Space launch information intelligent pushing method and system based on knowledge graph

Abstract

The invention discloses a spaceflight launching information intelligent pushing method and system based on a knowledge graph, and the method comprises the steps: collecting data information of a spaceflight launching site from multiple dimensions, and carrying out the preprocessing; constructing a knowledge graph based on the spaceflight launch site data information; constructing a user interest model based on the knowledge graph; based on the user interest model, constructing a spaceflight launch site information intelligent pushing model based on the knowledge graph; and realizing intelligent pushing of spaceflight launching information by using the intelligent pushing model of the spaceflight launching site information based on the knowledge graph. According to the method, a pushing method based on a knowledge graph and a graph convolutional neural network (GCN) is adopted, and intelligent pushing of spaceflight launching site information is realized through technologies such as data sampling and GCN; the high dimension and heterogeneity challenges of the knowledge graph are overcome, meanwhile, the complex relation between entities is captured through the GCN, and the accuracy and diversity of the push model are improved.

Description

Technical Field

[0001] The present invention belongs to the field of aerospace, and specifically relates to an intelligent push method and system for aerospace launch information based on a knowledge graph. Background Art

[0002] With the rapid rise of the information age, as a key technology to cope with the challenge of information overload, the recommendation system is increasingly becoming the dominant way for people to obtain personalized information. In this context, as one of the core fields of scientific and technological innovation, information transmission and sharing in the aerospace field have become increasingly crucial. In traditional recommendation systems, collaborative filtering is widely recognized because it assumes that similar interacting users have similar interests. However, with the continuous expansion of Internet resources, the information influx faced by users is becoming increasingly huge, and traditional recommendation systems show certain limitations in meeting personalized needs.

[0003] Currently, users' information acquisition needs are not only limited to product recommendations, but also involve knowledge acquisition, learning, and cross-domain information discovery. To help users better cope with this challenge, intelligent information push based on knowledge graphs has gradually become the focus of research. A knowledge graph, as a structured and semantically rich knowledge representation method, can more accurately capture changes in user interests and provide more accurate information for the recommendation system. Compared with traditional systems, recommendation systems based on knowledge graphs can understand user needs more comprehensively and adapt to interest changes more flexibly. Especially in the context of an aerospace launch site where information involves multiple fields and is highly complex, the push model based on knowledge graphs has more advantages.

[0004] In summary, for information push at aerospace launch sites, traditional recommendation systems have deficiencies in meeting users' personalized needs, and the push model based on knowledge graphs has more advantages. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention proposes an intelligent push method and system for aerospace launch information based on a knowledge graph. By analyzing users' historical behaviors, a precise user interest model is constructed to provide personalized support for intelligent push; a push method based on a knowledge graph and a graph convolutional neural network (GCN) is adopted, and through technologies such as data sampling and GCN, intelligent push of aerospace launch site information is realized; the challenges of high dimensionality and heterogeneity of the knowledge graph are overcome, and at the same time, complex relationships between entities are captured through GCN, improving the accuracy and diversity of the push model.

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

[0007] An intelligent push method for aerospace launch information based on a knowledge graph, comprising the following steps:

[0008] Collect space launch site data information from multiple dimensions and perform preprocessing; among them, the collected space launch site data information includes: the real-time status of the launch site infrastructure, the specific details of mission planning and execution, and the monitoring data of weather and environment;

[0009] Construct a knowledge graph based on the space launch site data information;

[0010] Construct a user interest model based on the knowledge graph;

[0011] Construct an intelligent push model of space launch site information based on the knowledge graph based on the user interest model;

[0012] Use the intelligent push model of space launch site information based on the knowledge graph to realize the intelligent push of space launch information.

[0013] Preferably, the method for preprocessing the collected space launch site data information includes: two stages of data cleaning and data transformation;

[0014] Among them, in the data cleaning stage: remove incorrect, duplicate or irrelevant data to eliminate noise, reasonably fill in missing values through context or historical data, and standardize the data in each stage of space launch to unify the measurement unit and format; in the data transformation stage: perform word segmentation, stop word removal, and stemming processing on the launch text information.

