Knowledge graph-based transportation tool matching recommendation method and system

Abstract

The invention is suitable for the field of artificial intelligence, and provides a transportation tool matching recommendation method and system based on a knowledge graph, and the method comprises the steps: converting a triple into a text form through employing an improved knowledge graph embedding model KGE, carrying out the coding of an input text through a BERT model, and obtaining the context representation of each Token; capturing global semantic information of the entities and the relationships based on the aggregation representation of the triple to obtain embedded representations of the entities and the relationships; integrating and extracting features of the heterogeneous information, and performing dimension reduction processing on feature data of the heterogeneous information by using principal component analysis (PCA) to obtain embedded representation of the heterogeneous information; the method comprises the following steps: performing feature dimension reduction on entities, relationships and heterogeneous information based on an embedded layer of a recommendation system algorithm NFM model, capturing a second-order interaction relationship between dimension reduction features through bilinear interaction operation, learning high-order feature interaction by using a feed-forward neural network FNN, outputting a matching probability of goods and a transportation tool, and obtaining a matching result of the goods and the transportation tool; and based on the matching probability, performing matching recommendation of the goods and the transportation tool.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to a transportation tool matching recommendation method and system based on knowledge graph. Background Art

[0002] With the development of the cold chain logistics industry, how to effectively match goods with suitable means of transportation has become an important issue.

[0003] In related technologies, link prediction in knowledge graphs is the task of predicting the exact relationship between entities. Traditional knowledge graph link prediction methods mainly rely on shallow models (such as TransE and DistMult). Although these models can be quickly expanded to large knowledge graphs, they are difficult to capture complex semantic relationships due to their limited expressive power. For example, the TransE model represents relationships through simple vector translation operations and cannot effectively handle complex relationships such as many-to-many and one-to-many. Although the DistMult model improves its expressive power through matrix decomposition, it performs poorly when processing noisy data.

[0004] At present, the technology of transport matching recommendation based on knowledge graph needs to comprehensively consider multi-dimensional information such as temperature sensitivity of goods, temperature control performance of transport tools, and user evaluation. Existing recommendation systems (such as collaborative filtering and content-based recommendation) often ignore the importance of these heterogeneous information, resulting in insufficient accuracy and interpretability of recommendation results. For example, traditional collaborative filtering methods only make recommendations based on user rating data, and cannot effectively utilize the complex relationship between the temperature requirements of goods and the temperature control capabilities of transport tools. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method and system for recommending transportation tools matching based on a knowledge graph, aiming to solve the technical problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions.

[0007] An embodiment of the present invention provides a method for recommending transportation means matching based on a knowledge graph, the method comprising the following steps: Based on the cold chain information data, a logistics information knowledge graph is constructed. The potential semantic relationships in the knowledge graph are mined through a graph neural network. The scoring data is associated with the entities in the knowledge graph to form a triple (h, r, t) of the user-goods-transportation tool multidimensional relationship network. Using the improved knowledge graph embedding model KGE, the triple (h, r, t) is converted into text form, and the input text is encoded through the BERT model to obtain the contextual representation of each token; based on the aggregate representation of the triple, the global semantic information of entities and relations is captured to obtain the embedded representation of entities and relations; Integrate and extract the features of heterogeneous information, use principal component analysis (PCA) to reduce the dimension of the feature data of heterogeneous information, and obtain the embedded representation of heterogeneous information; The embedding layer of the NFM model based on the recommendation system algorithm performs feature dimensionality reduction on entity embedding representation, relationship embedding representation and heterogeneous information embedding representation, captures the second-order interaction relationship between the reduced-dimensional features through bilinear interaction operations, uses the feedforward neural network FNN to learn high-order feature interactions, and outputs the matching probability of goods and transportation tools.

[0008] Furthermore, before the step of constructing a logistics information knowledge graph based on the cold chain information data, the method further includes a step of preprocessing the data, and the step of preprocessing the data includes: Remove noise from comment data through regular expressions and rule matching; Use entity linking to map entities in different data sources to a unified standard identifier; The BiLSTM-CRF model based on the attention mechanism is used to extract the relationship between entities from text data.

