Global knowledge graph management method and system

Abstract

The present invention relates to the technical field of text feature extraction, and specifically discloses a global knowledge graph management method and system, the method comprising receiving text containing author parameters uploaded by an uploader, and constructing a text library; determining the application probability of each text according to the author parameters, randomly selecting text in the text library based on the application probability, and creating at least one text set; performing topic identification on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic; creating a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic; the present invention provides a screening and visualization solution for text data, when retrieving data, performing text feature extraction on the text data, and constructing a knowledge graph, for the searcher, the data has a high degree of visualization, extremely high readability, and an extremely easy data selection process.

Description

Technical Field

[0001] The present invention relates to the technical field of text feature extraction, and specifically to a global knowledge graph management method and system. Background Art

[0002] With the development of network technology and the popularization of smart devices, there are more and more means to obtain information. Everyone can become a creator and share their knowledge with others. This means that for any field, there is a lot of relevant knowledge content. However, since creators will not use the same template to share information, the information on the Internet is extremely messy. When searching, the screening process is very difficult, difficult to read, and the selection process is very difficult. This is the technical problem that the technical solution of the present invention wants to solve. Summary of the invention

[0003] The purpose of the present invention is to provide a global knowledge graph management method and system to solve the problems raised in the above-mentioned background technology.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A global knowledge graph management method, the method comprising:

[0006] Receive the text containing the author parameters uploaded by the uploader and build a text library;

[0007] Determine the application probability of each text according to the author parameter, randomly select text from the text library based on the application probability, and create at least one text set;

[0008] Performing topic identification on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic;

[0009] Create a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic;

[0010] Among them, the step of creating a knowledge graph includes reading the topic distribution of each text set and the word distribution corresponding to its topic, constructing a knowledge graph corresponding to each text set; merging the knowledge graphs corresponding to all text sets to obtain the final knowledge graph.

[0011] Furthermore, the step of receiving the text containing the author parameters uploaded by the uploader and constructing the text library includes:

[0012] Receive the domain index input by the administrator, retrieve the text based on the domain index, and synchronously obtain the author parameters;

[0013] Inputting the author parameters into the trained evaluation model to determine the utility value of the text; the utility value is used to determine the application probability of the text;

[0014] Statistics texts and their utility values, and build a text library.

[0015] Furthermore, the step of performing topic identification on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic includes:

[0016] The number of topics entered by the receiving administrator;

[0017] For each text, extract a topic distribution from a Dirichlet distribution based on the number of topics;

[0018] For each topic, draw a word distribution from a Dirichlet distribution;

[0019] For each text, iteratively update the topic distribution and the word distribution based on Gibbs sampling;

[0020] After the iterative update is completed, the final topic distribution and final word distribution for each text are output.

[0021] Furthermore, the iterative update process is:

[0022] Generate document content based on the extracted topic distribution and word distribution of each topic;

[0023] For each word in the document content , calculate the conditional probability that the current word belongs to each topic, and reallocate the topic of the current word according to the calculated conditional probability;

[0024] The loop is executed until the preset number of loops is reached;

[0025] The document content generation process is as follows:

[0026] Distribution by topic Extract a topic from , from the selected topics Word distribution Extract a word from ;

[0027] The calculation process of conditional probability is:

[0028] ;

[0029] In the formula, represents the conditional probability, Representation Document Middle The topic assignment of words, Indicates to remove the document Middle The distribution of all topics except words, The actual vocabulary representing all the words; and are preset values, all of which are 0.1;

[0030] For Documentation Theme The number of words in , excluding the current word; For Documentation The total number of words in , excluding the current word, Is the theme Chinese words The number of occurrences, excluding the current word, Is the theme The total number of words in , excluding the current word; is the dimension of each element in the topic distribution; is the dimension of each element in the word distribution.

[0031] Furthermore, the step of creating a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic includes:

[0032] Read the word distribution of each topic in each text set, and create a circular node corresponding to each word; the name of the circular node is the word, and the radius of the circular node is proportional to the probability of the word under the topic;

[0033] The circular nodes are clustered according to the topics to which each word belongs, and each type of circular node is scaled proportionally according to the probability of the topic. After all the circular nodes are scaled, the knowledge graph corresponding to the text set is obtained; the scaling ratio is proportional to the probability of the topic;

[0034] Merge the knowledge graphs corresponding to all text sets to obtain the final knowledge graph.

