Enterprise credibility evaluation system and method based on knowledge graph

Abstract

The invention relates to an enterprise credibility evaluation system and method based on a knowledge graph. The evaluation system comprises an enterprise dictionary construction unit, a knowledge graph construction unit, an enterprise credibility calculation unit and an enterprise credibility evaluation unit. The enterprise dictionary construction unit is used for constructing a corpus of enterprise transaction vocabularies for the knowledge graph construction unit to call; the knowledge graph construction unit is used for linking enterprise entities with direct transactions and indirect transactions; the enterprise credibility calculation unit is used for calculating the credibility of the to-be-evaluated enterprise through the knowledge graph after obtaining the information and credit records of the to-be-evaluated enterprise; and the enterprise trust degree evaluation unit is used for evaluating the tradable trust degree of the to-be-evaluated enterprise according to the trust expected value of the user enterprise. According to the method, auxiliary calculation is performed on the trust degree of the to-be-evaluated enterprise through the knowledge graph, and the tradable trust level of the to-be-evaluated enterprise is further judged, so that a scientific auxiliary reference is provided for the user enterprise in the aspect of transaction object selection.

Description

Technical Field

[0001] The present invention relates to an enterprise trust evaluation system, and in particular to an enterprise trust evaluation system based on a knowledge graph. Background Art

[0002] With the rapid development of world economic integration and science and technology, the relationship between enterprises has changed from antagonistic competition to cooperative competition. Cooperation makes the development of enterprises tend to be win-win. Trust is the premise of business dealings between enterprises. Researching and evaluating the trustworthiness of enterprises has become the focus of current cooperation between enterprises. The organizational model of advantage sharing between enterprises is based on a cooperative relationship of mutual trust and mutual assistance, and can be a dynamic alliance organization. Enterprises need to make a series of evaluations and measurements on various aspects of transaction candidate enterprises to determine whether they are suitable for establishing a partnership and jointly achieve the overall goal.

[0003] Due to the current era of information expansion and overflow, it is difficult to evaluate the trust of enterprises by relying solely on media information. In current research, the trust judgment of enterprises is based on the analysis of the business information, financial status, scientific research capabilities and other factors of the enterprise and quantitative calculation or qualitative judgment. On the one hand, these factors do not take into account the credit situation of the enterprise, such as legal proceedings, breach of contract records, etc. On the other hand, they do not take into account the evaluation or influence of other enterprises on the enterprise. Therefore, this kind of analysis based only on the factors of the enterprise itself is relatively one-sided, and the calculated trust is not enough to truly reflect the trust level of the enterprise. The present invention introduces transaction trust and recommendation trust, combined with the direct trust of the enterprise itself, to conduct a comprehensive evaluation of the trust of the enterprise. The enterprise trust calculated based on these three trust components is more in line with actual needs, so that user enterprises have a more scientific basis in choosing partners. Summary of the invention

[0004] Purpose of the invention: The first purpose of the present invention is to provide a scientific enterprise trust evaluation system that can essentially reflect the trust of an enterprise. The second purpose of the present invention is to provide an enterprise trust evaluation system based on a knowledge graph.

[0005] Technical solution: The present invention is an enterprise trust evaluation system based on knowledge graph, including an enterprise dictionary construction unit, a knowledge graph construction unit, an enterprise trust calculation unit and an enterprise trust evaluation unit;

[0006] The enterprise dictionary construction unit is used to construct a corpus describing enterprise transaction vocabulary for use by the knowledge graph construction unit;

[0007] The knowledge graph building unit is used to link together business entities that have direct and indirect transactions;

[0008] The enterprise trust calculation unit is used to calculate the trust of the enterprise to be evaluated based on the knowledge graph after obtaining the basic information and credit record of the enterprise to be evaluated;

[0009] The enterprise trust evaluation unit is used to provide a trust expectation value that meets the enterprise's own expectations, judge the trustworthiness of the enterprise to be evaluated from the perspective of the expectation value, analyze and calculate the elements of trust, and finally add the enterprise to be evaluated to the trust rating library.