[0009] Furthermore, in the step of encoding the input text through the BERT model, the encoding of BERT is expressed as follows: BERT(x)=Transformer(Embedding(x)), wherein x is the input text, Transformer is a multi-layer self-attention mechanism, and Embedding represents the process of mapping the input text x to a low-dimensional vector space.

[0010] Furthermore, in the step of obtaining the embedded representation of entities and relations, the step of obtaining the embedded representation of entities includes: The representation of the CLS tag of the BERT model is used as the entity embedding of the head entity h or the tail entity t; or for the head entity h and the tail entity t, the representation of the corresponding token in the BERT output is extracted, and the token representation of the entity is averaged or concatenated to obtain a more fine-grained entity embedding.

[0011] Furthermore, in the step of obtaining the embedded representation of entities and relationships, the step of obtaining the embedded representation of relationships includes: using the Token representation corresponding to the relationship as the relationship embedding, or using the aggregate representation of triples as the relationship embedding.

[0012] Furthermore, the step of using principal component analysis (PCA) to perform dimensionality reduction processing on the feature data of heterogeneous information includes: Clean and organize the data into a form suitable for PCA; Calculate the mean value of each feature of each sample, and subtract the mean value of the feature from the data of each sample so that the data can be normalized; Calculate the covariance matrix of the data; find the eigenvalues ​​and eigenvectors of the covariance matrix, rearrange the eigenvalues ​​and eigenvectors in order from large to small, calculate the cumulative contribution rate of the eigenvalues, and select a subset of the eigenvector set according to the cumulative contribution rate according to a preset ratio; transform the original data to obtain the reduced-dimensional data.

[0013] Furthermore, in the step of capturing the second-order interaction relationship between the dimension reduction features through the bilinear interaction operation, the formula of the dual interaction layer is expressed as follows: ; ; Among them, x i and x j is the eigenvector; v i and v j is the embedding vector; T represents the transposition operation; Bi represents the output of the double interaction layer, that is, the second-order interaction relationship between the feature vectors; i and j represent the index of the feature vector; n represents the number of feature vectors; W represents the weight matrix.

[0014] Another embodiment of the present invention provides a transportation tool matching recommendation system based on knowledge graph, which includes the following modules: The knowledge graph construction module is used to build a logistics information knowledge graph based on cold chain information data, mine the potential semantic relationships in the knowledge graph through the graph neural network, associate the scoring data with the entities in the knowledge graph, and form a triple (h, r, t) of the user-goods-transportation tool multidimensional relationship network; The structural information embedding module is used to convert the triple (h, r, t) into text form using the improved knowledge graph embedding model KGE, and encode the input text through the BERT model to obtain the contextual representation of each token; based on the aggregate representation of the triple, the global semantic information of entities and relations is captured to obtain the embedded representation of entities and relations; The heterogeneous information embedding module is used to integrate and extract the features of heterogeneous information, and use principal component analysis (PCA) to reduce the dimension of the feature data of heterogeneous information to obtain the embedded representation of heterogeneous information; The transportation tool matching module is used to reduce the feature dimension of entity embedding representation, relationship embedding representation and heterogeneous information embedding representation based on the embedding layer of the recommendation system algorithm NFM model, capture the second-order interaction relationship between the reduced-dimensional features through bilinear interaction operations, use the feedforward neural network FNN to learn high-order feature interactions, and output the matching probability of goods and transportation tools.

