An evaluation weight calculation method based on a knowledge graph

Abstract

The application discloses a kind of based on knowledge graph's evaluation weight calculation method, including the following steps: according to evaluation related data to establish evaluation knowledge graph and keyword vector library;Combining the evaluation target set and the keyword vector similarity of calculation is searched in the adaptive index of index library Form evaluation index system;The established evaluation index system is converted into directed graph, and the knowledge reasoning probability is calculated by the keyword vector similarity between the nodes of directed graph, replaces the random walk probability in existing PageRank algorithm, forms improved PageRank algorithm;According to the PageRank value of each node calculated by improved PageRank algorithm as the objective weight of each evaluation index.The application can make full use of expert knowledge and other data, explore the potential connection between evaluation indexes, automatically calculate the weight of each index in evaluation index system, so that the determination of evaluation weight is not dependent on subjective judgment, improve the efficiency and objectivity of evaluation decision process.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of weight calculation and knowledge graph technology, and particularly relates to a knowledge graph-based evaluation weight calculation method. BACKGROUND

[0002] A knowledge graph is a knowledge base established by collecting data information from various sources and is a collection of knowledge. The knowledge graph was formally proposed by Google in 2012 to improve the capabilities of search engines and improve the search experience of users. The knowledge graph is essentially a semantic network that represents entities and their relationships in the objective world in the form of a graph. The nodes in the knowledge graph represent entities, and the edges between the nodes represent the relationships between the entities. Knowledge graph technology mainly includes knowledge representation, knowledge graph construction, and knowledge graph application. Among them, knowledge graph construction is the most critical link in forming a knowledge graph. At present, knowledge graphs have shown great application value in natural language understanding, big data analysis, intelligent question answering, Internet of Things, and other fields, and are an important driving force for the future development of artificial intelligence.

[0003] Evaluation weight calculation is the process of determining the importance of each index in the evaluation index system. Evaluation weight calculation methods are mainly divided into three categories: subjective evaluation weight calculation methods, objective evaluation weight calculation methods, and subjective and objective combined evaluation weight calculation methods. Subjective evaluation weight calculation methods include the analytic hierarchy process, network analysis method, decision laboratory analysis method, etc., objective evaluation weight calculation methods include entropy weight method, principal component analysis method, etc., and the subjective and objective combined evaluation weight calculation method is to combine the above two methods to obtain a comprehensive weight. Evaluation weight calculation is an important link in the evaluation decision-making process and has a significant impact on the final evaluation result. Reasonable evaluation weight can enhance the scientificity and rationality of the evaluation process.

[0004] Most current researches adopt a subjective and objective combined evaluation weight calculation method, with subjective evaluation weight as the main method and objective evaluation weight as the auxiliary method, i.e., emphasizing the role of people in the evaluation decision-making process and making objective corrections. However, most current objective evaluation weight calculation methods are based on single index data to find their distribution and variation rules to assign weights, which makes it difficult to explore deeper connections between indexes. SUMMARY

[0005] The present application overcomes the deficiencies in the prior art and provides a knowledge graph-based evaluation weight calculation method that can fully utilize expert knowledge and other data to explore potential connections between evaluation indexes and automatically calculate the weights of each index in the evaluation index system, making the determination of evaluation weights independent of human subjective judgment and improving the efficiency and objectivity of the evaluation decision-making process.

[0006] Technical solution: To achieve the above object, the application provides an evaluation weight calculation method based on a knowledge graph, comprising the following steps:

[0007] S1: According to the evaluation related data, an evaluation knowledge graph and a keyword vector library are established, the evaluation knowledge graph is used as an evaluation index library of the evaluation index system, and the keyword vector library is used as the basis for evaluation index retrieval;

[0008] S2: The evaluation index system is formed by combining the set evaluation target and the calculated keyword vector similarity of the adaptive index in the index library;

[0009] S3: The established evaluation index system is converted into a directed graph, and the knowledge reasoning probability is calculated through the keyword vector similarity between the nodes of the directed graph to replace the random walk probability in the existing PageRank algorithm, forming an improved PageRank algorithm;

[0010] S4: The PageRank value of each node calculated according to the improved PageRank algorithm is used as the objective weight of each evaluation index.