[0010] Furthermore, the enterprise dictionary construction unit constructs a vocabulary library containing enterprise names, transaction terms, and time terms. The system performs keyword extraction and uses the constructed enterprise dictionary to segment the enterprise data text of the enterprise alliance, providing a corpus for the construction of the knowledge graph, and using the corpus for pre-training of model calculations.

[0011] The knowledge graph construction unit includes the functions of four-tuple knowledge group extraction, knowledge reasoning and graph storage; by extracting the task, four-tuples with time knowledge are obtained. After the end of this unit, the user enterprise can obtain the knowledge graph about the transaction relationship with the enterprise to be evaluated, and use the graph database for visualization.

[0012] The enterprise trust calculation unit includes direct trust calculation, transaction trust calculation, and recommended trust calculation. The direct trust calculation is used to calculate the trust based on the basic information and credit record of the enterprise to be evaluated. The transaction trust calculation is used to calculate the trust based on the satisfaction between the user enterprise and the enterprise to be evaluated. The recommended trust calculation is used to calculate the transaction trust between the user enterprise and each enterprise in the transaction link with the enterprise to be evaluated.

[0013] The enterprise trust evaluation unit includes evaluation criteria division and grade library division classification. The evaluation criteria division is used to calculate and divide the evaluation criteria for whether the enterprise is trustworthy based on the historical transaction records of the user enterprise. The trust grade division classification is used to analyze the calculated trust components and classify the enterprise to be evaluated into the corresponding transaction trust grade.

[0014] The method for using the enterprise trust evaluation system based on the knowledge graph includes the following steps:

[0015] (1) The enterprise dictionary construction unit constructs a special dictionary containing vocabulary describing enterprises and enterprise transactions, providing a corpus for the subsequent knowledge graph construction unit;

[0016] (2) The knowledge graph construction unit is used to construct a knowledge graph of enterprise entities and whether there are transactions and indirect transaction relationships; first, the corpus is jointly annotated, and then the corpus text vector is converted into a knowledge quadruple with time knowledge. The obtained quadruple is used to mine hidden knowledge through the method of multi-relational cyclic knowledge reasoning, that is, the entity relationship of the quadruple corresponds to the domain information corresponding to all relationships under multiple entities, and the target entity domain information is used to obtain the vector representation of the domain feature. The obtained domain feature vector is then encoded through a temporal encoder, and the network layer output results of the long short-term memory network are extracted through a fully connected layer and classified by probability distribution calculation. Finally, all the obtained knowledge quadruple are input into the graph database for visualization.

[0017] (3) After the user enterprise inputs the enterprise to be evaluated, the enterprise trust calculation unit will query the basic information and credit record of the enterprise to be evaluated, and use the hierarchical analysis method to calculate the weights of each factor queried, and weight them to obtain the direct trust. At the same time, the enterprise trust calculation unit will use the constructed knowledge graph to query whether the user enterprise itself has a transaction record with the enterprise to be evaluated, return the transaction details and use formula (1) to calculate the transaction trust.

[0018]

[0019] Among them, B2 represents the transaction trust, S i represents the satisfaction of the i-th transaction, I i Indicates the importance of the i-th transaction, T i represents the time weight of the i-th transaction, and n is the number of enterprises that directly trade with member enterprise i. Based on the trust formula, a satisfaction penalty factor is introduced to count the number of satisfied and dissatisfied times in the transaction history records. If the number of satisfied times is greater than the number of satisfied times, a reward is given, otherwise a penalty is given. The transaction trust with the satisfaction penalty factor added is calculated according to formula (2).

[0020]

[0021] Among them, B deal is the final transaction trust, S sat The number of transactions that indicate satisfaction, S unsat The number of transactions that indicated dissatisfaction.

