Spectrum processing method, apparatus and computer-readable storage medium

Abstract

This disclosure provides a graph processing method, apparatus, and computer-readable storage medium, relating to the field of artificial intelligence technology. The method includes: obtaining a first word vector for each first word in a first knowledge graph, and a second word vector for each second word in at least one second word to be added to the first knowledge graph; determining a second association relationship between the at least one second word and each first word based on a graph deep learning model, an adjacency matrix, and an input feature matrix, wherein the adjacency matrix is ​​constructed based on the first association relationship between the first words in the first knowledge graph, and the input feature matrix is ​​constructed based on the first word vector and the second word vector; and adding the at least one second word to the first knowledge graph based on the second association relationship to obtain a second knowledge graph.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a graph processing method, apparatus, and computer-readable storage medium. Background Technology

[0002] A knowledge graph is a semantic network that reveals the relationships between entities, aiming to describe various entities existing in the real world and their interrelationships. A knowledge graph can also be viewed as a huge graph, where nodes represent entities and edges represent relationships between entities. Knowledge graphs provide the ability to analyze problems from the perspective of "relationships." Presenting updated knowledge from various domains in the form of knowledge graphs facilitates access to relevant information within those domains.

[0003] In related technologies, for each node to be added to the knowledge graph, the node information of the node to be added and the existing nodes in the knowledge graph are usually used for association analysis. When it is determined that there is a relationship between the two, the node to be added is added to the knowledge graph according to the relationship, so as to complete the update of the knowledge graph. Summary of the Invention

[0004] The inventors noted that in the related technologies, the association analysis is performed using only the node information of the node to be added and the existing node, which leads to inaccurate determination of the association between the node to be added and the existing node, resulting in an inaccurate updated knowledge graph.

[0005] To address the aforementioned problems, the present disclosure proposes the following solutions.

[0006] According to one aspect of the present disclosure, a graph processing method is provided, comprising: obtaining a first word vector for each first word in a first knowledge graph, and a second word vector for each second word in at least one second word to be added to the first knowledge graph; determining a second association relationship between the at least one second word and each first word based on a graph deep learning model, an adjacency matrix, and an input feature matrix, wherein the adjacency matrix is ​​constructed based on the first association relationship between the first words in the first knowledge graph, and the input feature matrix is ​​constructed based on the first word vector and the second word vector; and adding the at least one second word to the first knowledge graph based on the second association relationship to obtain a second knowledge graph.

[0007] In some embodiments, the method further includes: segmenting each word in at least one set of words to obtain multiple third words, wherein the at least one set of words includes at least one set of words from a first set of words and a second set of words, wherein the first set of words includes each first word in the first knowledge graph, and the second set of words includes the at least one second word; obtaining a third word vector for each of the multiple third words; wherein the adjacency matrix is ​​constructed based on the first association relationship and the membership relationship between the at least one set of words and the multiple third words, and the input feature matrix is ​​constructed based on the first word vector, the second word vector, and the third word vector.

[0008] In some embodiments, the method further includes: obtaining training text independent of the first knowledge graph, wherein the training text belongs to the same domain as the first knowledge graph and includes at least some of the third words among the plurality of third words; segmenting the training text according to a dictionary to obtain segmented training text, wherein the dictionary includes the plurality of third words; wherein obtaining the third word vector includes: obtaining the third word vector through a word vector model trained from the segmented training text.

[0009] In some embodiments, the dictionary further includes the first group of words, the training text includes at least a portion of the first words from the first group of words, and the first word vector is obtained through the word vector model.

[0010] In some embodiments, the dictionary further includes the second group of words, the training text includes at least a portion of the second words in the second group of words, and the second word vector is obtained through the word vector model.

[0011] In some embodiments, the at least one set of words includes the first set of words.

[0012] In some embodiments, the at least one set of words includes the second set of words.

[0013] In some embodiments, the at least one set of words includes the first set of words and the second set of words.