[0015] Compared with the prior art, the transportation tool matching recommendation method and system based on knowledge graph of the present invention has the following beneficial effects: First, the present invention constructs a logistics information knowledge graph based on cold chain information data, which contains multiple semantic relationships. By mining the potential semantic relationships in the knowledge graph through a graph neural network, the complex interactive information in cold chain logistics can be captured. Second, the present invention uses an improved knowledge graph embedding model KGE to convert triples (h, r, t) into text form, and encodes the input text through the BERT model to obtain the contextual representation of each Token; based on the aggregate representation of triples, the global semantic information of entities and relationships is captured, and the embedded representation of entities and relationships is obtained, so that the semantic capture capability of the present invention is strong, and the complex semantic relationships in the knowledge graph are effectively captured through the context perception capability of BERT; compared with the traditional model, the improved knowledge graph embedding model KGE can reduce the impact of noise and irrelevant information on model performance; Third, the present invention integrates and extracts the features of heterogeneous information, uses principal component analysis (PCA) to reduce the dimensionality of the feature data of heterogeneous information, obtains an embedded representation of the heterogeneous information, and improves the accuracy of recommendations. Specifically, by integrating heterogeneous information such as temperature sensitivity and temperature control performance, the accuracy of recommendations is significantly improved; PCA dimensionality reduction reduces the data dimension and improves the computational efficiency of the model; Fourth, the present invention performs feature dimensionality reduction on entities, relationships and heterogeneous information based on the embedding layer of the recommendation system algorithm NFM model, captures the second-order interaction relationship between the reduced-dimensional features through bilinear interaction operations, uses the feedforward neural network FNN to learn high-order feature interactions, and outputs the matching probability of goods and transportation tools, thereby realizing multimodal information integration. It can simultaneously process numerical, categorical and text features, and is suitable for multi-source data in cold chain logistics. The nonlinear relationship capture of the present invention captures the complex nonlinear relationship between features through neural networks, thereby improving the recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, 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.

[0017] Figure 1This is a flow chart of an implementation method of a transportation tool matching recommendation method based on a knowledge graph of the present invention; Figure 2 A sub-flow chart of a transportation tool matching recommendation method based on knowledge graph of the present invention; Figure 3 Another sub-flow chart of a transportation tool matching recommendation method based on knowledge graph of the present invention; Figure 4 A module block diagram of a transportation tool matching recommendation system based on knowledge graph of the present invention; Figure 5 The present invention is a structural block diagram of a computer device. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] In related technologies, link prediction of knowledge graphs is the task of predicting the exact relationships between entities.

[0020] Traditional knowledge graph link prediction methods mainly rely on shallow models (such as TransE and DistMult). Although these models can be quickly expanded to large knowledge graphs, they have difficulty capturing complex semantic relationships due to their limited expressive power. For example, the TransE model represents relationships through simple vector translation operations and cannot effectively handle complex relationships such as many-to-many and one-to-many. Although the DistMult model improves its expressive power through matrix decomposition, it performs poorly when processing noisy data.

[0021] At present, the technology of transport matching recommendation based on knowledge graph needs to comprehensively consider multi-dimensional information such as temperature sensitivity of goods, temperature control performance of transport tools, and user evaluation. Existing recommendation systems (such as collaborative filtering and content-based recommendation) often ignore the importance of these heterogeneous information, resulting in insufficient accuracy and interpretability of recommendation results. For example, traditional collaborative filtering methods only make recommendations based on user rating data, and cannot effectively utilize the complex relationship between the temperature requirements of goods and the temperature control capabilities of transport tools.

[0022] The present invention proposes a method and system for transport tool matching recommendation based on knowledge graph. The logistics information knowledge graph CCD-KG is constructed using the cold chain information dataset provided by the cold chain information network. The existing knowledge graph embedding model (KGE) is improved and an improved knowledge graph embedding model KGE is proposed to effectively capture the potential information in the knowledge graph. At the same time, the present invention also integrates other heterogeneous information to improve the prediction performance. Finally, NFM is used as a classifier to predict cold chain logistics recommendations.

[0023] The specific implementation of the transportation tool matching recommendation method and system based on knowledge graph of the present invention is described in detail below in conjunction with specific embodiments.

[0024] Please refer to Figure 1 In an embodiment of the present invention, a method for recommending transportation means matching based on a knowledge graph is provided, and the method comprises the following steps: S101: Build a logistics information knowledge graph based on cold chain information data, mine the potential semantic relationships in the knowledge graph through graph neural networks, associate the scoring data with the entities in the knowledge graph, and form a triple (h, r, t) of the user-goods-transportation tool multidimensional relationship network; In step S101 of the embodiment of the present invention, the scoring data mainly comes from the user's comments on cold chain distribution, which may include the user's specific evaluation of the satisfaction of the transportation tool, the integrity of the goods, the delivery time, etc.; through sentiment analysis, these comments can be converted into specific scoring data. For example, the user may give a score to the service quality of the transportation tool (such as 1 point to 10 points) in the comment, or evaluate the preservation effect of the goods during transportation; after the user gives the score, the scoring data is associated with the entities in the knowledge graph (such as users, goods, and transportation tools). For example, assuming that user A gives a score of 8 points to the goods C transported by transportation tool B, then a triple (user A, evaluation, goods C and transportation tool B) can be constructed in the knowledge graph, and its score of 8 points can be used as the weight or attribute of the relationship. In this way, the user's satisfaction with a specific transportation tool and goods combination can be recorded in the knowledge graph, thereby providing valuable reference information for the subsequent recommendation system.