[0022] At the same time, it is determined whether there is a link between the user enterprise and the enterprise to be evaluated through transaction connection. If so, the recommendation trust is calculated through the transaction trust between the enterprises on this transaction link. There are two types of links: single-path recommendation link and multi-path recommendation link. The recommendation trust of the single-path and multi-path recommendation links is calculated by formula (3) and formula (4).

[0023]

[0024] Among them, B3(i, j) represents the single-path recommendation trust from user enterprise i to enterprise j to be evaluated, depth(i, j-1) represents the link length between member enterprises and enterprises directly trading with the enterprise to be evaluated, and B2(i, i+1) represents the transaction trust between member enterprise i and its directly trading enterprise i+1. B3 represents the multi-path recommendation trust, which can be considered to exist n single-path recommendation links, and each link is given a different time weight and weighted summed, that is, the trust of the enterprise to be evaluated under multiple paths is obtained. B3(i, n, j) represents the single-path recommendation trust of member enterprise i to enterprise j to be evaluated through n enterprise recommendations.

[0025] (4) The enterprise trust evaluation unit will query the historical transaction records of the user enterprise itself, that is, query the trust of enterprises with which it has transactions and the trust of enterprises with which it has not transactions, obtain the evaluation criteria through the classification algorithm, compare the calculated trust with the evaluation criteria, classify according to the calculated three trust components and give the final result. The enterprise trust level is divided into three types according to direct trust, transaction trust and recommended trust, and is specifically divided into three major categories: tradable enterprises, enterprises to be considered for transaction and non-tradable enterprises. If the evaluation result is a tradable enterprise, the enterprise will be added to the trust level library of the user enterprise and the evaluation result will be returned to the user enterprise, showing the details of the trust level library of the user enterprise. If the evaluation result is a transaction enterprise to be considered or a non-tradable enterprise, the evaluation result will be recorded and added to the trust level library of the user enterprise, and then the next enterprise to be evaluated in the enterprise alliance will be randomly selected to perform a new round of trust calculation until an enterprise that meets the tradable trust level is found.

[0026] Beneficial effects: Compared with the existing enterprise trust evaluation method, the present invention has the following significant advantages: by constructing a knowledge graph related to enterprise transactions, it assists in the calculation of enterprise trust. On the basis of the direct trust of the user enterprise, transaction trust and recommended trust are proposed and calculated to achieve a comprehensive evaluation of enterprise trust. The enterprise trust calculated based on these three trust components is more in line with actual needs, so that user enterprises have a more scientific basis for selecting partners. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is the system function structure diagram of the present invention

[0028] Figure 2 It is a specific implementation flow chart of the present invention

[0029] Figure 3 This is the knowledge graph construction flow chart of the present invention DETAILED DESCRIPTION

[0030] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0031] The first core of the present invention is a corporate trust evaluation system based on knowledge graph, which not only realizes the calculation of corporate trust, but also provides a trust level classification, providing a scientific and effective reference for enterprises to trust each other from different angles. Figure 1 As shown, the enterprise trust evaluation system based on knowledge graph of the present invention includes an enterprise dictionary construction unit, a knowledge graph construction unit, an enterprise trust calculation unit and an enterprise trust evaluation unit.

[0032] Figure 2 It is a specific implementation flow chart of the present invention, comprising the following steps:

[0033] Step 1: The enterprise dictionary construction unit is used to construct a dictionary containing enterprise, transaction terms, and time terms. This dictionary can be used to build a corpus. First, all enterprise names in the enterprise alliance are collected for dictionary construction, and transaction and time terms are added to the dictionary. The TF-IDF algorithm is used to extract high-frequency words in the text and update the dictionary.

[0034] Step 2: Knowledge graph construction. The knowledge graph construction flow chart is as follows: Figure 3 The knowledge graph construction includes four-element knowledge group extraction, knowledge reasoning, and graph storage.