[0015] Preferably, the method for constructing a knowledge graph based on the space launch site data information includes:

[0016] Perform data annotation on the space launch site data information, use the annotated data to train the named entity recognition model, relation extraction model, and sequence annotation model, and apply the trained model to the data source to be extracted for information extraction;

[0017] Integrate the extracted data sources for multi-source entities, use the fuzzy matching algorithm to identify the same or similar entities, and eliminate naming differences through context information or domain knowledge base to unify the entity format standard;

[0018] Import the fused data source into the Neo4j database, construct the knowledge graph by using the Cypher language, represent the nodes as entities, the edges as the relationships between entities, and maintain the structural integrity of the graph.

[0019] Preferably, the method for constructing a user interest model based on the knowledge graph includes:

[0020] Extract the features and preferences of users through data mining, association analysis, and clustering analysis techniques;

[0021] Construct a user profile for each user based on the extracted user characteristics and preferences, and then construct a user model.

[0022] Based on the user model, construct a multi-dimensional user demand label system, namely the user interest model.

[0023] Preferably, the method for constructing an intelligent information push model of the space launch site based on the knowledge graph based on the user interest model includes:

[0024] Based on the user interest model, construct a user-information interaction matrix.

[0025] Based on the user-information interaction matrix, evaluate the importance of the relationship to the user.

[0026] Based on the importance of the relationship to the user, combine the node information and the neighborhood information to update the node information.

[0027] Based on the updated node information, calculate the predicted probability of the user for the information.

[0028] Based on the predicted probability of the user for the information, calculate the loss between the result predicted by the algorithm and the score in the scoring matrix, and continuously iterate the weights to make the algorithm converge, and complete the construction of the intelligent information push model of the space launch site based on the knowledge graph.

[0029] Preferably, the method for constructing a user-information interaction matrix based on the user interest model includes:

[0030] Construct a user-information interaction matrix Y∈R M×N , where U = {u1, u2,..., u M} represents M users, V = {v1, v2,..., v M} represents N pieces of information, and the value of y uv indicates whether the user has an interaction with the information. The following is an interaction matrix of 3 users and 4 space launch information:

[0031]

[0032] Preferably, the method for evaluating the importance of the relationship to the user based on the user-information interaction matrix:

[0033] The information v∈V corresponds to the entity e∈E;

[0034] represents the relationship between the entity e i and e j ;

[0035] describes the importance of the relationship r to the user u, where u represents the user and r represents the relationship.

[0036] Preferably, the method for calculating the predicted probability of a user for information includes:

[0037] The calculation formula for the predicted probability of user vector u for information vector v is as follows:

[0038] y u ′ v = predict(v, v′, u′, u),

[0039] where u represents the user vector; u′ represents the feature vector of the aggregated and combined user u; v represents the item vector; v′ represents the aggregated and combined feature vector; predict() represents an arbitrary function.

[0040] Preferably, the method for realizing intelligent push of space launch information by using the intelligent push model of space launch site information based on the knowledge graph includes:

[0041] The overall input of the intelligent push model of space launch information includes the space launch knowledge graph G v (ε, R), the knowledge graph G u (τ, R) of user requirements; the interaction matrix Y between the user and the space launch information; the domain sampling mapping S v , S u ; the hyperparameters predict(·), agg(·), combined with the calculated predicted probability of the user for the information, and the predicted probability y′ of the user for the information is obtained after the algorithm converges uv , and personalized push of the space launch site information is performed according to the predicted probability y′ uv .

[0042] The present invention also provides an intelligent push system for space launch information based on a knowledge graph, including: an acquisition module, a first construction module, a second construction module, a third construction module, and a push module;

[0043] The acquisition module is used to collect space launch site data information from multiple dimensions and perform preprocessing; among them, the collected space launch site data information includes: the real-time status of the launch site infrastructure, the specific details of the mission plan and execution, and the monitoring data of the weather and environment;

[0044] The first construction module is used to construct a knowledge graph based on the space launch site data information;

[0045] The second construction module is used to construct a user interest model based on the knowledge graph;

[0046] The third construction module is used to construct an intelligent push model of space launch site information based on the knowledge graph based on the user interest model;

[0047] The push module is used to realize intelligent push of space launch information by using the intelligent push model of space launch site information based on the knowledge graph.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] 1. Apply the knowledge graph to the field of intelligent push of space launch site information. Through the construction of the knowledge graph, the complex relationships in the space launch site information are successfully captured, improving the accuracy of the push model.

[0050] 2. The application of the knowledge graph not only achieves results in information extraction and fusion, but also promotes the performance improvement of the recommendation system through in-depth analysis of user behavior.