[0025] In the construction of the triple (h, r, t) of the embodiment of the present invention, by associating the rating data with the entities in the knowledge graph, the present invention constructs a triple (h, r, t) of the user-goods-transportation tool multidimensional relationship network; in this triple, h represents the head entity (such as user A), r represents the relationship (such as evaluation), and t represents the tail entity (such as goods C and transportation tool B). In addition, the present invention can also use the rating as an attribute of the relationship so as to consider the user's satisfaction factor in the subsequent recommendation process.

[0026] In the present invention, the latent semantic relations in the knowledge graph are mined through the graph neural network GNN. For example, GAT (Graph Attention Network) is used to capture the complex interactions between goods and transportation tools.

[0027] By adopting step S101, the present invention constructs a logistics information knowledge graph based on cold chain information data, including multiple semantic relationships, and mines potential semantic relationships in the knowledge graph through a graph neural network, so as to capture complex interactive information in cold chain logistics; In an optional embodiment, the present invention uses graph database tools such as Neo4j to intuitively display the relationship between goods and transportation tools, and supports efficient query and recommendation.

[0028] For further information, please refer to Figure 1 The recommendation method provided in the embodiment of the present invention further includes the following steps: S102: Using the improved knowledge graph embedding model KGE, convert the triple (h, r, t) into text form, and encode the input text through the BERT model to obtain the contextual representation of each token; based on the aggregate representation of the triple, capture the global semantic information of entities and relations, and obtain the embedded representation of entities and relations; Among them, the BERT model of the embodiment of the present invention encodes the input text through a multi-layer self-attention mechanism to obtain the contextual representation of each Token. These representations contain rich semantic information and can effectively capture the complex relationships in the text; the entity embedding and relationship embedding of the embodiment of the present invention are obtained through the output of the BERT model. The entity embedding can use the representation of the [CLS] tag, or average or concatenate the Token representation of the entity; the relationship embedding can use the Token representation corresponding to the relationship, or use the aggregate representation of the triple.

[0029] Furthermore, in an embodiment of the present invention: Triple encoding: Convert the triples (h, r, t) in the knowledge graph into text form, such as "[CLS] Truck A can transport cargo B[SEP]", where [CLS] and [SEP] are special tags of BERT; BERT encoding: Encode the input text through the BERT model to obtain the contextual representation of each Token; BERT encoding is represented as follows: BERT(x)=Transformer(Embedding(x)), where x is the input text, Transformer is a multi-layer self-attention mechanism, and Embedding represents the process of mapping the input text x to a low-dimensional vector space; BERT encoding of the present invention is one of the key steps in knowledge graph embedding. The input text is encoded through the BERT model to obtain the contextual representation of each Token. The specific implementation is as follows: (1) Use Hugging Face's Transformers library to load the pre-trained BERT model; (2) Segment the input text to obtain a Token sequence; (3) Use the BERT model to encode the Token sequence to obtain the contextual representation of each Token; Aggregate Representation: Use the [CLS] tagged representation as the aggregate representation of the entire triple to capture the global semantic information of entities and relations.

[0030] By adopting step S102, the semantic capture capability of the present invention is strong, and the complex semantic relationships in the knowledge graph are effectively captured through the context perception capability of BERT; compared with the traditional model, the improved knowledge graph embedding model KGE can reduce the impact of noise and irrelevant information on the model performance; S103: Integrate and extract the features of heterogeneous information, use principal component analysis PCA to perform dimensionality reduction processing on the feature data of the heterogeneous information, and obtain an embedded representation of the heterogeneous information; In the step of integrating and extracting features of heterogeneous information of the present invention, the integration of heterogeneous information includes data alignment, for example, aligning the temperature requirements of goods (such as refrigeration, freezing) with the temperature control performance of transportation tools (such as refrigeration capacity, temperature stability); extracting the core features of heterogeneous information; for example: the minimum temperature requirement of the goods and the minimum refrigeration temperature of the transportation tool.