[0035] The extraction of four-element knowledge groups includes entity extraction, relationship extraction, attribute extraction and time extraction. Before knowledge extraction, it is necessary to analyze the relationship attributes and mark the training corpus. After analyzing the relationship attributes to be extracted, the training corpus needs to be preprocessed. The original sentence is segmented and the data is marked through the dictionary constructed in step 1. The BIOES marking method is used here. BIOES marking is a more complete marking method than BIO marking. B (Begin) indicates the beginning of the entity name, E (End) indicates the end of the entity name, I (Intermediate) indicates the middle word between the beginning and the end of the entity name, O (Other) represents other, marking irrelevant characters, and S (Single) represents a single character. Since the entity-relationship joint learning model is adopted in this implementation, the strategy of joint tagging of entity relationships is adopted. The main rules for entity labeling are as follows: {B (entity start), I (entity inside), E (entity end), S (single entity)}; the type information of entity relationships is mainly as follows: {1 (entity 1), 2 (entity 2), 3 (time)}, and other word tags not in the quadruple are recorded as 0.

[0036] In the task of extracting four-tuple knowledge groups, the annotated corpus text vectors are converted into feature vectors based on the Word2Vec model and input into the long short-term memory network-conditional random field model. In order to build a dynamic knowledge graph, it is necessary to add time to the knowledge tuple. Each node in the graph represents an entity, and the directed edges between entities represent relationships. The edge of the directed edge adds the time dimension, that is, the four-tuple knowledge tuple (entity, relationship, entity, time) is jointly extracted. The input layer of the Word2Vec model converts the word vectors of the text into a feature vector sequence. The feature vector is composed of the text vector of the input word text and an additional feature vector. For the feature vector of the input text processing vocabulary, the word vector of the text itself can be used as an additional feature vector. The task of the input layer is to convert the text word vector into a sequence annotation feature vector and send it to the network layer. The long short-term memory network-conditional random field model is used to jointly extract data annotations. The model inputs the above feature vectors into a bidirectional long short-term memory network and encodes the feature vectors. The encoded results are then sent as input to the random field model to classify the annotation sequence.

[0037] The knowledge reasoning task is a cyclic process of extracting, integrating and reasoning knowledge through time information. Because transactions between enterprise entities may have multiple transaction relationships at the same time point, a dynamic knowledge reasoning method with multi-relationship cyclic time is adopted. The extracted knowledge quadruple can be regarded as (head entity, relationship, tail entity, time), and the symbol is (h, r, t, T), where h represents the head entity, r represents the relationship, t represents the tail entity, and τ represents the time point. The entity neighborhood information at each time point is extracted and integrated through multi-relationship neighbor clustering.

[0038] First, the domain information corresponding to all the relations under multiple entities corresponding to the entity relations of the four-tuple is used to retain the domain information under multiple relations corresponding to the target entity to be predicted. Then, the target entity domain information is subjected to the method of obtaining the cumulative mean vector to obtain the vector representation of the domain features. Multi-relationship proximity clustering is calculated according to formula (5).

[0039]

[0040] in, represents the set of neighborhood tail entities t corresponding to the head entity h at time point τ, Represents the number of all elements, e t is the vector representation of the tail entity t, according to the relationship set Get the collection of corresponding domain entities Finally, the vector representation f(h, τ) of the domain feature can be obtained. After obtaining the domain feature vector, it is passed through the temporal encoder, that is, the feature vector is input into the long short-term memory network layer, and then the output result after the long short-term memory network layer is extracted through the fully connected layer, and finally the Softmax activation function is used to perform probability distribution calculation and classification.

[0041] Finally, the graph storage task is performed. The quadruple knowledge extracted and inferred in the above process is stored in the graph database Neo4j, and the knowledge graph is visualized using data structures such as nodes, edges, and attributes. Each node represents an entity, and the edge represents a cooperative relationship. The time in the quadruple is stored on the edge.

[0042] Step 3: Calculate the enterprise trust. Use the knowledge graph constructed in step 2 to query the basic information and recommended links of the enterprise to be evaluated, and calculate the trust of each enterprise based on the queried information.