[0014] In some embodiments, the graph deep learning model includes a graph convolutional neural network (GCN) model as an encoder and a graph autoencoder (GAE) model as a decoder.

[0015] In some embodiments, the first knowledge graph is a domain knowledge graph in the field of communications.

[0016] According to another aspect of the present disclosure, a graph processing apparatus is provided, comprising: an acquisition module configured to acquire a first word vector of each first word in a first knowledge graph, and a second word vector of each second word in at least one second word to be added to the first knowledge graph; a determination module configured to determine a second association relationship between the at least one second word and each first word based on a graph deep learning model, an adjacency matrix, and an input feature matrix, wherein the adjacency matrix is ​​constructed based on the first association relationship between the first words in the first knowledge graph, and the input feature matrix is ​​constructed based on the first word vector and the second word vector; and an addition module configured to add the at least one second word to the first knowledge graph based on the second association relationship to obtain a second knowledge graph.

[0017] According to another aspect of the present disclosure, a map processing apparatus is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute the method described in any of the above embodiments based on instructions stored in the memory.

[0018] According to another aspect of the present disclosure, a computer-readable storage medium is provided, including computer program instructions, wherein the computer program instructions, when executed by a processor, implement the method described in any of the above embodiments.

[0019] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein the computer program, when executed by a processor, implements the method described in any of the above embodiments.

[0020] In this embodiment, an adjacency matrix constructed based on the first association relationship between first words in a first knowledge graph, and an input feature matrix constructed based on the first word vector of each first word and the second word vector of each of at least one second word to be added to the first knowledge graph, are input into a graph deep learning model to determine the second association relationship between the at least one second word and each first word in the first knowledge graph. Then, based on the second association relationship, at least one second word is added to the first knowledge graph to obtain a second knowledge graph. Thus, in the process of determining the second association relationship between the at least one second word to be added to the first knowledge graph and the existing first words in the first knowledge graph, not only is the information of each second word and each first word as nodes (i.e., node information) utilized, but also the information of the first association relationship as edges in the first knowledge graph (i.e., edge information), improving the accuracy of the determined second association relationship between the node to be added and the existing nodes, thereby improving the accuracy of the updated second knowledge graph.

[0021] The technical solutions of this disclosure will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic flowchart of a spectrum processing method according to some embodiments of the present disclosure;

[0024] Figure 2 This is a schematic flowchart of a spectrum processing method according to other embodiments of this disclosure;

[0025] Figure 3 This is a schematic diagram of heterogeneous diagrams according to some embodiments of the present disclosure;

[0026] Figure 4 This is a schematic flowchart of a spectrum processing method according to some embodiments of the present disclosure;

[0027] Figure 5 This is a schematic diagram of the structure of a spectrum processing apparatus according to some embodiments of the present disclosure;

[0028] Figure 6 This is a schematic diagram of the structure of a spectrum processing apparatus according to other embodiments of the present disclosure. Detailed Implementation

[0029] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0030] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this disclosure.

[0031] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0032] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0033] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0034] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0035] Figure 1 This is a schematic flowchart of a spectrum processing method according to some embodiments of the present disclosure.

[0036] In step 102, the first word vector of each first word in the first knowledge graph is obtained, as well as the second word vector of each second word in at least one second word to be added to the first knowledge graph.

[0037] In some embodiments, the first word vector of each first word and the second word vector of each second word can be obtained based on the one-hot encoding algorithm. In this way, each first word vector and each second word vector are one-hot vectors.

[0038] In some embodiments, each first word vector and each second word vector can be obtained using a word vector model trained on each first word and each second word (e.g., Word2vec or GloVe model). In other embodiments, each first word vector and each second word vector can also be obtained using a word vector model trained in other ways, which will be further explained later.

[0039] In some embodiments, the first knowledge graph can be a domain knowledge graph in the field of communications, such as a knowledge graph of call detail records (CDRs) from an operator. For example, both the first term and the second term can be metadata from a communications domain database, wherein different metadata can come from the same data table or different data tables, and the different data tables can be different data tables in the same database or different data tables in different databases.