[0035] Furthermore, the step of merging the knowledge graphs corresponding to all text sets to obtain the final knowledge graph includes:

[0036] Read the knowledge graph corresponding to each text set and merge them;

[0037] During the merge process:

[0038] For each circular node, query the corresponding text set and the topic corresponding to the circular node in the text set;

[0039] When the same circular node corresponds to multiple text sets and the corresponding topics are the same, the circular node is scaled twice according to the number of topics;

[0040] When the same circular node corresponds to multiple text sets and the corresponding topics are different, the circular nodes corresponding to each topic are retained and connected;

[0041] When a circular node corresponds to a text set, the circular node corresponding to each topic is retained.

[0042] The technical solution of the present invention also provides a global knowledge graph management system, the system comprising:

[0043] The text library construction module is used to receive the text containing the author parameters uploaded by the uploader and construct the text library;

[0044] A text set creation module, used to determine the application probability of each text according to the author parameter, randomly select texts from the text library based on the application probability, and create at least one text set;

[0045] A text recognition module is used to perform topic recognition on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic;

[0046] A knowledge graph creation module is used to create a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic;

[0047] Among them, the step of creating a knowledge graph includes reading the topic distribution of each text set and the word distribution corresponding to its topic, constructing a knowledge graph corresponding to each text set; merging the knowledge graphs corresponding to all text sets to obtain the final knowledge graph.

[0048] Furthermore, the text library construction module includes:

[0049] A text retrieval unit, used to receive a domain index input by the management party, retrieve text based on the domain index, and synchronously obtain author parameters;

[0050] A text evaluation unit, which is used to input the author parameters into the trained evaluation model to determine the utility value of the text; the utility value is used to determine the application probability of the text;

[0051] The text statistics unit is used to count texts and their utility values ​​and build a text library.

[0052] Furthermore, the text recognition module includes:

[0053] A quantity receiving unit, used to receive the quantity of topics input by the management party;

[0054] A topic distribution extraction unit, configured to extract a topic distribution from a Dirichlet distribution based on the number of topics for each text;

[0055] A word distribution extraction unit, used for extracting word distribution from Dirichlet distribution for each topic;

[0056] An iterative updating unit, used for iteratively updating the topic distribution and the word distribution for each text based on Gibbs sampling;

[0057] The result output unit is used to output the final topic distribution and final word distribution for each text after the iterative update is completed.

[0058] Furthermore, the knowledge graph creation module includes:

[0059] A node creation unit is used to read the word distribution of each topic in each text set and create a circular node corresponding to each word; the name of the circular node is the word, and the radius of the circular node is proportional to the probability of the word under the topic;

[0060] The node clustering unit is used to cluster the circular nodes according to the topics to which each word belongs, and to scale each type of circular node geometrically according to the probability of the topic. After all the circular nodes are scaled, the knowledge graph corresponding to the text set is obtained; the scaling ratio is proportional to the probability of the topic;

[0061] The merging unit is used to merge the knowledge graphs corresponding to all text sets to obtain the final knowledge graph.

[0062] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a screening and visualization solution for text data. When retrieving data, text features are extracted from the text data and a knowledge graph is constructed. For the searcher, the data is highly visualized and highly readable, and the data selection process is extremely easy. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0064] Figure 1 This is a flowchart of the global knowledge graph management method.

[0065] Figure 2 This is the structural block diagram of the global knowledge graph management system. DETAILED DESCRIPTION

[0066] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0067] Figure 1The flowchart of the global knowledge graph management method is as follows. In an embodiment of the present invention, a global knowledge graph management method is provided, and the method includes:

[0068] Step S100: receiving the text containing the author parameters uploaded by the uploader and constructing a text library;

[0069] The uploader refers to the party that uploads the text, including individual creators or creators from units such as schools and enterprises. The uploaded text refers to the text content related to a certain field. These texts are counted to obtain a text library. The process of step S100 in actual application is that the administrator pre-sets a field, and then the administrator consults the information, obtains the text containing the author parameters, and obtains the text library. The content of the obtained text is not limited. Whether it is a paper on a professional website or a personal sharing on some forums, it is a text that can be obtained by this application.

[0070] Step S200: determining the application probability of each text according to the author parameter, randomly selecting texts from the text library based on the application probability, and creating at least one text set;

[0071] The author parameters are analyzed, and the author parameters are used to characterize the author's professionalism in the field. The application probability of each text is determined according to the author parameters, and then texts are randomly selected from the text library based on the application probability to obtain a text set; the text set is a subset of the text library, which is equivalent to a simplification of the text library. Since the text selection process is a random selection process based on the application probability, different text sets are obtained with different selection times.