[0011] Further, the establishment method of the evaluation knowledge graph in step S1 is: according to the evaluation task, collecting related text data, and using knowledge graph construction technology to establish the evaluation knowledge graph;

[0012] The establishment method of the keyword vector library is: for the collected text data, using the word embedding algorithm to obtain the word vector library, then extracting the keywords of each node of the knowledge graph, and finding the corresponding word vector to form the keyword vector library.

[0013] Further, the formation of the evaluation index system in step S2 is: according to the set evaluation target, screening out the nodes with specific labels in the knowledge graph, and calculating the keyword vector similarity with the evaluation target for further retrieval, then, all evaluation indexes and their dependency relationships that meet the conditions are summarized to form the evaluation index system.

[0014] Further, the evaluation target is set by experts according to the evaluation task;

[0015] The calculation method of the keyword vector similarity is:

[0016]

[0017] Wherein, represents the keyword vector of the evaluation target, represents the keyword vector of the knowledge graph node.

[0018] Further, the step S3 is specifically:

[0019] A1: Constructing the directed graph of the evaluation index system

[0020] After the construction of the evaluation index system, each evaluation index in the evaluation index system is converted into a node in the definition of the directed graph, and the directed dependence relationship between the indexes is converted into a directed edge in the definition of the directed graph, realizing the conversion process from the evaluation index system to the directed graph;

[0021] A2: Establishing an improved random walk model

[0022] Define the directed graph of the evaluation index system:

[0023] G = (V, E)

[0024] Where V and E represent the set of nodes and directed edges respectively, and each directed edge corresponds to a pair of ordered nodes;

[0025] The random walk model is essentially a Markov chain defined on a directed graph, and its Markov property is represented as the state transition probability from the current node state to the next node state, which only relates to the current node state. In the improved random walk model, the nodes are no longer transferred to their directed nodes with equal probability, but with knowledge reasoning probability. The directed graph of the evaluation index system is essentially established based on the knowledge graph, so the nodes in the directed graph, i.e. the evaluation indexes, correspond to the nodes in the knowledge graph, and the keyword vector of the node can be queried through the knowledge graph;

[0026] Define the knowledge similarity of nodes v i and v j in the directed graph G of the evaluation index system as:

[0027]

[0028] Where, is the highest ordered keyword vector of nodes v i and v j in the knowledge graph;

[0029] Assuming that node j in the directed graph has k directed edges out, the state transition probability of node j to each node can be represented by the knowledge similarity as:

[0030]

[0031] Where, p i ' j is the knowledge reasoning probability, representing the probability of transferring from evaluation index j to evaluation index i with knowledge as the measure;

[0032] Assuming that the directed graph G has m nodes, the state transition probability matrix of the improved random walk model is represented as:

[0033]

[0034] Obviously, p i ' j ≥0, The state transition probability matrix meets the properties of state transition probability matrix.

[0035] Further, the calculation method of the PageRank value in the step S4 is:

[0036] The evaluation index system directed graph established based on the knowledge graph may not meet the conditions of aperiodicity and strong connectivity, and thus the Markov chain thereof may not have a steady state distribution.

[0037] For a general evaluation index system directed graph G, assuming that the directed graph G has m nodes, the state transition probability matrix P based on the improved random walk model is obtained, if the steady state distribution of the Markov chain is required, in addition to defining the improved random walk model of the directed graph G, a completely random walk model needs to be defined again, and the element values of the state transition probability matrix P' of the completely random walk model are all The new state transition probability matrix is composed of the linear combination of P and P', and the Markov chain obtained on this basis can be proved to have a steady distribution R, and R has the following equation:

[0038]

[0039] wherein, the linear combination coefficient d (0≤d≤1) is a damping factor, the value is taken according to experience, E is a full 1 matrix of m x m; l is an m-dimensional vector with all components being 1, R represents the steady distribution in the general case, and each component of R is the PageRank value of each node, that is, the objective weight of each evaluation index:

[0040] R=[PR(v1),PR(v2),…,PR(v m )] T , v1,v2,...,v m ∈V

[0041] wherein, PR(v i )>0, The first term of the formula represents the probability distribution of the model walking to each node according to the state probability transition matrix P when the state is steady, and such probability distribution accounts for the weight d; the second term represents the probability distribution of the model walking to each node according to the state probability transition matrix P' when the state is steady, and such probability distribution accounts for the weight 1-d.