[0031] It can be seen that the present invention improves the accuracy of recommendations. Specifically, by integrating heterogeneous information such as temperature sensitivity and temperature control performance, the accuracy of recommendations is significantly improved. The present invention adopts PCA dimensionality reduction to reduce the data dimension and improve the computational efficiency of the model. S104: Based on the embedding layer of the recommendation system algorithm NFM model, the entity embedding representation, relationship embedding representation and heterogeneous information embedding representation are subjected to feature dimensionality reduction. The second-order interaction relationship between the reduced-dimensional features is captured through bilinear interaction operations. The feedforward neural network FNN is used to learn high-order feature interactions and output the matching probability of goods and transportation tools.

[0032] The present invention realizes multimodal information integration and can process numerical, categorical and textual features at the same time, and is suitable for multi-source data in cold chain logistics. The nonlinear relationship capture of the present invention captures the complex nonlinear relationship between features through a neural network, thereby improving the recommendation effect.

[0033] In the step of using a feedforward neural network FNN to learn high-order feature interactions described in the present invention, the feedforward neural network (FNN) is used to deeply learn and extract high-order feature interactions. After being processed by the embedding layer and the double interaction layer, the obtained features already contain rich information, including entities, relationships, and second-order interaction relationships of heterogeneous data. However, feature interactions are often complex and nonlinear, and cannot be fully captured by only linear combinations or simple bilinear interactions.

[0034] The FNN of the present invention can perform nonlinear transformation on these features through a multi-layer neural network structure, and gradually learn higher-order feature combinations. Each layer of the neural network is composed of multiple neurons, each of which performs a weighted summation on the input features and applies a nonlinear activation function (such as ReLU); this nonlinear transformation enables the network to capture the complex relationships and high-order interactions between features; for example, the nonlinear relationship between the sensitivity of goods to temperature and the temperature control performance of transportation tools, and the interaction between user evaluation and transportation tool selection; In practical applications, the training process of the FNN of the present invention optimizes the network parameters through the back propagation algorithm to minimize the error between the prediction results and the actual matching situation. Finally, the output layer of the FNN gives the matching probability of the goods and the means of transport, which reflects their degree of adaptation and the credibility of the recommendation.

[0035] Therefore, the dual interaction layer of the embodiment of the present invention captures the second-order interaction relationship between the dimensionality reduction features through a bilinear interaction operation; this interaction relationship can enhance the expressiveness of the model and improve the accuracy of the recommendation; the feedforward neural network (FNN) of the present invention is used to learn high-order feature interactions and output the matching probability of goods and transportation tools; wherein, the FNN performs feedforward calculation on the input features through a multi-layer perceptron (MLP) model, which can capture the complex nonlinear relationship between the features.

[0036] For further information, please refer to Figure 2 In an embodiment of the present invention, before the step of constructing a logistics information knowledge graph based on the cold chain information data, the method further includes a step of preprocessing the data, and the step of preprocessing the data includes: S201: removing noise from comment data through regular expressions and rule matching; S202: Map entities in different data sources to a unified standard identifier using entity linking; S203: Use the BiLSTM-CRF model based on the attention mechanism to extract the relationship between entities from text data.

[0037] In step S201 of the present invention, noise in the comment data, such as advertisements and irrelevant symbols, is removed by regular expression and rule matching. In this step, regular expression matching can be used to match and delete HTML tags, special characters, etc. in the comments, and comment content irrelevant to cold chain logistics can be deleted by rule matching. in: The regular expression provided in the embodiment of the present disclosure uses a single string to describe or match a series of strings. When removing noise from comment data, the regular expression can identify and filter out text with specific patterns, such as advertising links, irrelevant symbols and special characters. For example, for advertising links, which usually contain patterns such as "http" and "www", the present invention can also use the regular expression / http[s]?: / / \S+ / i to match and filter out these links; The rule matching provided in the embodiments of the present disclosure refers to filtering and processing text according to predefined rules, which can be based on text features such as length, format, specific keywords, etc. For example, a rule can be defined to remove comments containing a large number of consecutive repeated letters (such as "aaaaa") or special symbols (such as "@#$%^").