[0051] 3. Research the method based on the knowledge graph and graph convolutional neural network (KGCN), bringing new technical means to the push model. Based on the method of combining graph convolutional neural network and knowledge graph, through steps such as node message combination, data sampling, topological neighbor structure information, and prediction, an intelligent push model of space launch site information is constructed. This model not only improves the recommendation effect, but also makes the push results more interpretable, providing a better user experience for users. Description of the Drawings

[0052] In order to more clearly illustrate the technical solutions of the present invention, the following briefly introduces the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0053] Figure 1 It is the flowchart for constructing an intelligent push model of space launch site information based on the knowledge graph according to an embodiment of the present invention;

[0054] Figure 2 It is the schematic diagram of the knowledge graph according to an embodiment of the present invention;

[0055] Figure 3 It is the flowchart for data collection and preprocessing according to an embodiment of the present invention;

[0056] Figure 4 It is the flowchart for constructing the knowledge graph according to an embodiment of the present invention;

[0057] Figure 5 It is the flowchart for constructing a user interest model according to an embodiment of the present invention;

[0058] Figure 6 It is the intelligent push flowchart of space launch site information based on the knowledge graph according to an embodiment of the present invention. Detailed Embodiments

[0059] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0060] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0061] Embodiment 1

[0062] The present invention provides an intelligent push method for space launch information based on a knowledge graph, including the following steps:

[0063] Collect space launch site data information from multiple dimensions and perform preprocessing; among them, the collected space launch site data information includes: the real-time status of the launch site infrastructure, the specific details of mission plans and executions, and the monitoring data of weather and environment;

[0064] Build a knowledge graph based on the space launch site data information;

[0065] Build a user interest model based on the knowledge graph;

[0066] Build an intelligent push model for space launch site information based on the knowledge graph based on the user interest model;

[0067] Use the intelligent push model for space launch site information based on the knowledge graph to realize the intelligent push of space launch information.

[0068] As Figure 1 shown, this process includes the following steps:

[0069] S11, data collection and preprocessing.

[0070] In the process of building an intelligent push model for space launch site information based on a knowledge graph, data collection and preprocessing are key links. Collect and preprocess space launch site data information from multiple dimensions.

[0071] S12, build a knowledge graph.

[0072] A knowledge graph is a data graph designed to accumulate and convey real-world knowledge, where nodes represent entities and edges represent relationships between entities, usually represented by a triple G=(h, r, t), where h∈E represents the head entity, r∈R represents the relationship, and t∈E represents the tail entity. For example, (launch mission, person in charge, Zhang San) can form a triple. AsFigure 2 as shown

[0073] Triangles represent classes, rectangles represent entities, and rhombuses represent relationships. From this knowledge graph, we can obtain the following knowledge:

[0074] 1) The person in charge of the launch mission is Zhang San; 2) The launch mission is a task; 3) Zhang San is a person; 4) The participating organization of the launch mission is XX organization; 5) XX organization is an organization.

[0075] First, natural language processing (NLP) technology and information extraction algorithms are used to extract key information such as entities, relationships, and attributes from the space launch site information, forming a triple structure of "entity - relationship - entity". Then, multi-source knowledge is comprehensively and accurately integrated, and finally, the Neo4j graph database is used for the storage of the knowledge graph.

[0076] S13, construct a user interest model.

[0077] Adopt the method of implicit feedback data to construct a user interest model. Through technologies such as data mining, association analysis, and clustering analysis, the implicit feedback data is carefully mined to extract the characteristics and preferences of users, so as to accurately describe user needs and behaviors, and thus construct a more accurate user model.

[0078] S14, construct an intelligent push model for space launch site information based on the knowledge graph.

[0079] First, construct a user - information relationship matrix, then evaluate the importance of the relationship to the user, then combine the node with the neighborhood information and update the node information, calculate the predicted attention of the user to the information, and calculate the loss between the prediction result and the score in the score matrix. Finally, by calculating the loss, continuously update and iterate the weights to make the algorithm converge.

[0080] Furthermore, data collection and preprocessing are as Figure 3 shown, including the following steps:

[0081] S21, collect data from multiple dimensions.