[0043] Direct trust calculation, through the hierarchical analysis method to quantitatively calculate the basic information of the enterprise to be evaluated, such as debt-to-asset ratio, credit record, etc. To determine the indicator weight, first construct a judgment matrix. The same level of the judgment matrix represents the judgment value of the relative importance of each indicator, which is judged by various experts. The judgment value uses a scale of 1-9. Each element in the matrix represents the pairwise comparison value of the relative importance of the horizontal indicator to each column indicator. After constructing the judgment matrix, start to calculate the maximum eigenvalue of the judgment matrix and its corresponding eigenvector and thereby obtain the weight coefficient of each indicator. The square root method is used to solve the eigenvalue and its eigenvector. The specific steps are as follows:

[0044] First, calculate the product Q of each row element of the judgment matrix according to formula (6): i .

[0045]

[0046] Next, calculate Q according to formula (7) i The nth root of

[0047]

[0048] Then, according to formula (8), the vector Perform normalization:

[0049]

[0050] The solution is R = [R1, R2, .... R n ] T is the eigenvector.

[0051] The direct trust B1 is calculated by taking the solved feature vector as the weight and multiplying it by the corresponding index vector C according to formula (9).

[0052] B1=R T C (9)

[0053] The transaction trust is calculated according to formula (10). Each transaction between enterprises in the enterprise alliance will be evaluated. The evaluation includes the time, satisfaction and importance of the transaction content of both enterprises. The satisfaction range is 1-10, and the importance range is 0.1-1. Based on the evaluation, the transaction trust between enterprises in the enterprise alliance can be calculated.

[0054]

[0055] Where S i represents the satisfaction of the i-th transaction, I i Indicates the importance of the i-th transaction, T i It represents the time weight of the i-th transaction. The more recent the transaction, the greater the time weight. On the contrary, the older the time, the lower the weight. The value of n is the number of enterprises that directly trade with the member enterprise. The present invention adds a satisfaction penalty factor, which divides satisfaction into satisfied and dissatisfied. In each transaction evaluation, a score greater than 4 indicates satisfaction, while a score less than or equal to 4 indicates dissatisfied. The transaction trust is further expressed as formula (11):

[0056]

[0057] B deal is the final transaction trust. The above formula indicates that in the historical transactions of enterprises in the enterprise alliance, when the number of satisfactory transactions S sat Greater than the number of unsatisfactory transactions S unsat When the number of satisfactory transactions S sat Less than the number of dissatisfaction S unsat When a person fails to do so, he will be given a corresponding "punishment".

[0058] The recommendation trust calculation is to find the connection path with the enterprise to be evaluated by querying the knowledge graph, that is, to find one or more paths between the user enterprise and the enterprise to be evaluated by using the idea of ​​deep traversal. The starting point of each path is the user enterprise i, and the end point is the enterprise to be evaluated j. The nodes on the path are the enterprises in the other enterprise alliances that have transaction relationships with each other. The paths are divided into single-path recommendation links and multi-path recommendation links.

[0059] For a single-path recommendation link, there is only one unidirectional recommendation link between the member enterprises in the enterprise alliance and the enterprise to be evaluated, denoted as R = [i, a, ... b, j]. The transaction trust between the member enterprise and the enterprise to be evaluated and their direct transaction enterprises is calculated respectively. Since the member enterprise does not have direct transaction behavior with other enterprises in the link, the trust of the enterprise through which the member enterprise directly trades will decay as the link depth increases. The specific calculation is based on formula (12).

[0060]

[0061] Among them, depth(i,j-1) represents the link length between the member enterprise and the enterprise to be evaluated that directly trades, B2(i,i+1) represents the transaction trust between the member enterprise and its directly trading enterprise, and the value of m is the number of enterprises with pairwise direct trading relations between company i and company j.

[0062] For multi-path recommendation links, they can be regarded as multiple single-path recommendation links. The recommendation trust of each single-path recommendation link is calculated, and then the recommendation trust calculated for each single path is given a weight t according to the time between the member enterprise and its multiple direct transaction enterprises. n The closer the transaction time point is, the greater the weight is given, and vice versa. The specific calculation is based on formula (13).