[0038] Further, step S202 of the embodiment of the present invention is an entity standardization step, in which entity linking technology (such as DBpedia Spotlight) is used to map entities in different data sources to a unified standard identifier; for example, "refrigerated truck" and "frozen truck" are mapped to "cold chain transportation tool"; this step can use DBpedia Spotlight API for entity linking to map entity names in different data sources to a unified standard identifier; Furthermore, step S203 of the embodiment of the present invention is a step of relation extraction, in which the BiLSTM-CRF model based on the attention mechanism is used to extract the relationship between entities from the text data. For example, the relationship "can be transported" is extracted from the sentence "truck A can transport cargo B"; this step can use the BiLSTM-CRF model for relation extraction, and when training the model, the labeled data set is used for training to extract the relationship between entities.

[0039] Furthermore, in the step of obtaining the embedded representation of entities and relations, the step of obtaining the embedded representation of entities includes: The representation of the CLS tag of the BERT model is used as the entity embedding of the head entity h or the tail entity t; or for the head entity h and the tail entity t, the representation of the corresponding token in the BERT output is extracted, and the token representation of the entity is averaged or concatenated to obtain a more fine-grained entity embedding.

[0040] Furthermore, in the step of obtaining the embedded representation of entities and relationships, the step of obtaining the embedded representation of relationships includes: using the Token representation corresponding to the relationship as the relationship embedding, or using the aggregate representation of triples as the relationship embedding.

[0041] For further information, please refer to Figure 3 In an embodiment of the present invention, the step of using principal component analysis PCA to perform dimensionality reduction processing on feature data of heterogeneous information includes: S301: sort and organize the data into a form suitable for PCA; S302: Calculate the average value of each feature of each sample, and subtract the average value of the feature from the data of each sample so that the data can be normalized; S303: Calculate the covariance matrix of the data; find the eigenvalues ​​and eigenvectors of the covariance matrix, rearrange the eigenvalues ​​and eigenvectors in descending order, calculate the cumulative contribution rate of the eigenvalues, and select a subset of the eigenvector set according to the cumulative contribution rate and a preset ratio; S304: Convert the original data to obtain data after dimension reduction.

[0042] Step S301 of the embodiment of the present invention is a data standardization process, which normalizes heterogeneous information to eliminate dimensional effects; in the covariance matrix calculation in step S303, the covariance matrix of the data is calculated to extract the main feature directions; in dimensionality reduction and embedding, the feature vector with the highest cumulative contribution rate is selected, and the high-dimensional data is mapped to the low-dimensional space to obtain the embedded representation of the heterogeneous information.

[0043] The present invention significantly improves the accuracy of recommendations by integrating heterogeneous information such as temperature sensitivity and temperature control performance; the PCA dimensionality reduction of the present invention reduces the data dimension and improves the computational efficiency of the model.

[0044] Furthermore, in the step of capturing the second-order interaction relationship between the dimension reduction features through the bilinear interaction operation, the formula of the dual interaction layer is expressed as follows: ; ; Among them, x i and x j is the eigenvector; v i and v jis the embedding vector; T represents the transposition operation; Bi represents the output of the double interaction layer, that is, the second-order interaction relationship between the feature vectors; i and j represent the index of the feature vector; n represents the number of feature vectors; W represents the weight matrix; The feedforward neural network (FNN) of the present invention is used to learn high-order feature interactions and output the matching probability of goods and transportation tools. The specific implementation is as follows: (1) constructing a multi-layer perceptron (MLP) model; (2) using the MLP model to perform feedforward calculations on input features and output the matching probability.