[0082] Comprehensively utilize multi-source data, including space launch site infrastructure information, mission plan and execution data, weather and environment data, personnel and equipment status data, etc., to ensure that the model has a comprehensive and accurate information basis. The diversity of space launch site information requires data collection from multiple dimensions. This includes collecting the real-time status of the launch site infrastructure, the specific details of mission plans and executions, and the monitoring data of weather and environment. Data sources include but are not limited to various sensors, mission execution systems, meteorological data centers, etc.

[0083] S22, data preprocessing.

[0084] Perform data preprocessing on multi-source data, which mainly includes two stages: data cleaning and data transformation. In the data cleaning stage, remove incorrect, duplicate or irrelevant data to eliminate noise, reasonably fill in missing values through context or historical data, and standardize the data at each stage of space launch to unify the measurement units and formats. In the data transformation stage, perform operations such as word segmentation, stop word removal, and stemming on the launch text information. Finally, integrate the data into a unified space launch dataset, providing a solid foundation for subsequent knowledge graph construction.

[0085] Furthermore, the knowledge graph construction process is as Figure 4 shown, including the following steps:

[0086] S31, information extraction.

[0087] In the construction process of the information intelligent push model, information extraction is a crucial step. Information extraction aims to extract key information such as entities, relationships, and attributes from structured data and semi-structured data, forming a triple structure of "entity-relationship-entity", providing a basis for the establishment of the knowledge graph.

[0088] Through a detailed analysis of the space launch site information, deeply understand the characteristics and requirements of the information. Include the relationships between different entities, clarify the key characteristics such as the structure, relationships, and attributes of the information, providing clear guidance for the direction and goals of information extraction. First, clarify and define key entities, such as launch sites, rocket models, satellite missions, etc., and their respective attributes, such as geographical locations, launch capabilities, mission objectives, etc.; then determine the relationships between entities, including the adaptation relationship between the launch site and the rocket, the carrying relationship between the rocket and the satellite, the execution relationship between the mission and the time, etc.; select a graph database model for the construction of the knowledge graph to better organize and manage the information, ensuring the structural integrity and data accuracy of the knowledge graph.

[0089] To achieve efficient information extraction, first annotate part of the space launch information, and then use the annotated data to train the named entity recognition (NER) model, relationship extraction model, and sequence annotation model, and further apply the trained models to the data source to be extracted for information extraction.

[0090] The NER model aims to identify entities with specific meanings in text, such as person names, place names, organization names, times, dates, etc. The input layer receives the text data to be processed; the feature extraction layer extracts features useful for named entity recognition from the text; the encoding layer converts the extracted features into a format that the model can understand; the decoding layer classifies each word or character in the text according to the features output by the encoding layer to determine whether it belongs to a certain named entity category and gives the boundaries of the entity; the output layer outputs the results of named entity recognition, including the category and boundary information of the entity. The training process of this model is to first prepare a labeled corpus in the aerospace field, and then use these labeled data to train the model. By continuously optimizing the parameters of the model, it can accurately identify the named entities in the text. Finally, a model that can effectively perform named entity recognition on unseen text is obtained.

[0091] The input layer in the relation extraction model receives the preprocessed text data; the feature extraction layer extracts features useful for relation extraction from the text; the encoding layer converts the extracted features into a format that the model can understand and further captures the sequence information and semantic information in the text; the relation classification layer classifies the entity pairs in the text to judge whether there is a specific relationship between them and the type of the relationship; the output layer outputs the results of relation extraction, including whether there is a relationship between the entity pairs and the type of the relationship. The specific training process is to first collect and label a corpus containing the relationships between entities, and then use these labeled data to train the model. By adjusting the model parameters, its ability to recognize and extract the relationships between entities from the text is optimized. During the training process, the model will learn how to understand the context, identify entities and judge the specific relationship types existing between these entities. Finally, the trained model can be applied to new text data to automatically extract and output the relationship information between entities.

[0092] The input layer in the sequence labeling model converts text into numerical vectors, usually using word embedding or character embedding techniques to capture the semantic information of words. The feature extraction layer uses RNNs (such as LSTM or GRU) to process sequence data, capture long-range dependencies, and extract context features. The context encoding layer further enhances the feature representation, possibly by introducing gating mechanisms (such as Highway Networks) or attention mechanisms for optimization. The sequence labeling layer adopts structures such as CRF, considers the labeling information of the entire sequence, and calculates the globally optimal labeling sequence. The decoding layer uses algorithms such as Viterbi decoding to find the optimal labeling sequence from the output of CRF as the final output of the model. The specific training process is to prepare sequence data containing labeling information as the training set, which labels the category of each element in the sequence. Use this data to train the model, and through iterative optimization of the model parameters, enable the model to learn the dependencies and context information between elements in the sequence to accurately predict the labeling category of each element in the sequence. Finally, a model that can accurately label unseen sequence data is obtained.