[0042] Intuitively, the second term as a smoothing term is introduced to make the node have a probability of 1-d to transfer completely randomly, and to walk to each node with equal probability The state transition to any node ensures that the state transition of the node without a directed edge can also be realized, thereby ensuring that the Markov chain has a stationary distribution.

[0043] Further, the calculation process of the PageRank value in the step S4 comprises the following steps.

[0044] B1: Calculate the state transition probability matrix P of the directed graph G of the evaluation index system based on knowledge measurement.

[0045] B2: Let t = 0, and select an initial state distribution x0.

[0046] B3: Calculate the general state transition probability matrix A of the directed graph.

[0047]

[0048] B4: Iterate and normalize the result vector.

[0049] y t+1 = Ax t

[0050]

[0051] B5: When ||x t+1 -x t ||≤ε, let R = x t , and stop iteration.

[0052] B6: Otherwise, let t = t + 1, and execute the step B4.

[0053] B7: Normalize the R to represent a probability distribution.

[0054] The R obtained after the above steps is the PageRank value of each node of the directed graph of the evaluation index system, that is, the objective weight of each index.

[0055] Advantages: Compared with the prior art, the present application has the following advantages:

[0056] 1: According to the evaluation task related data, the present application establishes the evaluation knowledge graph and the word vector library of the keywords of each node of the knowledge graph as the basis for constructing the evaluation index system and weight calculation, reduces the degree of human participation in the evaluation process, and improves the efficiency of the evaluation.

[0057] 2: The present application defines the conversion process of the evaluation index system to the directed graph, so that the algorithm of graph theory can be applied to the evaluation index system, and a new weight calculation idea is provided.

[0058] 3、The application improves the PageRank algorithm, calculates the knowledge reasoning probability through the similarity of the keyword vector between the nodes in the evaluation index system directed graph, replaces the equal probability transfer probability in the PageRank random walk model, so that the calculated PageRank value of each index can reflect the objective weight measured by knowledge, and avoids the theme deviation problem existing in the traditional PageRank algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 The application provides a knowledge graph-based evaluation weight calculation method.

[0060] Figure 2 The application provides a knowledge graph.

[0061] Figure 3 The application provides an evaluation index system based on a knowledge graph.

[0062] Figure 4 The application provides an error change graph in the iteration process of the improved PageRank algorithm.

[0063] Figure 5 The application provides each bottom-layer evaluation weight based on the improved PageRank algorithm. DETAILED DESCRIPTION

[0064] The application will be further illustrated below in combination with the drawings and specific embodiments, and it should be understood that the embodiments are only used for illustrating the application and are not used for limiting the scope of the application, and after reading the application, the modification of various equivalent forms of the application by those skilled in the art falls within the scope defined by the claims attached to the application.

[0065] The application provides a knowledge graph-based evaluation weight calculation method, as shown in Figure 1 The method comprises the following steps:

[0066] S1: according to the evaluation related data, an evaluation knowledge graph and a keyword vector library are established, the evaluation knowledge graph is used as an evaluation index library of the evaluation index system, and the keyword vector library is used as the basis for evaluation index retrieval;

[0067] S2: in combination with the set evaluation target and the calculated keyword vector similarity, the adaptive index in the index library is retrieved to form an evaluation index system;

[0068] S3: the established evaluation index system is converted into a directed graph, the knowledge reasoning probability is calculated through the keyword vector similarity between the nodes in the directed graph, the random walk probability in the existing PageRank algorithm is replaced, and an improved PageRank algorithm is formed;

[0069] S4: Calculate the PageRank value of each node as the objective weight of each evaluation index according to the improved PageRank algorithm.

[0070] The method for establishing the evaluation knowledge graph in step S1 is: according to the evaluation task, collecting related text data, and establishing the evaluation knowledge graph by using the knowledge graph construction technology;

[0071] The method for establishing the keyword vector library is: obtaining the keyword vector library by using the word embedding algorithm for the collected text data, then extracting the keywords of each node of the knowledge graph, and finding the corresponding keyword vector to form the keyword vector library.

