A Process Knowledge Recommendation Method, System, Electronic Device, and Storage Medium

Abstract

The present invention provides a process knowledge recommendation method, system, electronic device and storage medium, belonging to the field of knowledge recommendation. The process knowledge recommendation method includes: obtaining the domain-role vectors of each user and the category vectors of each knowledge item, and calculating the demand values of each user for each knowledge item to obtain a user-knowledge demand matrix; determining an initial user-knowledge score matrix according to the historical knowledge behavior data of the user; performing matrix decomposition and score prediction on the initial user-knowledge score matrix by using a matrix decomposition model and the alternating least squares method to obtain a user-knowledge score prediction matrix; performing weighted summation on the user-knowledge demand matrix and the user-knowledge score prediction matrix to obtain a user-knowledge hybrid matrix; sorting the final scores of each knowledge item in the user-knowledge hybrid matrix, and recommending to the user the knowledge items whose final scores are greater than a set threshold. It can recommend the most appropriate process knowledge to the user in real time, efficiently and accurately.

Description

Technical Field

[0001] The present invention relates to the field of knowledge recommendation, and particularly to a process knowledge recommendation method, system, electronic device and storage medium. Background Art

[0002] In order to solve the problem of lack of effective knowledge recommendation and inheritance in the process design and manufacturing of civil aviation aircraft parts, especially long-cycle and long-process parts, process personnel need to carry out a large amount of repetitive design work based on personal experience. Moreover, due to the difficult extraction of effective knowledge in the process design, low knowledge reuse rate and sharing ability, and poor supply chain collaborative design and manufacturing ability in the process design, it is necessary to integrate new-generation information technologies such as 5G, industrial Internet, and artificial intelligence, focus on the collaborative manufacturing of the large aircraft supply chain, use methods such as artificial intelligence to carry out knowledge-based intelligent process design, break through the digital collaborative core technology of the joint development of the supply chain, achieve a high degree of parallelism of the supply chain, promote the transformation of the new generation of aircraft to the digital design and manufacturing collaborative mode, so as to achieve the purpose of optimizing resources, saving costs, improving quality, and accelerating progress, and better meet the airworthiness management requirements for the review, appraisal, supervision, and management of all links in the whole life cycle of civil aviation aircraft parts.

[0003] Knowledge recommendation is to integrate knowledge service and information push technologies. Knowledge recommendation needs to center around the personalized needs of users, analyze user interests, and provide timely and proactive knowledge services to users based on data such as user personal information or historical behaviors. Its essence is to match a huge knowledge base according to the knowledge needs of users, screen out the most appropriate knowledge, and actively push it to designers, greatly reducing the time spent by designers in obtaining the required knowledge. Knowledge recommendation can accelerate knowledge transfer, eliminate the phenomenon of knowledge asymmetry, and promote knowledge application and innovation.

[0004] Currently, there have been many related studies on recommendation algorithms. The most main recommendation methods are content-based recommendation algorithms, rule-based recommendation algorithms, and collaborative filtering knowledge recommendation algorithms. The content-based recommendation algorithm is to obtain user information and use the matching degree between the information and user interests for recommendation. The rule-based recommendation technology generates association rules by analyzing user operation records to find the correlation within the information set and user set, and recommends relevant information to users based on this rule. The collaborative filtering technology is the most widely recognized and adopted recommendation technology at present. This technology finds the set of similar users of the current user by analyzing user ratings and makes recommendations based on the rating behaviors of similar users.