[0093] Specifically, it includes identifying key entities in space launch missions from a large amount of text and data: spacecraft names, launch centers, launch times, etc.; extracting complex relationships between entities: "Spacecraft XX" was successfully launched at "YY Launch Center" on "XX month XX day, ZZ year", etc.; extracting detailed attribute information related to the mission: the weight of the spacecraft, orbit type, scientific experimental equipment carried, etc.

[0094] S32, Knowledge fusion.

[0095] The entity formats in space launch information from multiple sources are not standardized, and the names may also vary. To solve these problems, it is necessary to comprehensively and accurately integrate multi-source knowledge. During the knowledge fusion process, first, the integration of multi-source entities needs to be carried out, and a fuzzy matching algorithm is used to identify the same or similar entities. The process of the fuzzy matching algorithm is as follows: Use the edit distance to measure the difference. Let two strings S = S1S2S3…Sm and T = T1T2T3…Tn, construct the matrix LD[m + 1, n + 1], and use the idea of dynamic programming to loop and calculate the value of each cell LD(i,j) in the matrix. The LD(m,n) in the lower right corner is the size of the required edit distance, and the calculation formula is as follows:

[0096]

[0097] Among them, Min = min{LD(i - 1, j)+1, LD(i, j - 1)+1, LD(i - 1, j - 1)+f(i, j)}, where when the i-th word of S is not equal to the j-th word of T, f(i, j)=1; otherwise, f(i, j)=0. The larger the LD, the smaller the similarity. When the value of LD exceeds the set threshold, the two entities can be considered similar or identical.

[0098] Then, eliminate naming differences through context information or domain knowledge bases, and unify the entity format standards, including names, data types, precisions, units, etc., to ensure the consistency and accuracy of the knowledge graph. Knowledge fusion enables the space launch site information intelligent push model to better process knowledge data from multiple sources and improve the data quality and overall performance of the model.

[0099] S33, Knowledge storage.

[0100] The storage of the knowledge graph is a key link in the research. In the storage stage, the Neo4j graph database is selected, which is a high-performance non-relational database. Graph databases are popular for their ability to store entities and relationships in a graph structure form, and are faster in query speed compared to relational databases. The visualization display ability of Neo4j further improves the clarity of the knowledge graph. During the process of knowledge storage, the knowledge graph file of the pre-processed information is imported into the Neo4j database. By using the Cypher language to construct the knowledge graph, nodes are represented as entities and edges are represented as relationships between entities to maintain the structural integrity of the graph. The following is a code example for constructing the knowledge graph:

[0101]

[0102] Furthermore, the process of constructing the user interest model is as Figure 5 shown, including the following steps:

[0103] S41, Extract the features and preferences of the user.

[0104] In-depth mining and analysis of implicit feedback data has become a key means to understand user behavior and preferences. As a highly complex and technology-intensive field, the space launch site generates implicit feedback data during its operation. Implicit feedback data refers to various behavioral data generated by users when using recommendation systems or related services that do not explicitly express their intentions. These data do not rely on users' direct evaluation or selection, but indirectly reflect their preferences and interests through their behavior. For example, the online browsing behavior, search query records, virtual tour paths, etc. of users (including researchers, engineers, audiences and potential partners) contain a wealth of valuable information about interest orientation, technical focus and cooperation intentions. Through data mining, association analysis, cluster analysis and other technologies, this information can be effectively extracted to more accurately grasp user characteristics and preferences.

[0105] Data mining is used to extract useful information for space launch site operations from massive amounts of implicit feedback data. Data mining reveals hot spots visited by users, technical fields of interest, and potential service demand trends, including the growing interest of users in new rocket technologies and the increased attention paid to the safety of launch processes, thereby providing data support for facility upgrades, technology introduction, and market promotion at launch sites.

[0106] Correlation analysis further explores the inherent connections between these implicit feedback data. By analyzing the behavior pattern of users searching for "rocket engine efficiency" and then paying attention to "fuel technology", correlation analysis can reveal the close relationship between different technical concerns, helping launch site managers identify potential technical cooperation opportunities or market trends, optimize resource allocation, and promote technological innovation.