[0040] In step 104, based on the graph deep learning model, the adjacency matrix, and the input feature matrix, a second association relationship between at least one second word and each first word is determined.

[0041] Here, the adjacency matrix is ​​constructed based on the first association between the first words in the first knowledge graph, and the input feature matrix is ​​constructed based on the first word vector of each first word and the second word vector of each second word. It should be understood that each word can be considered a node in the knowledge graph, and the association between words can be considered an edge in the knowledge graph.

[0042] In some embodiments, a graph deep learning model may include a graph convolutional network (GCN) model as an encoder and a graph autoencoder (GAE) model as a decoder, wherein the GCN model as an encoder can generate feature vectors for each node, and the GAE model as a decoder can generate the probability that there are edges between nodes based on the feature vectors of each node.

[0043] As one implementation, a graph deep learning model can be constructed by connecting a two-layer GCN model, which acts as an encoder, to a GAE model, which acts as a decoder.

[0044] For example, the forward propagation calculation formula for a two-layer GCN model can be:

[0045]

[0046] in, A is the adjacency matrix, X is the input feature matrix, D is the degree matrix of matrix A, σ is the nonlinear activation function, and W... (0) and W (1) These are all weight matrices used for linear transformations.

[0047] For example, the calculation formula for the GAE model can be:

[0048]

[0049] The loss function corresponding to the graph deep learning model constructed in this way can be:

[0050]

[0051] Where 1≤m≤M, M is the total number of nodes in the knowledge graph, and σ is the nonlinear activation function.

[0052] It should be understood that the degree matrix D is a diagonal matrix, and each element on its diagonal represents the number of edges associated with each node. v can be any of the nonlinear activation functions such as softmax, ReLU, and tanh. For example, the above formula (1) can be:

[0053] In this implementation, the adjacency matrix A, constructed based on the first association relationship between the first words in the first knowledge graph, and the input feature matrix X, constructed based on the first word vector of each first word and the second word vector of each second word, can be input into the graph deep learning model for learning and training. When the loss value of the loss function L reaches a preset threshold, the output matrix is... Based on matrix This allows us to determine a second association between at least one second word and each first word.

[0054] In step 106, at least one second word is added to the first knowledge graph based on the second association relationship to obtain the second knowledge graph.

[0055] For example, a second knowledge graph can be obtained by creating an edge between a second word and a first word that have a second association relationship, thereby adding the second word to the first knowledge graph.

[0056] In the above embodiments, an adjacency matrix constructed based on the first association relationship between first words in the first knowledge graph, and an input feature matrix constructed based on the first word vector of each first word and the second word vector of each second word in at least one second word to be added to the first knowledge graph, are input into the graph deep learning model to determine the second association relationship between the at least one second word and each first word in the first knowledge graph. Then, based on the second association relationship, at least one second word is added to the first knowledge graph to obtain the second knowledge graph. Thus, in the process of determining the second association relationship between at least one second word to be added to the first knowledge graph and existing first words in the first knowledge graph, not only is the information of each second word and each first word as nodes (i.e., node information) utilized, but also the information of the first association relationship as edges in the first knowledge graph (i.e., edge information), improving the accuracy of the determined second association relationship between the node to be added and the existing nodes, thereby improving the accuracy of the updated second knowledge graph.

[0057] Figure 2 This is a schematic flowchart of a spectrum processing method according to other embodiments of the present disclosure.

[0058] In step 202, the first word vector of each first word in the first knowledge graph is obtained, as well as the second word vector of each second word in at least one second word to be added to the first knowledge graph.

[0059] In step 204, each word in at least one group of words is segmented to obtain multiple third words.

[0060] Here, at least one set of words includes at least one set of words from the first set of words and the second set of words. The first set of words includes each first word in the first knowledge graph, and the second set of words includes at least one second word.

[0061] In some embodiments, at least one set of words may include only the first set of words, or at least one set of words may include only the second set of words, or at least one set of words may include both the first set of words and the second set of words.