[0072] Step S300: performing topic identification on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic;

[0073] By performing topic analysis on a text set, we can obtain a topic distribution, that is, the distribution of each topic in the text set. A text can include multiple topics; for each topic, we can obtain a word distribution; this process is a specific application of the LDA algorithm, which is a topic model algorithm for text analysis.

[0074] Step S400: Creating a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic;

[0075] Read the topic distribution of each text set and the word distribution corresponding to its topic, and construct the knowledge graph corresponding to each text set. This process is actually a two-dimensional display process of the topic distribution and the word distribution corresponding to its topic. The topic distribution and word distribution are represented by a node-based graph structure to obtain a knowledge graph; wherein each text set corresponds to a knowledge graph, and step S200 may generate more than one knowledge graph. The knowledge graphs corresponding to all text sets are merged to obtain the final knowledge graph.

[0076] Regarding step S100, the step of receiving the text containing the author parameters uploaded by the uploader and constructing the text library includes:

[0077] Receive the domain index input by the administrator, retrieve the text based on the domain index, and synchronously obtain the author parameters;

[0078] Inputting the author parameters into the trained evaluation model to determine the utility value of the text; the utility value is used to determine the application probability of the text;

[0079] Statistics texts and their utility values, and build a text library.

[0080] The author parameters include the total number of views, total number of likes and total number of collections of all the author's texts, etc., and the author's academic qualifications can also be introduced; the management party first determines a field index (such as the field name), retrieves the text based on the field index, and obtains the author parameters when the text is retrieved. If the author parameters cannot be obtained, the default parameters are used as the author parameters (which can be compared to anonymous). The default parameters generally use the author parameters of the new author; after obtaining the author parameters, the author parameters are input into the trained evaluation model to obtain the utility value of each text. The utility value is used to represent the author's ability. Generally, the stronger the author's ability reflected by the author parameters, the higher the utility value and the greater the application probability of the corresponding text.

[0081] About the evaluation model:

[0082] The evaluation model of this application is actually a mapping model from author parameters to utility values. The simplest one is to use a power series summation function, that is, ;in, The terms are constant terms, and each term represents an author parameter. The exponents and coefficients are determined in advance by the staff, and then they are accumulated to obtain a function from the author parameter (multi-dimensional) to the utility value (one-dimensional), which is called an evaluation model.

[0083] In addition, there is a more complex solution that does not require manual parameter setting: count multiple author parameters, and have the staff pre-calibrate the utility value corresponding to each author parameter to obtain a sample set. The characteristics of the samples in the sample set are the author parameters, and the labels are the utility values. The neural network model is trained in this way. When the error is small enough, the obtained neural network model can be used as an evaluation model.

[0084] Regarding step S300, the step of performing topic identification on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic includes:

[0085] The number of topics entered by the receiving administrator;

[0086] For each text, extract a topic distribution from a Dirichlet distribution based on the number of topics;

[0087] For each topic, draw a word distribution from a Dirichlet distribution;

[0088] For each text, iteratively update the topic distribution and the word distribution based on Gibbs sampling;

[0089] After the iterative update is completed, the final topic distribution and final word distribution for each text are output.

[0090] The number of topics is pre-entered by the administrator, for example, ten; for each document in the text library, a topic distribution is extracted from the Dirichlet distribution , , is a dimensional vector, is the number of topics; for each topic , from the Dirichlet distribution Extract a word distribution from , , is a dimensional vector, representing the probability of a word under the topic.

[0091] The topic distribution and word distribution obtained by the above extraction are a baseline model. For each text, the topic distribution and the word distribution are iteratively updated based on Gibbs sampling, and the baseline model is continuously updated to obtain the final topic distribution in each document and the word distribution under each topic.

[0092] It should be noted that Dirichlet Distribution is a continuous multivariate probability distribution and a commonly used prior distribution in Bayesian statistics. It is particularly suitable for modeling multiple distribution parameters. The main feature of Dirichlet Distribution is that it can describe the probability distribution of multiple random variables, and the sum of these random variables is 1, which makes it particularly suitable for modeling scenarios such as probability vectors and weight vectors.