[0063] B3=∑t n B3(i,n,j) (13)

[0064] Among them, B3(i, n, j) represents the single-path recommendation trust of enterprise j to be evaluated obtained by member enterprise i through the recommendation of n enterprises, and the value of n is the number of enterprises with which the member enterprise directly trades; B3 represents the multi-path recommendation trust, which can be considered that there are n single-path recommendation links, and each link is assigned a different time weight and weighted summed, that is, the trust of the enterprise to be evaluated under multiple paths is obtained.

[0065] Step 4: Classification of evaluation criteria and trust level classification. The present invention does not formulate a unified scoring standard, but provides rules for the classification of scoring criteria for member enterprises in the enterprise alliance.

[0066] Through the knowledge graph constructed above, all enterprises in the enterprise alliance are found, and the direct trust of the user enterprise itself with and without transactions is found. Based on the idea of ​​classification, these enterprises are classified, and the trust is used as input to find the boundary between transactions and non-transactions through the support vector machine algorithm. This boundary is the judgment standard. For each member enterprise, due to their different historical transactions, the judgment standard is also different. Compare the trust calculated in step 3 with the judgment standard to determine whether the enterprise is trustworthy.

[0067] Trust level classification is designed to classify the trust of the evaluated enterprises in detail. The trust level library is generally divided into three categories: tradable enterprises, enterprises to be considered for trading, and non-tradable enterprises. Tradable means that the calculated trust is higher than the evaluation standard of the user enterprise, non-tradable means that the calculated trust is less than the evaluation standard, and to be considered for trading means that the calculated trust is less than the evaluation standard, and then the division is made according to the calculated trust components. For some cases of tradable and to be considered for trading, the calculated trust components (direct trust, transaction trust, and recommended trust) are analyzed and divided for reference by user enterprises. An example of the rating division table of the trust level library of a user enterprise is shown in Table 1.

[0068] Finally, the enterprise to be evaluated is added to the trust level library of member enterprises. If the trust level of the enterprise to be evaluated input by the user enterprise is higher than the user enterprise's own expected trust value, it is trustworthy. The result will be fed back to the user enterprise and other enterprises in its trust level library will be displayed. If the result of the evaluation is untrustworthy, the record of the evaluation result will be retained, and other enterprises in the enterprise alliance will be reselected in a random manner. The operations in the present invention will be repeated to continue the trust calculation and evaluation until an enterprise that meets the transaction trust level is found.

[0069] Table 1 Trust rating library rating classification table

[0070]

Claims

1. A knowledge graph-based enterprise trust evaluation system, characterized in that: It includes an enterprise dictionary construction unit, a knowledge graph construction unit, an enterprise trust calculation unit and an enterprise trust evaluation unit; The enterprise dictionary construction unit is used to construct a corpus describing enterprise transaction vocabulary for the knowledge graph construction unit to call; The knowledge graph construction unit is used to link together business entities that have direct transactions and indirect transactions; The enterprise trust calculation unit is used to calculate the trust of the enterprise to be evaluated based on the knowledge graph after obtaining the basic information and credit record of the enterprise to be evaluated; The enterprise trust evaluation unit is used to provide a trust expectation value that meets the user enterprise's own trust, evaluate the trustworthiness of the enterprise to be evaluated from the perspective of the expectation value, analyze and calculate the components of the trust, and finally add the enterprise to be evaluated to the trust rating library.

2. The enterprise trust evaluation system based on knowledge graph according to claim 1 is characterized by: The enterprise dictionary building unit includes data collection, keyword extraction and corpus storage; The data collection is used to collect data such as enterprise names and enterprise transaction terms in the enterprise alliance; Keyword extraction is used to extract candidate keywords describing corporate transactions from the collected words; corpus storage is used to use the candidate keywords as a corpus for subsequent deep learning.