[0062] In some embodiments, a preset word segmentation algorithm can be used to segment at least one group of words to obtain multiple third words. The preset word segmentation algorithm may be, for example, a conditional random field algorithm (CRF) or a neural network word segmentation algorithm. Here, word segmentation processing includes preprocessing such as stop word removal. Stop words can include interjections, punctuation marks, function words, and other words without actual semantic meaning. For example, segmenting "mobile voice behavior" from at least one group of words can yield multiple third words including "mobile," "voice," and "behavior"; as another example, segmenting "terminal product label" from at least one group of words can yield multiple third words including "terminal," "product," and "label."

[0063] In step 206, the third word vector of each of the multiple third words is obtained.

[0064] In some embodiments, the third word vector of each third word can be obtained by a word vector model trained from multiple third words obtained after segmenting at least one set of words. In these embodiments, the first word vector of each first word and the second word vector of each second word can also be obtained by the word vector model.

[0065] In step 208, based on the graph deep learning model, the adjacency matrix, and the input feature matrix, a second association relationship between at least one second word and each first word is determined.

[0066] Here, the adjacency matrix is ​​constructed based on the first association relationship and the membership relationship between at least one set of words and multiple third words, and the input feature matrix is ​​constructed based on the first word vector of each first word, the second word vector of each second word, and the third word vector of each third word.

[0067] In step 210, at least one second word is added to the first knowledge graph based on the second association relationship to obtain the second knowledge graph.

[0068] The specific implementation methods of steps 202, 208 and 210 are similar to those of steps 102 to 106 described above. For detailed descriptions, please refer to the relevant embodiments in steps 102 to 106 described above, which will not be repeated here.

[0069] In the above embodiments, in determining the second association between at least one second word to be added to the first knowledge graph and existing first words in the first knowledge graph, the node information used includes not only information about each second word and each first word, but also information about each third word. The edge information used includes not only information about the first association, but also information about the membership relationships between at least one group of words and multiple third words. That is, by fully utilizing the implicit node information and edge information in the newly added nodes and / or known nodes to determine the second association between the node to be added and existing nodes, the accuracy of the determined second association is further improved, thereby further improving the accuracy of the updated second knowledge graph.

[0070] In some embodiments, at least one set of words may include a first set of words. In these embodiments, each third word obtained by segmenting each first word in the first knowledge graph can be used as a node in the graph, thereby expanding the original first knowledge graph into a heterogeneous graph to reflect the first association relationship between the first words in the first knowledge graph and the membership relationship between each first word and multiple third words.

[0071] To facilitate understanding of the above relationships and hierarchical relationships, the following will combine... Figure 3 Further explanation is needed.

[0072] Figure 3 This is a schematic diagram of heterogeneous diagrams according to some embodiments of the present disclosure.

[0073] like Figure 3 As shown, the heterogeneous graph contains 6 nodes and 6 edges. The 6 nodes are the first terms "4GLTE call detail records" and "5G call detail records", and the multiple third terms "4G", "LTE", "call detail records", and "5G" obtained after segmenting these two first terms. Among the 6 edges, the edge between "4GLTE call detail records" and "5G call detail records" represents the first association relationship between these two first terms; the 3 edges between "4G", "LTE", and "call detail records" and "4GLTE call detail records" represent the membership relationship between these 3 third terms and the first term "4GLTE call detail records"; and the 2 edges between "5G", "call detail records", and "5G call detail records" represent the membership relationship between these 2 third terms and the first term "5G call detail records".

[0074] In the above embodiments, the adjacency matrix of the input graph deep learning model is constructed based on the first association relationship between first words and the membership relationship between each first word and multiple third words. The input feature matrix is ​​constructed based on the first word vector of each first word, the second word vector of each second word, and the third word vector of each of the multiple third words obtained after segmenting each first word. Thus, in determining the second association relationship, the known information (including node information and edge information) implicit in the first knowledge graph (which is a known graph) is utilized more fully. That is, to update the first knowledge graph, relevant information in the original first knowledge graph is used more fully to determine the second association relationship, thereby further improving the accuracy of the updated second knowledge graph.