[0107] Cluster analysis focuses on segmenting user groups with similar interests, needs or behavioral characteristics. This helps provide customized space launch information push, visit experience or cooperation plans for different user groups.

[0108] S42, build a more accurate user model.

[0109] By carefully mining implicit feedback data, extracting user characteristics and preferences, and based on the extracted characteristics, building a detailed user portrait for each user, including basic information, behavioral characteristics and interest characteristics, to accurately describe user needs and behaviors, thereby building a more accurate user model and more accurately recognizing user characteristics.

[0110] S43, establish a user demand labeling system.

[0111] Based on the user model, a multi-dimensional user demand labeling system is constructed to more comprehensively describe and meet user needs.

[0112] Construct user interest tags based on the user's browsing history, followed content, etc. on platforms such as the official website of the space launch site and social media, directly reflecting the user's interest points and preferences for space technology. By analyzing the user's behavior data on platforms related to the space launch site, extract the user's behavior characteristic tags to describe the user's behavior patterns and habits, which helps to understand the user's activity and participation. The space launch site involves a diverse user group, including scientific researchers, engineers, educators, media practitioners, and the general public, etc. Construct professional role tags to distinguish the professional backgrounds and needs of different users.

[0113] Further, the intelligent information push process of the space launch site based on the knowledge graph is as Figure 6 shown, including the following steps:

[0114] S51, construct a user-information interaction matrix.

[0115] Constructing a user-information interaction matrix is a key step in the recommendation system and personalized service. It can help us understand and analyze the interaction relationship between users and information. This matrix can be constructed based on the user's implicit feedback data (such as browsing history, click behavior, etc.) to reveal the user's interest and preference for different information.

[0116] The user-information interaction matrix is usually a two-dimensional array, where the rows represent users, the columns represent items, and each element in the matrix records the interaction information between the user and the item. The information can be represented as binary, that is, whether the user has had an interaction with the item.

[0117] Specifically, in the research of the intelligent information push scenario of the space launch site, construct the user interaction matrix Y∈R M×N , where U = {u1, u2,..., u M} represents M users, V = {v1, v2,..., v N} represents N pieces of information, and the value of y uv indicates whether the user has an interaction with the information. If the user u has an interaction (such as subscription, push, etc.) with the information v, then y uv = 1, otherwise y uv = 0. The following is the interaction matrix of 3 users and 4 space launch information:

[0118]

[0119] S52, evaluate the importance of the relationship to the user.

[0120] Given the user-information interaction matrix Y and the knowledge graph G, evaluate the importance of the relationship to the user and predict whether the user u has potential interest in the information v that he has not interacted with before.

[0121] Specifically, the information v ∈ V corresponds to the entity e ∈ E. For example, the information "launch time" appears as an entity with the same name in the knowledge graph. Represents the entity e i With e j The relationship between them. Describes the importance of the relationship r to the user u, where u represents the user and r represents the relationship.

[0122] To describe the topological neighborhood structure of the information v, the linear combination of the v neighborhood is calculated as follows: Where Is the normalized user-relationship score. Represents the linear combination of the neighborhoods for the user u and the entity e; N represents the neighborhood set, that is, the set of other entities directly connected to the entity e. When calculating the neighborhood representation of an entity, since targeted neighborhood aggregation of specific user scores is required, the user relationship score can act as a personalized filter.

[0123] S53, Combine the node information with the neighborhood information to update the node information.

[0124] Specifically, it is necessary to combine the node information v with its neighborhood information To update the node information. The three combination methods are as follows:

[0125]

[0126] Sum combination: Take the sum of the two representation vectors, and then perform a non-linear transformation. For example, sum the information vector of the user u and the information vector of the space launcher type, and then apply a non-linear transformation such as ReLU. Where w and b are the weights and biases of the transformation respectively, and σ is a non-linear function such as the rectified linear unit function.

[0127]

[0128] Concatenation combination: First concatenate the two representation vectors before applying the non-linear transformation, and the dimension of the concatenated vector becomes twice the original.

[0129]

[0130] Neighbor combination: Directly use the neighborhood representation of the entity v as the output representation. Message combination is a key step. Through message combination, the representation of the message can be combined with its neighbors.

[0131] S54, Calculate the prediction probability of the user for the information.