[0093] The iterative update process is:

[0094] Generate document content based on the extracted topic distribution and word distribution of each topic;

[0095] For each word in the document content , calculate the conditional probability that the current word belongs to each topic, and reallocate the topic of the current word according to the calculated conditional probability;

[0096] The loop is executed until the preset number of loops is reached;

[0097] The document content generation process is as follows:

[0098] Distribution by topic Extract a topic from , from the selected topics Word distribution Extract a word from ;

[0099] The calculation process of conditional probability is:

[0100] ;

[0101] In the formula, represents the conditional probability, Representation Document Middle The topic assignment of words, Indicates to remove the document Middle The distribution of all topics except words, The actual vocabulary representing all the words; and are preset values, all of which are 0.1;

[0102] For Documentation Theme The number of words in , excluding the current word; For Documentation The total number of words in , excluding the current word, Is the theme Chinese words The number of occurrences, excluding the current word, Is the theme The total number of words in , excluding the current word; is the dimension of each element in the topic distribution; is the dimension of each element in the word distribution.

[0103] The above content is a specific application of the Gibbs sampling process, which approximates the true topic distribution of the document and the true word distribution of each topic by iteratively updating the topic assignment of each word.

[0104] Regarding step S400, the step of creating a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic includes:

[0105] Read the word distribution of each topic in each text set, and create a circular node corresponding to each word; the name of the circular node is the word, and the radius of the circular node is proportional to the probability of the word under the topic;

[0106] The circular nodes are clustered according to the topics to which each word belongs, and each type of circular node is scaled proportionally according to the probability of the topic. After all the circular nodes are scaled, the knowledge graph corresponding to the text set is obtained; the scaling ratio is proportional to the probability of the topic;

[0107] Merge the knowledge graphs corresponding to all text sets to obtain the final knowledge graph.

[0108] In an example of the technical solution of the present invention, the word distribution of each topic in each text set is read, and the word distribution represents the probability of occurrence of each word under the topic. Circular nodes corresponding to the words are created according to the probability. The greater the probability, the larger the radius of the circular node. Then, the circular nodes corresponding to all words of the same topic are classified into one category. On this basis, since the topic distribution also represents the probability of occurrence of each topic, each type of circular node is scaled in combination with the probability of the topic, and the scaling ratio is proportional to the probability. In actual situations, the average probability of the topic is calculated, and the corresponding scaling ratio is set to 100%. Based on this corresponding relationship, the scaling ratio of different probabilities can be determined. Generally, the scaling ratio will have a preset upper limit and a preset lower limit.

[0109] After creating and clustering the circular nodes, the knowledge graph corresponding to each text set is obtained. When the text set is not unique, the knowledge graphs corresponding to all text sets are merged to obtain the final knowledge graph as the global knowledge graph of the field selected by the management.

[0110] The step of merging the knowledge graphs corresponding to all text sets to obtain the final knowledge graph includes:

[0111] Read the knowledge graph corresponding to each text set and merge them;

[0112] During the merge process:

[0113] For each circular node, query the corresponding text set and the topic corresponding to the circular node in the text set;

[0114] When the same circular node corresponds to multiple text sets and the corresponding topics are the same, the circular node is scaled twice according to the number of topics;

[0115] When the same circular node corresponds to multiple text sets and the corresponding topics are different, the circular nodes corresponding to each topic are retained and connected;

[0116] When a circular node corresponds to a text set, the circular node corresponding to each topic is retained.

[0117] The merging process mainly includes the above three situations. Taking the circular node as the benchmark, for each circular node, query its corresponding text set and the topic corresponding to the circular node in the text set. If the same circular node corresponds to multiple text sets and the corresponding topics are the same, the corresponding circular node is scaled twice. The actual situation of this situation is that, assuming there are five text sets, if a word appears in three text sets, and two of the corresponding topics in the three text sets are the same (a word corresponds to only one topic in a text set), then the number of topics is two, and the circular node can be enlarged according to the number of topics.

[0118] If the same circular node corresponds to multiple text sets and the corresponding topics are different, then the circular nodes of all words corresponding to each topic are retained. After merging, the same circular node will appear in the node clusters corresponding to different topics. They are connected to indicate that they are the same word but correspond to different topics.

[0119] If the circular node corresponds to a text set, it means that only a certain topic in a certain text contains the word. In this case, just keep the merge.

[0120] It is worth mentioning that the scaling ratio in the secondary scaling process adopts a logarithmic function, such as: ; In the formula, is the scaling factor, is the number of topics, The preset correction factor, adding one to the number of topics means that no negative numbers are introduced during merging, which means that only amplification occurs during merging.