[0075] In other embodiments, at least one set of words may include a first set of words and a second set of words. In these embodiments, the adjacency matrix of the input graph deep learning model is constructed based on the first association between the first words, the membership relationship between each first word and multiple third words, and the membership relationship between each second word and multiple third words. The input feature matrix is ​​constructed based on the first word vector of each first word, the second word vector of each second word, and the third word vector of each of the multiple third words obtained after segmenting each first word and each second word.

[0076] In this way, in the process of determining the second association, not only is the node information and edge information implicit in the newly added nodes made fuller use, but the node information and edge information implicit in the known nodes are also made fuller use, which further improves the accuracy of the determined second association, thereby further improving the accuracy of the updated second knowledge graph.

[0077] However, the inventors also noted that whether word vectors are obtained by using word vector models trained from each first word and at least one second word in the first knowledge graph, or by using word vector models trained from multiple third words obtained after segmenting each first word and / or at least one second word, the word vector models trained are inaccurate due to the lack of semantic information of each word in its corresponding domain during the training process. Consequently, the word vectors obtained through these word vector models are inaccurate, which in turn leads to the inaccuracy of the subsequently determined second association relationship and the updated second knowledge graph.

[0078] Based on this, this application also proposes the following solutions.

[0079] Figure 4 This is a schematic flowchart of a spectrum processing method according to some embodiments of the present disclosure.

[0080] and Figure 2Compared to the embodiments shown, Figure 4 The illustrated spectrum processing method further includes steps 402 and 404.

[0081] In step 402, training text independent of the first knowledge graph is obtained.

[0082] Here, the training text belongs to the same domain as the first knowledge graph and includes at least some of the third words from multiple third words.

[0083] In some embodiments, the training text may be in the same domain as the first knowledge graph and include each of a plurality of third words. For example, if the first knowledge graph is in the field of communications, the training text may include relevant texts such as communications case studies, communications news, and communications papers, wherein these texts include each of a plurality of third words.

[0084] In step 404, the training text is segmented according to the dictionary to obtain the segmented training text. Here, the dictionary includes multiple third-party words.

[0085] It should be understood that at least a portion of the training text after word segmentation is identical to a portion of the third word in the dictionary.

[0086] As some implementation methods of step 404, for third words that exist in both the dictionary and the training text, the dictionary can be used directly to segment the text containing these third words; for other words that do not exist in the dictionary but exist in the training text, word segmentation tools such as jieba can be used to segment the text containing these other words; for third words that exist in the dictionary but not in the training text, these third words can be treated as out-of-vocabulary words and randomly initialized. In this way, the segmented training text can be obtained.

[0087] exist Figure 4 In the illustrated embodiment, step 206 can be implemented as follows:

[0088] The third word vector of each third word is obtained by training a word vector model from the segmented training text.

[0089] In the above embodiments, training text containing at least some of the third words from a plurality of third words, belonging to the same domain as the first knowledge graph, is introduced. A word vector model is trained on the segmented training text obtained by segmenting the text using a dictionary containing the third words, so that each third word vector can be obtained through the word vector model trained in this way. Thus, since the segmented training text includes not only multiple third words but also the semantic information of each third word in its corresponding domain, the word vector model trained in this way is more accurate. Consequently, the third word vectors obtained through the word vector model trained in this way are also more accurate, further improving the accuracy of the subsequently determined second association relationship, and thus further improving the accuracy of the updated second knowledge graph.

[0090] In some embodiments, the dictionary may further include a first set of words, and the training text may further include at least a portion of the first words from the first set of words. In these embodiments, the first word vector of each first word can be obtained by a word vector model trained from the segmented training text. This improves the accuracy of the obtained first word vectors, thereby further improving the accuracy of the updated second knowledge graph.