[0072] The formation of the evaluation index system in step S2 is specifically: according to the set evaluation target, screening out the nodes with specific labels in the knowledge graph, and calculating the keyword vector similarity between the nodes and the evaluation target for further retrieval, then, all the evaluation indexes and the dependency relationship therebetween that meet the conditions are summarized to form the evaluation index system.

[0073] The evaluation target is set by experts according to the evaluation task;

[0074] The calculation method of the keyword vector similarity is:

[0075]

[0076] wherein, represents the keyword vector of the evaluation target, represents the keyword vector of the node of the knowledge graph.

[0077] The specific process of step S3 is:

[0078] A1: Constructing the directed graph of the evaluation index system

[0079] After the construction of the evaluation index system is completed, each evaluation index in the evaluation index system is converted into a node in the definition of the directed graph, and the directed dependency relationship between the indexes is converted into a directed edge in the definition of the directed graph, so as to realize the conversion process from the evaluation index system to the directed graph;

[0080] A2: Establishing the improved random walk model

[0081] Defining the directed graph of the evaluation index system:

[0082] G=(V,E)

[0083] wherein V and E represent the set of nodes and directed edges respectively, and each directed edge corresponds to a pair of ordered nodes;

[0084] The random walk model is essentially a Markov chain defined on a directed graph, and the Markov property is represented as the state transition probability from the current node state to the next node state only related to the current node state. In the improved random walk model, the node is no longer transferred to the node it points to with equal probability, but is transferred with knowledge reasoning probability. The directed graph of the evaluation index system is essentially established based on the knowledge graph, so the node in the directed graph, that is, the evaluation index corresponds to the node in the knowledge graph, can query the keyword vector of the node through the knowledge graph;

[0085] The knowledge similarity of the nodes v i and v j in the directed graph G of the evaluation index system is defined as:

[0086]

[0087] Wherein, is the highest ranked keyword vector of the nodes v i and v j in the knowledge graph;

[0088] Assuming that the node j in the directed graph has k directed edges out, the state transition probability of the node j to each node can be represented by the knowledge similarity as:

[0089]

[0090] Wherein, p′ ij is the knowledge reasoning probability, representing the probability of transferring from the evaluation index j to the evaluation index i with knowledge as the measurement;

[0091] Assuming that the directed graph G has m nodes, the state transition probability matrix of the improved random walk model is represented as:

[0092]

[0093] Obviously, p′ ij ≥ 0, complying with the properties of the state transition probability matrix.

[0094] The calculation method of the PageRank value in step S4 is:

[0095] The directed graph of the evaluation index system established based on the knowledge graph may not satisfy the conditions of aperiodicity and strong connectivity, so the Markov chain thereof may not have a steady-state distribution.

[0096] For the general evaluation index system has a directed graph G, assuming it has m nodes, based on the improved random walk model state transition probability matrix is P, if you want to get its Markov chain steady state distribution, in the directed graph G in addition to define its own improved random walk model, also need to define a completely random walk model, its state transition probability matrix P' element value is The new state transition probability matrix is composed of the linear combination of P and P', the Markov chain obtained on this basis can be proved to have a stationary distribution R, R has the following equation:

[0097]

[0098] Wherein, the linear combination coefficient d (0≤d≤1) is the damping factor, according to the experience value, E is the m×m full 1 matrix; L is an m-dimensional vector with each component being 1, R represents the stationary distribution in general case, each component of R is the PageRank value of each node, that is, the objective weight of each evaluation index:

[0099] R=[PR(v1),PR(v2),…,PR(v m )] T , v1,v2,...,v m ∈V

[0100] Wherein, PR(v i )>0, The first term of the formula represents the probability distribution of the model walking to each node according to the state probability transition matrix P when the state is stationary, and such probability distribution accounts for weight d; The second term represents the probability distribution of the model walking to each node according to the state probability transition matrix P' when the state is stationary, and such probability distribution accounts for weight 1-d.