[0045] The embedding layer of the NFM model of the recommendation system algorithm of the present invention maps features such as goods, transportation tools, and user ratings to a low-dimensional vector space to solve the data sparsity problem; The NFM model of the recommendation system algorithm of the present invention realizes multimodal information integration, can process numerical, categorical and textual features at the same time, and is suitable for multi-source data in cold chain logistics; the present invention captures the complex nonlinear relationship between features through a neural network to improve the recommendation effect.

[0046] Please refer to Figure 4 In another embodiment of the present invention, a transportation tool matching recommendation system based on knowledge graph is provided, and the system includes the following modules: The knowledge graph construction module 401 is used to construct a logistics information knowledge graph based on the cold chain information data, mine the potential semantic relationship in the knowledge graph through the graph neural network, associate the scoring data with the entities in the knowledge graph, and form a triple (h, r, t) of the user-goods-transportation tool multidimensional relationship network; The structural information embedding module 402 is used to convert the triple (h, r, t) into text form by using the improved knowledge graph embedding model KGE, and encode the input text through the BERT model to obtain the context representation of each token; based on the aggregate representation of the triple, the global semantic information of the entity and the relationship is captured to obtain the embedding representation of the entity and the relationship; The heterogeneous information embedding module 403 is used to integrate and extract the features of the heterogeneous information, and use principal component analysis PCA to perform dimensionality reduction processing on the feature data of the heterogeneous information to obtain an embedded representation of the heterogeneous information; The transportation tool matching module 404 is used to perform feature dimensionality reduction on the entity embedding representation, relationship embedding representation and heterogeneous information embedding representation based on the embedding layer of the recommendation system algorithm NFM model, capture the second-order interaction relationship between the reduced-dimensional features through bilinear interaction operations, use the feedforward neural network FNN to learn high-order feature interactions, and output the matching probability of goods and transportation tools.

[0047] Please refer to Figure 5In another embodiment of the present disclosure, a computer device is further provided. The computer device 500 of the embodiment of the present disclosure includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the steps of the transportation tool matching recommendation method based on the knowledge graph in any of the above embodiments are implemented; The computer device 500 may also include: an input system and an output system; the processor, memory, input system and output system may be connected via a bus or other means, and the input system may receive input digital or character information, and generate signal input related to transportation tool matching recommendations based on the knowledge graph.

[0048] The output system may include display devices such as display screens.

[0049] Furthermore, the present invention also provides a computer-readable storage medium having computer-readable instructions stored thereon, and when the computer-readable instructions are executed by a processor, the steps of the transportation tool matching recommendation method based on the knowledge graph as provided above are implemented.

[0050] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0051] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, the functional units in the various embodiments of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0052] In a typical configuration of an embodiment of the present invention, the terminal, the device of the service network and the computing device include one or more processors (CPU), an input / output interface, a network interface and a memory.

[0053] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0054] Computer readable media include permanent and non-permanent, removable and non-removable media and can store information by any method or technology. The information can be computer readable instructions, data structures, program modules or other data.

[0055] Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0056] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0057] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some or all of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

[0058] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A transportation tool matching recommendation method based on knowledge graph, characterized in that: The recommended approach includes the following steps: Based on the cold chain information data, a logistics information knowledge graph is constructed. The potential semantic relationships in the knowledge graph are mined through a graph neural network. The scoring data is associated with the entities in the knowledge graph to form a triple (h, r, t) of the user-goods-transportation tool multidimensional relationship network. Using the improved knowledge graph embedding model KGE, the triple (h, r, t) is converted into text form, and the input text is encoded through the BERT model to obtain the contextual representation of each token; based on the aggregate representation of the triple, the global semantic information of entities and relations is captured to obtain the embedded representation of entities and relations; Integrate and extract the features of heterogeneous information, use principal component analysis (PCA) to reduce the dimension of the feature data of heterogeneous information, and obtain the embedded representation of heterogeneous information; The embedding layer of the NFM model based on the recommendation system algorithm performs feature dimensionality reduction on entity embedding representation, relationship embedding representation and heterogeneous information embedding representation, captures the second-order interaction relationship between the reduced-dimensional features through bilinear interaction operations, uses the feedforward neural network FNN to learn high-order feature interactions, and outputs the matching probability of goods and transportation tools.