[0132] The formula for calculating the prediction probability of the user vector u for the information vector v is as follows:

[0133] y u ′v = predict(v, v′, u′, u)

[0134] u represents the user vector, and the feature vector of user u after aggregation and combination is u′; v represents the item vector, and the feature vector after aggregation and combination is v′. predict() can be any function, such as the inner product, and other forms can also be adopted to adapt to specific requirements and scenarios.

[0135] S55. Calculate the loss between the result predicted by the algorithm and the score in the scoring matrix, and continuously iterate the weights to make the algorithm converge.

[0136] To obtain better recommendation results, a negative sampling strategy is used during training to calculate the loss between the result predicted by the algorithm and the score in the scoring matrix. By calculating the loss, the weights are continuously updated and iterated to make the algorithm converge. The following loss function is adopted in the research of the intelligent information push scenario of the space launch site:

[0137]

[0138] where y uv′ represents the preference value of user u for item v′, the predicted preference value of user u for item v, is the cross-entropy loss, P is the negative sampling distribution, and T u is the number of negative samples of user u. Here, T u = |{v: y uv = 1}| and P follows a uniform distribution. The last term is the L2 regularizer (λ is the hyperparameter of the regularization strength, which controls the weight of the regularization term in the loss function).

[0139] S56. Information push

[0140] The overall input of the intelligent information push model for space launch information includes the space launch knowledge graph G v (ε, R), the knowledge graph G u (τ, R) of user requirements, where ε represents the attributes of the space launch knowledge graph, τ represents the attributes of the user requirement knowledge graph, and R represents the set of relationships, describing the connections between different entities; the interaction matrix Y between the user and the space launch information; the domain sampling mapping S v of the space launch field and the user requirement field, S u ; the hyperparameters predict(·), agg(·), combined with the calculation of the prediction probability of the user for the information in step S54. After the algorithm converges, the prediction probability y′ of the user for the information can be obtained uv , and the space launch site information can be personalized pushed according to the prediction probability.

[0141] Example 2

[0142] The present invention also provides an intelligent push system for space launch information based on a knowledge graph, including: a collection module, a first construction module, a second construction module, a third construction module, and a push module;

[0143] The collection module is used to collect space launch site data information from multiple dimensions and perform preprocessing; among them, the collected space launch site data information includes: the real-time status of the launch site infrastructure, the specific details of mission planning and execution, and the monitoring data of weather and environment;

[0144] The first construction module is used to construct a knowledge graph based on the space launch site data information;

[0145] The second construction module is used to construct a user interest model based on the knowledge graph;

[0146] The third construction module is used to construct an intelligent push model for space launch site information based on the knowledge graph based on the user interest model;

[0147] The push module is used to utilize the intelligent push model for space launch site information based on the knowledge graph to realize the intelligent push of space launch information.

[0148] 1. By analyzing the user's historical behavior, an accurate user interest model is constructed, providing personalized support for intelligent push;

[0149] 2. Adopt a push method based on a knowledge graph and a graph convolutional neural network (GCN); through technologies such as data sampling and GCN, the intelligent push of space launch site information is realized;

[0150] 3. Overcome the challenges of high dimensionality and heterogeneity of the knowledge graph, and at the same time capture the complex relationships between entities through GCN, improving the accuracy and diversity of the push model.

[0151] The above-described embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. An intelligent push method for space launch information based on a knowledge graph, characterized in that, It includes the following steps: Collect space launch site data information from multiple dimensions and perform preprocessing; among them, the collected space launch site data information includes: the real-time status of the launch site infrastructure, the specific details of mission plans and executions, and the monitoring data of weather and environment; Construct a knowledge graph based on the space launch site data information; Construct a user interest model based on the knowledge graph; Construct an intelligent push model of space launch site information based on the knowledge graph based on the user interest model; Use the intelligent push model of space launch site information based on the knowledge graph to realize the intelligent push of space launch information.

2. The intelligent push method for space launch information based on a knowledge graph according to claim 1, wherein The method for preprocessing the collected space launch site data information includes: two stages of data cleaning and data transformation; Among them, in the data cleaning stage: remove incorrect, duplicate or irrelevant data to eliminate noise, reasonably fill in missing values through context or historical data, and standardize the data in each stage of space launch to unify the measurement unit and format; in the data transformation stage: perform word segmentation, stop word removal, and stemming processing on the launch text information.

