vertical spare parts e-commerce recommendation method based on knowledge graph according to claim 8, characterized in that: The step S4021, using a principal component analysis method to perform data preprocessing on the characteristics of the target type nodes, comprises the following steps: S40211, constructing a node feature data matrix X; S40212, performing Z-mean normalization processing on the node feature matrix X; S40213. Calculate the covariance matrix R of the matrix Q; S40214. Calculate the eigenvalues ​​and eigenvectors of the matrix R; S40215. Sort the eigenvalues ​​in descending order; S40216. retain the eigenvectors corresponding to the first n largest eigenvalues; S40217, convert the original data into the new space; S40218. Complete the conversion of original data.

10. A vertical spare parts e-commerce recommendation system based on knowledge graph, characterized in that: The system includes: a data collection and preprocessing module, a spare parts vertical knowledge graph construction module, a graph mining model construction module, a model optimization module and a model verification module. The data acquisition preprocessing module is used to acquire test data and preprocess the test data to obtain preprocessed test data. The spare parts vertical knowledge graph construction module is used to construct the spare parts vertical knowledge graph based on the preprocessed test data. The graph mining model building module is used to build a graph mining model based on the spare parts vertical knowledge graph. The model optimization module is used to determine the spare parts knowledge graph meta-path of the graph mining model, embed the target type node features, and perform importance sampling design on the nodes in the spare parts knowledge graph. The model verification module is used to conduct experimental comparison and analysis through pre-collected test data sets to verify the effectiveness of the graph mining model.

Images