[0121] Figure 2: is a structural block diagram of a global knowledge graph management system. In an embodiment of the present invention, a global knowledge graph management system, the system 10 includes:

[0122] The text library construction module 11 is used to receive the text containing the author parameters uploaded by the uploader and construct the text library;

[0123] A text set creation module 12, configured to determine the application probability of each text according to the author parameter, randomly select texts from the text library based on the application probability, and create at least one text set;

[0124] A text recognition module 13 is used to perform topic recognition on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic;

[0125] A knowledge graph creation module 14 is used to create a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic;

[0126] Among them, the step of creating a knowledge graph includes reading the topic distribution of each text set and the word distribution corresponding to its topic, constructing a knowledge graph corresponding to each text set; merging the knowledge graphs corresponding to all text sets to obtain the final knowledge graph.

[0127] Furthermore, the text library construction module 11 includes:

[0128] A text retrieval unit, used to receive a domain index input by the management party, retrieve text based on the domain index, and synchronously obtain author parameters;

[0129] A text evaluation unit, which is used to input the author parameters into the trained evaluation model to determine the utility value of the text; the utility value is used to determine the application probability of the text;

[0130] The text statistics unit is used to count texts and their utility values ​​and build a text library.

[0131] The text recognition module 12 includes:

[0132] A quantity receiving unit, used to receive the quantity of topics input by the management party;

[0133] A topic distribution extraction unit, configured to extract a topic distribution from a Dirichlet distribution based on the number of topics for each text;

[0134] A word distribution extraction unit, used for extracting word distribution from Dirichlet distribution for each topic;

[0135] An iterative updating unit, used for iteratively updating the topic distribution and the word distribution for each text based on Gibbs sampling;

[0136] The result output unit is used to output the final topic distribution and final word distribution for each text after the iterative update is completed.

[0137] The knowledge graph creation module 14 includes:

[0138] A node creation unit is used to read the word distribution of each topic in each text set and create a circular node corresponding to each word; the name of the circular node is the word, and the radius of the circular node is proportional to the probability of the word under the topic;

[0139] The node clustering unit is used to cluster the circular nodes according to the topics to which each word belongs, and to scale each type of circular node proportionally according to the probability of the topic. After all the circular nodes are scaled, the knowledge graph corresponding to the text set is obtained; the scaling ratio is proportional to the probability of the topic;

[0140] The merging unit is used to merge the knowledge graphs corresponding to all text sets to obtain the final knowledge graph.

[0141] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A global knowledge graph management method, characterized in that: The method comprises: Receive the text containing the author parameters uploaded by the uploader and build a text library; Determine the application probability of each text according to the author parameter, randomly select text from the text library based on the application probability, and create at least one text set; Performing topic identification on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic; Create a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic; The steps of creating a knowledge graph include reading the topic distribution of each text set and the word distribution corresponding to the topic, building a knowledge graph corresponding to each text set; merging the knowledge graphs corresponding to all text sets to obtain the final knowledge graph; The step of creating a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic includes: Read the word distribution of each topic in each text set, and create a circular node corresponding to each word; the name of the circular node is the word, and the radius of the circular node is proportional to the probability of the word under the topic; The circular nodes are clustered according to the topics to which each word belongs, and each type of circular node is scaled proportionally according to the probability of the topic. After all the circular nodes are scaled, the knowledge graph corresponding to the text set is obtained; the scaling ratio is proportional to the probability of the topic; Merge the knowledge graphs corresponding to all text sets to obtain the final knowledge graph; The step of merging the knowledge graphs corresponding to all text sets to obtain the final knowledge graph includes: Read the knowledge graph corresponding to each text set and merge them; During the merge process: For each circular node, query the corresponding text set and the topic corresponding to the circular node in the text set; When the same circular node corresponds to multiple text sets and the corresponding topics are the same, the circular node is scaled twice according to the number of topics; When the same circular node corresponds to multiple text sets and the corresponding topics are different, the circular nodes corresponding to each topic are retained and connected; When a circular node corresponds to a text set, the circular node corresponding to each topic is retained.

2. The global knowledge graph management method according to claim 1, characterized in that: The step of receiving the text containing the author parameters uploaded by the uploader and constructing the text library comprises: Receive the domain index input by the administrator, retrieve the text based on the domain index, and synchronously obtain the author parameters; Inputting the author parameters into the trained evaluation model to determine the utility value of the text; the utility value is used to determine the application probability of the text; Count texts and their utility values ​​and build a text library.