[0091] In other embodiments, where at least one set of words includes the first set of words, the dictionary may also include the first set of words, and the training text may also include at least some of the first words from the first set of words. In this case, the first word vector of each first word can be obtained through a word vector model trained from the segmented training text. Thus, for updating the first knowledge graph, not only is more relevant information in the original first knowledge graph utilized to further improve the accuracy of the determined second association, but the accuracy of the obtained first word vectors is also further improved, thereby further improving the accuracy of the updated second knowledge graph.

[0092] In some embodiments, the dictionary may further include a second set of words, and the training text may further include at least some of the second words from the second set of words. In these embodiments, the second word vector of each second word can be obtained by a word vector model trained from the segmented training text. This improves the accuracy of the obtained second word vectors, thereby further improving the accuracy of the updated second knowledge graph.

[0093] In other embodiments, where at least one set of words includes a second set of words, the dictionary may also include the second set of words, and the training text may also include at least some of the second words from the second set of words. In this case, the second word vector for each second word can be obtained through a word vector model trained from the segmented training text. This further improves the accuracy of the updated second knowledge graph.

[0094] In some embodiments, the dictionary may further include a first group of words and a second group of words, and the training text may further include at least some of the first words from the first group of words and at least some of the second words from the second group of words. In these embodiments, the first word vector of each first word and the second word vector of each second word can be obtained by a word vector model trained from the segmented training text. This improves the accuracy of the obtained first and second word vectors, thereby further improving the accuracy of the updated second knowledge graph.

[0095] In other embodiments, where at least one set of words includes a first set of words and a second set of words, the dictionary may also include both the first set of words and the second set of words. The training text may also include at least some of the first words from the first set of words and at least some of the second words from the second set of words. In this case, the first word vector for each first word and the second word vector for each second word can be obtained through a word vector model trained from the segmented training text. Thus, for updating the first knowledge graph, not only is the accuracy of the determined second association further improved, but the accuracy of the obtained first and second word vectors is also further improved, thereby further improving the accuracy of the updated second knowledge graph.

[0096] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus embodiments, since they largely correspond to the method embodiments, the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0097] Figure 5 This is a schematic diagram of the structure of a spectrum processing apparatus according to some embodiments of the present disclosure.

[0098] like Figure 5 As shown, the spectrum processing device 500 includes an acquisition module 501, a determination module 502, and an addition module 503.

[0099] The acquisition module 501 can be configured to acquire the first word vector of each first word in the first knowledge graph, and the second word vector of each second word in at least one second word to be added to the first knowledge graph.

[0100] The determination module 502 can be configured to determine a second association between at least one second word and each first word based on a graph deep learning model, an adjacency matrix, and an input feature matrix, wherein the adjacency matrix is ​​constructed based on the first association between first words in a first knowledge graph, and the input feature matrix is ​​constructed based on the first word vector and the second word vector.

[0101] The newly added module 503 can be configured to add at least one second word to the first knowledge graph based on the second association relationship, so as to obtain the second knowledge graph.

[0102] In some embodiments, the map processing apparatus 500 may also include other modules that perform other operations in any of the above embodiments.

[0103] Figure 6 This is a schematic diagram of the structure of a spectrum processing apparatus according to other embodiments of the present disclosure.

[0104] like Figure 6 As shown, the map processing apparatus 600 includes a memory 601 and a processor 602 coupled to the memory 601. The processor 602 is configured to execute the method of any of the foregoing embodiments based on instructions stored in the memory 601.

[0105] The memory 601 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory may store, for example, an operating system, application programs, a boot loader, and other programs.

[0106] The image processing device 600 may also include an input / output interface 603, a network interface 604, and a storage interface 605. These interfaces 603, 604, and 605, as well as the memory 601 and processor 602, can be connected, for example, via a bus 606. The input / output interface 603 provides a connection interface for input / output devices such as displays, mice, keyboards, and touchscreens. The network interface 604 provides a connection interface for various networked devices. The storage interface 605 provides a connection interface for external storage devices such as SD cards and USB flash drives.