[0005] In the initial stage of the application of the process knowledge management system, the knowledge behavior of users is too little, and there is not much data available for the recommendation algorithm. At this time, the existing recommendation algorithms do not perform well, and there is a "cold start" problem. Moreover, the number of knowledge entries and users is huge, and the computational complexity of traditional recommendation methods will increase accordingly, and the efficiency will be greatly reduced. The increase in knowledge behavior in a short period of time will basically not cause changes in the recommendation results, and it is impossible to achieve real-time recommendation of information using traditional recommendation methods. Summary of the Invention

[0006] The object of the present invention is to provide a process knowledge recommendation method, system, electronic device and storage medium, which can recommend the most appropriate process knowledge to users in real time, efficiently and accurately.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A process knowledge recommendation method includes:

[0009] Obtain the field to which each user belongs and the role of the user in the field to obtain the field-role vector of each user;

[0010] Obtain the knowledge entries in the knowledge base and the categories of each knowledge entry, and determine the category vector of each knowledge entry; the knowledge base includes a plurality of knowledge entries and the categories of each knowledge entry; the categories of each knowledge entry include a large category and a small category, the large category corresponds to the field of the user, and the small category corresponds to the role of the user;

[0011] According to the field-role vector of each user and the category vector of each knowledge entry, calculate the demand value of each user for each knowledge entry to obtain a user-knowledge demand matrix;

[0012] Obtain the historical knowledge behavior data of each user; the historical knowledge behavior data includes the browsing, liking, collecting and blocking behaviors of the user for each knowledge entry;

[0013] For any user, calculate the initial score of the user for each knowledge entry according to the historical knowledge behavior data of the user;

[0014] According to the initial scores of each user for each knowledge entry, determine an initial user-knowledge score matrix;

[0015] Use a matrix decomposition model and the alternating least squares method to perform matrix decomposition and score prediction on the initial user-knowledge score matrix to obtain a user-knowledge score prediction matrix;

[0016] Weighted sum the user-knowledge demand matrix and the user-knowledge score prediction matrix to obtain a user-knowledge hybrid matrix; the user-knowledge hybrid matrix includes the final scores of each user for each knowledge entry;

[0017] For any user, sort the final scores of the user for each knowledge item in the user-knowledge hybrid matrix, and recommend to the user the knowledge items whose final scores are greater than a set threshold.

[0018] Optionally, the fields to which the user belongs include turning, milling, grinding, and drilling;

[0019] The roles in the turning field, the milling field, the grinding field, and the drilling field all include tool management, equipment management, programming, and process operation;

[0020] The major categories of each knowledge item include turning knowledge, milling knowledge, grinding knowledge, and drilling knowledge;

[0021] The sub-categories under the major category of turning knowledge, the sub-categories under the major category of milling knowledge, the sub-categories under the major category of grinding knowledge, and the sub-categories under the major category of drilling knowledge all include tool knowledge, equipment knowledge, programming knowledge, and process knowledge.

[0022] Optionally, the field-role vector of each user includes a field vector and a role vector; the category vector of each knowledge item includes a major category vector and a sub-category vector.

[0023] Optionally, use the following formula to calculate the demand value of user i for knowledge item j:

[0024]

[0025] where A ij is the demand value of user i for knowledge item j, a i is the field vector of user i, b i is the role vector of user i, c j is the major category vector of knowledge item j, and d j is the sub-category vector of knowledge item j.

[0026] Optionally, use the following formula to calculate the initial score of user i for knowledge item j:

[0027] G ij =(1 - e)(1×f + 2×g + 4×h);

[0028] where G ijThe initial score of user i for knowledge item j, e represents whether user i blocks knowledge item j, e = 1 indicates that user i has blocked knowledge item j, e = 0 indicates that user i has not blocked knowledge item j, f is the number of times user i browses knowledge item j, f ≥ 0, g represents whether user i likes knowledge item j, g = 1 indicates that user i has liked knowledge item j, g = 0 indicates that user i has not liked knowledge item j, h represents whether user i collects knowledge item j, h = 1 indicates that user i has collected knowledge item j, h = 0 indicates that user i has not collected knowledge item j.