3. The global knowledge graph management method according to claim 1, characterized in that: The step of performing topic identification on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic includes: The number of topics entered by the receiving administrator; For each text, extract a topic distribution from a Dirichlet distribution based on the number of topics; For each topic, draw a word distribution from a Dirichlet distribution; For each text, iteratively update the topic distribution and the word distribution based on Gibbs sampling; After the iterative update is completed, the final topic distribution and final word distribution for each text are output.

4. The global knowledge graph management method according to claim 3 is characterized in that: The iterative update process is: Generate document content based on the extracted topic distribution and word distribution of each topic; For each word in the document content , calculate the conditional probability that the current word belongs to each topic, and reallocate the topic of the current word according to the calculated conditional probability; The loop is executed until the preset number of loops is reached; The document content generation process is as follows: Distribution by topic Extract a topic from , from the selected topics Word distribution Extract a word from ; The calculation process of conditional probability is: ； In the formula, represents the conditional probability, Representation Document Middle The topic assignment of words, Indicates to remove the document Middle The distribution of all topics except words, The actual vocabulary representing all the words; and are preset values, all of which are 0.1; For Documentation Theme The number of words in , excluding the current word; For Documentation The total number of words in , excluding the current word, Is the theme Chinese words The number of occurrences, excluding the current word, Is the theme The total number of words in , excluding the current word; is the dimension of each element in the topic distribution; is the dimension of each element in the word distribution.

5. A global knowledge graph management system, characterized in that: The system comprises: The text library construction module is used to receive the text containing the author parameters uploaded by the uploader and construct the text library; A text set creation module, used to determine the application probability of each text according to the author parameter, randomly select texts from the text library based on the application probability, and create at least one text set; A text recognition module is used to perform topic recognition on the text set to obtain the topic distribution of the text set and the word distribution corresponding to each topic; A knowledge graph creation module is used to create a knowledge graph based on the topic distribution of all text sets and the word distribution corresponding to each topic; The steps of creating a knowledge graph include reading the topic distribution of each text set and the word distribution corresponding to the topic, building a knowledge graph corresponding to each text set; merging the knowledge graphs corresponding to all text sets to obtain the final knowledge graph; The knowledge graph creation module includes: A node creation unit is used to read the word distribution of each topic in each text set and create a circular node corresponding to each word; the name of the circular node is the word, and the radius of the circular node is proportional to the probability of the word under the topic; The node clustering unit is used to cluster the circular nodes according to the topics to which each word belongs, and to scale each type of circular node proportionally according to the probability of the topic. After all the circular nodes are scaled, the knowledge graph corresponding to the text set is obtained; the scaling ratio is proportional to the probability of the topic; The merging unit is used to merge the knowledge graphs corresponding to all text sets to obtain the final knowledge graph; The knowledge graphs corresponding to all text sets are merged to obtain the final knowledge graph, which includes: Read the knowledge graph corresponding to each text set and merge them; During the merge process: For each circular node, query the corresponding text set and the topic corresponding to the circular node in the text set; When the same circular node corresponds to multiple text sets and the corresponding topics are the same, the circular node is scaled twice according to the number of topics; When the same circular node corresponds to multiple text sets and the corresponding topics are different, the circular nodes corresponding to each topic are retained and connected; When a circular node corresponds to a text set, the circular node corresponding to each topic is retained.

6. The global knowledge graph management system according to claim 5, characterized in that: The text library construction module includes: A text retrieval unit, used to receive a domain index input by the management party, retrieve text based on the domain index, and synchronously obtain author parameters; A text evaluation unit, which is used to input the author parameters into the trained evaluation model to determine the utility value of the text; the utility value is used to determine the application probability of the text; The text statistics unit is used to count texts and their utility values ​​and build a text library.

7. The global knowledge graph management system according to claim 5, characterized in that: The text recognition module comprises: A quantity receiving unit, used to receive the quantity of topics input by the management party; A topic distribution extraction unit, configured to extract a topic distribution from a Dirichlet distribution based on the number of topics for each text; A word distribution extraction unit, used for extracting word distribution from Dirichlet distribution for each topic; An iterative updating unit, used for iteratively updating the topic distribution and the word distribution for each text based on Gibbs sampling; The result output unit is used to output the final topic distribution and final word distribution for each text after the iterative update is completed.

Images