3. The intelligent push method for space launch information based on a knowledge graph according to claim 1, wherein The method for constructing a knowledge graph based on the space launch site data information includes: Perform data annotation on the space launch site data information, use the annotated data to train the named entity recognition model, relationship extraction model, and sequence annotation model, and apply the trained models to the data source to be extracted for information extraction; Integrate multi-source entities of the extracted data source, use the fuzzy matching algorithm to identify the same or similar entities, and eliminate naming differences through context information or domain knowledge base to unify the entity format standard; Import the fused data source into the Neo4j database, construct the knowledge graph by using the Cypher language, represent the nodes as entities, the edges as the relationships between entities, and maintain the structural integrity of the graph.

4. The intelligent push method for space launch information based on a knowledge graph according to claim 1, wherein The method for constructing a user interest model based on the knowledge graph includes: Extract the features and preferences of users through data mining, association analysis, and clustering analysis techniques; Based on the extracted features and preferences of users, construct a user portrait for each user, and then construct a user model; Based on the user model, construct a multi-dimensional user demand label system, that is, a user interest model.

5. The intelligent push method for space launch information based on a knowledge graph according to claim 1, wherein The method for constructing an intelligent push model of space launch site information based on the knowledge graph based on the user interest model includes: Construct a user-information interaction matrix based on the user interest model; Evaluate the importance of the relationship to the user based on the user-information interaction matrix; Based on the importance of the relationship to the user, combine the node information and neighborhood information to update the node information; Based on the updated node information, calculate the predicted probability of the user for the information; Based on the predicted probability of the user for the information, calculate the loss between the result predicted by the algorithm and the score in the scoring matrix, and continuously iterate the weights to make the algorithm converge, and complete the construction of the intelligent push model of space launch site information based on the knowledge graph.

6. The intelligent push method for space launch information based on a knowledge graph according to claim 5, wherein The method for constructing a user-information interaction matrix based on the user interest model includes: Construct the user-information interaction matrix Y∈R M×N , where U = {u1, u2, …, u M} represents M users, V = {v1, v2, …, v M} represents N pieces of information, and the value of y uv indicates whether there is interaction between the user and the information. The following shows the interaction matrix of 3 users and 4 space launch information:

7. The intelligent push method for space launch information based on a knowledge graph according to claim 5, characterized in that The method for evaluating the importance of the relationship to the user based on the user-information interaction matrix: The information v ∈ V corresponds to the entity e ∈ E; Represent entity e i with e j the relationship between Describes the importance of relationship r to user u, where u represents the user and r represents the relationship.

8. The intelligent push method for space launch information based on a knowledge graph according to claim 5, wherein The method for calculating the predicted probability of information for a user includes: The calculation formula for the predicted probability of the user vector u for the information vector v is as follows: y′ uv = predict(v, v′, u′, u), Wherein, u represents the user vector; u' represents the feature vector of the aggregated and combined user u; v represents the item vector; v' represents the aggregated and combined feature vector; Predict() represents any function.

9. The intelligent push method for space launch information based on a knowledge graph according to claim 5, wherein The method for implementing intelligent push of space launch information by using the intelligent push model of space launch site information based on a knowledge graph includes: The overall input of the intelligent push model for space launch information includes the space launch knowledge graph G v (ε, R), the knowledge graph G u (τ, R) of user requirements; the interaction matrix Y between the user and space launch information; the domain sampling mapping S v , S u ; hyperparameters predict(·), agg(·), the predicted probability of the user's information obtained by combining calculations, and the predicted probability y' of the user's information is obtained after the algorithm converges uv , according to the predicted probability y' uv Personalized push of space launch site information is performed 10. An intelligent push system for space launch information based on a knowledge graph, characterized in that, Including: An acquisition module, a first construction module, a second construction module, a third construction module, and a push module; The acquisition module is used to collect space launch site data information from multiple dimensions and perform preprocessing; among them, the collected space launch site data information includes: the real-time status of the launch site infrastructure, the specific details of mission plans and executions, and the monitoring data of weather and environment; The first construction module is used to construct a knowledge graph based on the space launch site data information; The second construction module is used to construct a user interest model based on the knowledge graph; The third construction module is used to construct an intelligent push model of space launch site information based on a knowledge graph based on the user interest model; The push module is used to use the intelligent push model of space launch site information based on a knowledge graph to realize intelligent push of space launch information.

Images