[0101] The calculation process of PageRank value includes the following steps:

[0102] B1: Calculate the state transition probability matrix P of the evaluation index system directed graph G based on knowledge measurement;

[0103] B2: Let t = 0, select the initial state distribution x0;

[0104] B3: Calculate the general state transition probability matrix A of the directed graph

[0105]

[0106] B4: Iteration and normalization of the result vector

[0107] y t+1 =Ax t

[0108]

[0109] B5: when ||x t+1 -x t ||≤ε, let R=x t , stop iteration;

[0110] B6: otherwise, let t=t+1, execute step B4;

[0111] B7: normalize R to represent a probability distribution;

[0112] The R obtained after the above steps is the PageRank value of each node of the evaluation index system directed graph, that is, the objective weight of each index.

[0113] In order to verify the actual effect of the above scheme, the evaluation weight calculation method is applied in this embodiment, which is to calculate the evaluation weight of the evaluation index of the evaluation target of the mechanical hand fault risk of the production line of a certain automobile manufacturing enterprise, which is as follows:

[0114] First step: collect evaluation task related data, build evaluation knowledge graph and keyword vector library of each node thereof.

[0115] The data of this embodiment comes from a certain automobile manufacturing company. The company produced a large number of equipment problem solving reports during a certain period of production. These reports describe in detail the common fault problems on the production line during the production process of the company, the causes of the fault and the corresponding maintenance process, etc., which contain specific evaluation indexes related to the evaluation target and rich expert knowledge.

[0116] Part of the data in the equipment problem solving report of the company is labeled and processed as the training, verification and test data set of the knowledge extraction Casrel model, and another part is input into the trained Casrel model to realize knowledge extraction. This section adopts the method of manual labeling to ensure the quality of the data set, so as to improve the training effect of the model.

[0117] The text description of the production line fault in the equipment problem solving report is analyzed, and the following five types of triples are summarized and determined to describe the entity categories and their relationships in the equipment problem solving report, which are: (specific fault, trace, fault cause), (fault cause, cause, fault cause), (handler, use, maintenance method), (fault cause, involve, spare part), (equipment model, belong to, equipment classification).

[0118] The text of the equipment problem solving report is annotated in the form of the above-mentioned triplets to obtain a triplet data set, and the annotated data set is divided into a training set, a verification set and a test set according to a ratio of 8:1:1. The Casrel model after completing the training performs knowledge extraction on the remaining data, and the extracted triplets are stored in a neo4j graph database to form an evaluation knowledge graph as shown in Figure 2 .

[0119] The text data of the equipment problem solving report adopts a Word2vec word embedding algorithm to obtain a word vector library, and the node information in the knowledge graph is referenced to define the keywords of the nodes in the knowledge graph using a TF-IDF document keyword algorithm, which is defined as follows:

[0120]

[0121]

[0122] TF-IDF = TF * IDF

[0123] Wherein, c represents the number of times a certain word appears in the node, m represents the total number of words in the node, N represents the total number of nodes in the knowledge graph, and n represents the number of nodes containing the word.

[0124] The calculated TF-IDF values are arranged in descending order, and the highest ranked word or several higher ranked words are taken as the keywords. The keyword vector corresponding to the keyword vector in the word vector library is obtained.

[0125] Second step: According to the evaluation target and keyword retrieval algorithm, search the adaptive index in the evaluation knowledge graph to construct the evaluation index system.

[0126] The production line manipulator fault risk evaluation target is analyzed, and the specific evaluation indexes of the next level can be corresponded to the “specific fault” category in the knowledge graph entity category, so the evaluation knowledge graph is queried with the “manipulator” keyword vector, and the nodes with high similarity to the evaluation target and the “specific fault” label are searched. Five nodes with the highest similarity to the evaluation target are searched, which are: manipulator clamping fault, manipulator zero return fault, manipulator grabbing fault, manipulator power failure, and manipulator non-action. These five nodes are taken as the specific fault element set, and then the nodes related to the five nodes are further queried, and the nodes with the label “fault reason” are selected as the bottom-level evaluation indexes after analysis, a total of 30, and the weight of the bottom-level indexes is taken as the evaluation weight to be calculated. The original relationship of the retrieved evaluation indexes in the knowledge graph is retained, and the constructed evaluation index system is as shown in Figure 3 .