2. According to the knowledge graph-based transportation tool matching recommendation method of claim 1, it is characterized in that: Before the step of constructing a logistics information knowledge graph based on the cold chain information data, the method further includes a step of preprocessing the data, and the step of preprocessing the data includes: Remove noise from comment data through regular expressions and rule matching; Use entity linking to map entities in different data sources to a unified standard identifier; The BiLSTM-CRF model based on the attention mechanism is used to extract the relationship between entities from text data.

3. The method for recommending transportation tools matching based on knowledge graph according to claim 2, characterized in that: In the step of encoding the input text through the BERT model, the encoding of BERT is expressed as follows: BERT(x)=Transformer(Embedding(x)), where x is the input text, Transformer is a multi-layer self-attention mechanism, and Embedding represents the process of mapping the input text x to a low-dimensional vector space.

4. The method for recommending transportation tools matching based on knowledge graph according to claim 3 is characterized in that: In the step of obtaining the embedded representation of entities and relations, the step of obtaining the embedded representation of entities includes: The representation of the CLS tag of the BERT model is used as the entity embedding of the head entity h or the tail entity t; or for the head entity h and the tail entity t, the representation of the corresponding token in the BERT output is extracted, and the token representation of the entity is averaged or concatenated to obtain a more fine-grained entity embedding.

5. The method for recommending transportation tools matching based on knowledge graph according to claim 4, characterized in that: In the step of obtaining the embedding representation of entities and relationships, the step of obtaining the embedding representation of relationships includes: using the Token representation corresponding to the relationship as the relationship embedding, or using the aggregate representation of triples as the relationship embedding.

6. The method for recommending transportation tools matching based on knowledge graph according to claim 5, characterized in that: The step of using principal component analysis PCA to perform dimensionality reduction processing on the feature data of heterogeneous information includes: Clean and organize the data into a form suitable for PCA; Calculate the mean value of each feature of each sample, and subtract the mean value of the feature from the data of each sample so that the data can be normalized; Calculate the covariance matrix of the data; find the eigenvalues ​​and eigenvectors of the covariance matrix, rearrange the eigenvalues ​​and eigenvectors in descending order, calculate the cumulative contribution rate of the eigenvalues, and select a subset of the eigenvector set according to the cumulative contribution rate and a preset ratio; The original data is transformed to obtain the data after dimensionality reduction.

7. The method for recommending transportation tools matching based on knowledge graph according to claim 6, characterized in that: In the step of capturing the second-order interaction relationship between the dimension reduction features through the bilinear interaction operation, the formula of the double interaction layer is expressed as follows: ； ； Among them, x i and x j is the eigenvector; v i and v j is the embedding vector; T represents the transposition operation; Bi represents the output of the double interaction layer, that is, the second-order interaction relationship between the feature vectors; i and j represent the index of the feature vector; n represents the number of feature vectors; W represents the weight matrix.

8. A recommendation system for implementing the transportation tool matching recommendation method based on knowledge graph as claimed in any one of claims 1 to 7, characterized in that: The recommendation system includes the following modules: The knowledge graph construction module is used to build a logistics information knowledge graph based on cold chain information data, mine the potential semantic relationships in the knowledge graph through the graph neural network, associate the scoring data with the entities in the knowledge graph, and form a triple (h, r, t) of the user-goods-transportation tool multidimensional relationship network; The structural information embedding module is used to convert the triple (h, r, t) into text form using the improved knowledge graph embedding model KGE, and encode the input text through the BERT model to obtain the contextual representation of each token; based on the aggregate representation of the triple, the global semantic information of entities and relations is captured to obtain the embedded representation of entities and relations; The heterogeneous information embedding module is used to integrate and extract the features of heterogeneous information, and use principal component analysis (PCA) to reduce the dimension of the feature data of heterogeneous information to obtain the embedded representation of heterogeneous information; The transportation tool matching module is used to reduce the feature dimension of entity embedding representation, relationship embedding representation and heterogeneous information embedding representation based on the embedding layer of the recommendation system algorithm NFM model, capture the second-order interaction relationship between the reduced-dimensional features through bilinear interaction operations, use the feedforward neural network FNN to learn high-order feature interactions, and output the matching probability of goods and transportation tools.

Images