[0107] This disclosure also provides a computer-readable storage medium including computer program instructions that, when executed by a processor, implement the method of any of the above embodiments.

[0108] This disclosure also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, it implements the method of any of the above embodiments.

[0109] The embodiments of this disclosure have now been described in detail. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.

[0110] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0111] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that the functions specified in one or more flowchart illustrations and / or one or more blocks in a block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate functions for implementing the functions in the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0112] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0113] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0114] While specific embodiments of this disclosure have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments or equivalent substitutions can be made to some technical features without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A spectrum processing method, comprising: Obtain the first word vector of each first word in the first knowledge graph, and the second word vector of each second word in at least one second word to be added to the first knowledge graph; Each word in at least one group of words is segmented to obtain multiple third words. The at least one group of words includes at least one group of words from a first group of words and a second group of words. The first group of words includes each first word in the first knowledge graph, and the second group of words includes the at least one second word. Obtain the third word vector for each of the plurality of third words; Based on a graph deep learning model, an adjacency matrix, and an input feature matrix, a second association relationship between the at least one second word and each first word is determined. The adjacency matrix is ​​constructed based on the first association relationship between first words in the first knowledge graph and the membership relationship between the at least one group of words and the plurality of third words. The input feature matrix is ​​constructed based on the first word vector, the second word vector, and the third word vector. Based on the second association, at least one second word is added to the first knowledge graph to obtain a second knowledge graph.

2. The method according to claim 1, further comprising: Obtain training text independent of the first knowledge graph, wherein the training text belongs to the same domain as the first knowledge graph and includes at least some of the third words among the plurality of third words; The training text is segmented according to a dictionary to obtain segmented training text, wherein the dictionary includes the plurality of third words; Obtaining the third word vector includes: The third word vector is obtained by training a word vector model from the segmented training text.

3. The method according to claim 2, wherein, The dictionary also includes the first group of words, the training text includes at least some of the first words in the first group of words, and the first word vector is obtained through the word vector model.

4. The method according to claim 2 or 3, wherein, The dictionary also includes the second group of words, the training text includes at least some of the second words in the second group of words, and the second word vector is obtained through the word vector model.

5. The method according to claim 1 or 3, wherein, The at least one group of words includes the first group of words.

6. The method according to claim 4, wherein, The at least one set of words includes the second set of words.

7. The method according to claim 1 or 2, wherein, The at least one set of words includes the first set of words and the second set of words.

8. The method according to claim 1, wherein, The graph deep learning model includes a graph convolutional neural network (GCN) model as an encoder and a graph autoencoder (GAE) model as a decoder.

9. A spectrum processing apparatus, comprising: The acquisition module is configured to acquire the first word vector of each first word in the first knowledge graph, and the second word vector of each second word in at least one second word to be added to the first knowledge graph; The determination module is configured to determine a second association between the at least one second word and each first word based on a graph deep learning model, an adjacency matrix, and an input feature matrix. The adjacency matrix is ​​constructed based on the first association between first words in the first knowledge graph and the membership relationships between at least one set of words and multiple third words. The input feature matrix is ​​constructed based on the first word vector, the second word vector, and the third word vector of each of the multiple third words. The at least one set of words includes at least one set from a first set of words and a second set of words. The first set of words includes each first word in the first knowledge graph, and the second set of words includes the at least one second word. The new module is configured to add at least one second word to the first knowledge graph based on the second association relationship, so as to obtain a second knowledge graph. The graph processing device is configured to segment each word in the at least one set of words to obtain the plurality of third words, and to obtain the third word vector of each of the plurality of third words.

10. A spectrum processing apparatus, comprising: Memory; as well as A processor coupled to the memory is configured to execute the method of any one of claims 1-8 based on instructions stored in the memory.

11. A computer-readable storage medium comprising computer program instructions, wherein, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1-8.

12. A computer program product comprising a computer program, wherein, When the computer program is executed by a processor, it implements the method described in any one of claims 1-8.

Images