3. The enterprise trust evaluation system based on knowledge graph according to claim 1 is characterized by: The knowledge graph construction unit includes extraction of four-element knowledge groups (head entity, relationship, tail entity, time), knowledge reasoning and graph storage; The four-element knowledge group extraction is used to perform name recognition and entity extraction on the text description of enterprises and transactions; Knowledge reasoning is used to establish connections between enterprises in an enterprise alliance where there is an indirect transaction relationship; Graph storage is used to store all quadruple after knowledge reasoning through a graph database.

4. The enterprise trust evaluation system based on knowledge graph according to claim 1 is characterized by: The enterprise trust calculation unit includes direct trust calculation, transaction trust calculation and recommendation trust calculation; The direct trust calculation is used to calculate the trust based on the basic information and credit record of the enterprise to be evaluated. The transaction trust calculation is used to calculate the trust based on the satisfaction between the user enterprise and the enterprise to be evaluated. The recommended trust calculation is used by the user enterprise to calculate the transaction trust between each enterprise in the transaction link with the enterprise to be evaluated.

5. The enterprise trust evaluation system based on knowledge graph according to claim 1 is characterized by: The enterprise trust evaluation unit includes evaluation criteria classification and trust level classification; The evaluation criteria division is used to calculate and divide the evaluation criteria for whether the enterprise is trustworthy based on the historical transaction records of the user enterprise; The trust level classification is used to analyze the calculated trust components and classify the enterprise to be evaluated into the corresponding transaction trust level.

6. A method for judging enterprise trust based on knowledge graph, characterized in that: The following steps are involved: (1) The enterprise dictionary construction unit constructs a special dictionary containing vocabulary describing enterprises and enterprise transactions, providing a corpus for the subsequent knowledge graph construction unit; (2) The knowledge graph construction unit constructs a knowledge graph of enterprise entities and whether there are direct and indirect transaction relationships; (3) After the user enterprise inputs the enterprise to be evaluated, the enterprise trust calculation unit will calculate its three trust components based on the input enterprise to be evaluated, namely direct trust, transaction trust and recommendation trust; (4) The enterprise trust evaluation unit queries the historical transaction records of the user enterprise, obtains the evaluation criteria through the classification algorithm, compares and calculates the trust, classifies the trust level and gives the final result.

7. According to the enterprise trust evaluation method based on knowledge graph according to claim 6, in step (3), a satisfaction penalty factor is added to the calculation of transaction trust, and the calculation is performed according to formula (1) and formula (2). in, B2 represents the transaction trust before adding the satisfaction penalty factor, S i represents the satisfaction of the i-th transaction, I i Indicates the importance of the i-th transaction, T i represents the time weight of the i-th transaction, B deal is the final transaction trust, S sat The number of transactions that indicate satisfaction, S unsat The number of transactions that indicate dissatisfaction, n is the number of enterprises that directly trade with the member enterprise; and the recommendation trust based on transaction trust is calculated according to formula (3) and formula (4): B3=∑t n B3(i,n,j) (4) Among them, B3(i, j) represents the single-path recommendation trust from member enterprise i to enterprise j to be evaluated in the enterprise alliance, depth(i, j-1) represents the link length between member enterprise i and enterprise j to be evaluated that have a direct transaction relationship, B2(i, i+1) represents the transaction trust between member enterprise i and enterprise i+1 with a direct transaction relationship, and the value of m is the number of enterprises with pairwise direct transaction relationships between company i and company j; B3 represents the multi-path recommendation trust, which can be considered to have n single-path recommendation links, and each link is assigned a different time weight and weighted summed, that is, the trust of the enterprise to be evaluated under multiple paths is obtained, and B3(i, n, j) represents the single-path recommendation trust of member enterprise i to enterprise j to be evaluated obtained through the recommendation of n enterprises.

8. The enterprise trust evaluation method based on knowledge graph according to claim 6 is characterized in that: In step (4), the trust level library is divided into three categories: tradable level, pending tradable level and non-tradable level, based on the user enterprise's own evaluation criteria combined with the trust component.

Images