[0127] Third step: Calculate the objective evaluation weight of the bottom-level evaluation indexes according to the improved PageRank algorithm.

[0128] The evaluation index system is converted into a corresponding directed graph according to the improved PageRank algorithm. First, the similarity of keyword vectors between the indexes with existing relationships is calculated to obtain a state transition matrix based on knowledge reasoning probability.

[0129] In the iterative calculation of the improved PageRank, the allowed error ε is set to 0.00001, and the damping coefficient d is set to 0.85. At the same time, because the final stationary distribution is irrelevant to the initial distribution, in order to be more in line with the actual meaning, the initial distribution is set to a 35-dimensional vector with each component being , indicating an initial equal weight state. In the iterative calculation process of the improved PageRank, the change of the error is shown in FIG. 4. It can be seen that the error gradually decreases and tends to be stable, indicating that the finally obtained PageRank value conforms to the stationary distribution.

[0130] The final stationary distribution is obtained, and the PageRank values of each bottom-level evaluation index are extracted and normalized to obtain an objective evaluation weight measured by knowledge as shown in Figure 5 .

[0131] The embodiment also provides a knowledge graph-based evaluation weight calculation system, which comprises a network interface, a memory and a processor. The network interface is used for receiving and sending signals in the process of transmitting information with other external network elements. The memory is used for storing computer program instructions capable of running on the processor. The processor is used for executing the steps of the above consensus method when running the computer program instructions.

[0132] The embodiment also provides a computer storage medium storing a computer program, which can implement the above-described method when the processor executes the computer program. The computer readable medium can be considered as tangible and non-transitory. Non-limiting examples of non-transitory tangible computer readable media include non-volatile memory circuits (such as flash memory circuits, erasable programmable read-only memory circuits or mask read-only memory circuits), volatile memory circuits (such as static random access memory circuits or dynamic random access memory circuits), magnetic storage media (such as analog or digital magnetic tapes or hard disk drives), and optical storage media (such as CDs, DVDs or Blu-ray discs), etc. The computer program includes processor executable instructions stored on at least one non-transitory tangible computer readable medium. The computer program can also include or depend on stored data. The computer program can include a basic input / output system (BIOS) that interacts with the hardware of a special-purpose computer, a device driver that interacts with a specific device of a special-purpose computer, one or more operating systems, user application programs, background services, background application programs, etc.

[0133] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-readable program code.

[0134] The present application is described in relation to flow diagrams and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It is understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams and / or block diagrams, 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, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in the present application illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product. In some embodiments, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product. Figure 1 The flow diagrams and / or block diagrams in the present application illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product. In some embodiments, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product.

[0135] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in the present application illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product. In some embodiments, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product. Figure 1 The flow diagrams and / or block diagrams in the present application illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product. In some embodiments, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product.

[0136] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in the present application illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product. In some embodiments, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product. Figure 1 The flow diagrams and / or block diagrams in the present application illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product. In some embodiments, each flow diagram and / or block diagram can represent a method, a system, and / or a computer program product.

Claims

1.A knowledge graph-based evaluation weight calculation method, characterized in that, Comprise the following steps: S1: according to the evaluation related data establishes the evaluation knowledge graph and the keyword vector library, the evaluation knowledge graph is used as the evaluation index library of the evaluation index system, and the keyword vector library is used as the basis for the evaluation index retrieval; S2: combining the set evaluation target and the calculated keyword vector similarity, the adaptive indexes in the index library are searched to form the evaluation index system; S3: the established evaluation index system is converted into a directed graph, and the knowledge reasoning probability is calculated through the keyword vector similarity between the nodes of the directed graph to replace the random walk probability in the existing PageRank algorithm, forming an improved PageRank algorithm; S4: according to the improved PageRank algorithm, the PageRank value of each node is calculated as the objective weight of each evaluation index; Step S3 is specifically: A1: constructing the directed graph of the evaluation index system After the construction of the evaluation index system is completed, each evaluation index in the evaluation index system is converted into a node in the definition of the directed graph, and the directed dependence relationship between the indexes is converted into a directed edge in the definition of the directed graph, realizing the conversion process of the evaluation index system to the directed graph; A2: establish an improved random walk model Define the directed graph of the evaluation index system: G=(V,E) Wherein V and E represent the set of nodes and directed edges respectively, and each directed edge corresponds to a pair of ordered nodes; Definition of the knowledge similarity between the nodes v i and v j in the directed graph G is as follows: wherein, is the highest ranked keyword vector for the node v i in the knowledge graph, j in the knowledge graph. Assuming that node j has k directed edges out of the directed graph, the state transition probability of node j to each node can be represented by the knowledge similarity as follows: Among them, p' ij It is the knowledge reasoning probability, which represents the probability of shifting from evaluation index j to evaluation index i using knowledge as a measure; Assuming that the directed graph G has m nodes, the state transition probability matrix of the improved random walk model is represented as: Obviously, p' ij ≥ 0, comply with the properties of a state transition probability matrix. 2.The knowledge graph-based evaluation weight calculation method according to claim 1, characterized in that, The establishment method of the evaluation knowledge graph in the step S1 is: according to the evaluation task, the related text data is collected, and the knowledge graph construction technology is used to establish the evaluation knowledge graph; The establishment method of the keyword vector library is: for the collected text data, the word embedding algorithm is used to obtain the word vector library, and then the keywords of each node of the knowledge graph are extracted, and the corresponding word vector is searched to form the keyword vector library. 3.The knowledge graph-based evaluation weight calculation method of claim 1, wherein, The formation of the evaluation index system in the step S2 is specifically: according to the set evaluation target, the nodes with specific labels in the knowledge graph are screened out, and the keyword vector similarity between the nodes and the evaluation target is calculated for further retrieval, and then all the evaluation indexes and the dependence relationship between them that meet the conditions are summarized to form the evaluation index system. 4.The knowledge graph-based evaluation weight calculation method of claim 3, wherein, The evaluation target is set by experts according to the evaluation task; The calculation method of the keyword vector similarity is: wherein, represents a keyword vector of an evaluation target, represents a keyword vector of a knowledge graph node. 5.The knowledge graph-based evaluation weight calculation method of claim 1, wherein, The calculation method of the PageRank value in the step S4 is: For the evaluation index system directed graph G, assuming that it has m nodes, the state transition probability matrix is P based on the improved random walk model, if the steady state distribution of the Markov chain is required, in addition to defining the improved random walk model on the directed graph G, a completely random walk model also needs to be defined, the element value of the state transition probability matrix P' of the completely random walk model is The new state transition probability matrix is composed of the linear combination of P and P', the Markov chain obtained on this basis can be proved to have a stationary distribution R, R has the following equation: Wherein, the linear combination coefficient d is a damping factor, 0≤d≤1, and the value is taken according to experience; E is a full 1 matrix of m*m; l is an m-dimensional vector with all components being 1, R represents a stationary distribution in general case, and each component of R is the PageRank value of each node, that is, the objective weight of each evaluation index: R = [PR(v1), PR(v2),..., PR(v m )] T , v1, v2,..., v m ∈ V where PR(v i ) > 0, The first term of the formula represents the probability distribution of the model walking to each node according to the state probability transition matrix P when the state stationary distribution is reached, and such a probability distribution accounts for the weight d; the second term represents the probability distribution of the model walking to each node with equal probability according to the state probability transition matrix P' when the state stationary distribution is reached, and such a probability distribution accounts for the weight 1-d. 6.The knowledge graph-based evaluation weight calculation method of claim 5, wherein, The calculation process of the PageRank value in the step S4 includes the following steps: B1: calculate the state transition probability matrix P of the evaluation index system directed graph G based on knowledge measurement; B2: let t=0, select the initial state distribution x0; B3: Calculate the general state transition probability matrix A of the directed graph B4: Iterate and normalize the result vector y t+1 = Ax t B5: when ||x t+1 - x t || < ε, let R = x t , stop iteration; B6: Otherwise, let t = t + 1, and perform step B4; B7: Normalize R to represent a probability distribution; After the above steps, R obtained is the PageRank value of each node of the directed graph of the evaluation index system, that is, the objective weight of each index.

Images