[0029] Optionally, the matrix factorization model and the alternating least squares method are used to perform matrix factorization and score prediction on the initial user-knowledge score matrix to obtain a user-knowledge score prediction matrix, specifically including:

[0030] Using a matrix factorization algorithm, the initial user-knowledge score matrix is factorized into two matrices: a first matrix and a second matrix;

[0031] According to the non-empty values in the initial user-knowledge score matrix, the first matrix, and the second matrix, the loss function of the matrix factorization model is determined;

[0032] According to the loss function, the alternating least squares method is used to perform optimization calculations on the first matrix and the second matrix until the loss function converges, and then a first optimal matrix and a second optimal matrix are obtained;

[0033] Multiply the first optimal matrix and the second optimal matrix to obtain an initial score prediction matrix; the initial score prediction matrix includes the predicted score values of each user for each knowledge item;

[0034] Normalize each predicted score value in the initial score prediction matrix to obtain a user-knowledge score prediction matrix.

[0035] Optionally, the following formula is used to perform weighted summation on the user-knowledge demand matrix and the user-knowledge score prediction matrix to obtain a user-knowledge hybrid matrix:

[0036]

[0037] Where R is the user-knowledge hybrid matrix, n is the number of non-empty values in the initial user-knowledge score matrix, x is the number of users, y is the number of knowledge items, A is the user-knowledge demand matrix, and B is the user-knowledge score prediction matrix.

[0038] To achieve the above object, the present invention also provides the following solutions:

[0039] A process knowledge recommendation system, including:

[0040] A domain role acquisition unit, configured to acquire the domain to which each user belongs and the role of the user in the domain to obtain the domain-role vectors of each user;

[0041] A knowledge category acquisition unit, configured to acquire the knowledge entries in the knowledge base and the categories of each knowledge entry, and determine the category vectors of each knowledge entry; the knowledge base includes a plurality of knowledge entries and the categories of each knowledge entry; the categories of each knowledge entry include a major category and a minor category, the major category corresponds to the domain of the user, and the minor category corresponds to the role of the user;

[0042] A demand calculation unit, connected to the domain role acquisition unit and the knowledge category acquisition unit respectively, configured to calculate the demand values of each user for each knowledge entry according to the domain-role vectors of each user and the category vectors of each knowledge entry, and obtain a user-knowledge demand matrix;

[0043] A historical behavior acquisition unit, configured to acquire the historical knowledge behavior data of each user; the historical knowledge behavior data includes the browsing, liking, collecting, and blocking behaviors of the user for each knowledge entry;

[0044] An initial score calculation unit, connected to the historical behavior acquisition unit, configured to calculate, for any user, the initial scores of the user for each knowledge entry according to the historical knowledge behavior data of the user;

[0045] An initial score matrix determination unit, connected to the initial score calculation unit, configured to determine an initial user-knowledge score matrix according to the initial scores of each user for each knowledge entry;

[0046] A score prediction unit, connected to the initial score matrix determination unit, configured to perform matrix decomposition and score prediction on the initial user-knowledge score matrix by using a matrix decomposition model and the alternating least squares method to obtain a user-knowledge score prediction matrix;

[0047] A final score calculation unit, connected to the demand calculation unit and the score prediction unit respectively, configured to perform weighted summation on the user-knowledge demand matrix and the user-knowledge score prediction matrix to obtain a user-knowledge hybrid matrix; the user-knowledge hybrid matrix includes the final scores of each user for each knowledge entry;

[0048] A recommendation unit, connected to the final score calculation unit, configured to, for any user, sort the final scores of the user for each knowledge entry in the user-knowledge hybrid matrix, and recommend to the user the knowledge entries whose final scores are greater than a set threshold.

[0049] To achieve the above object, the present invention also provides the following solutions:

[0050] An electronic device includes a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned process knowledge recommendation method.

[0051] To achieve the above object, the present invention also provides the following solutions:

[0052] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned process knowledge recommendation method.

[0053] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention: First, obtain the field to which each user belongs and the role of the user in the field, obtain the knowledge entries in the knowledge base and the categories of each knowledge entry, and obtain the historical knowledge behavior data of each user; then, according to the field-role vector of each user and the category vector of each knowledge entry, calculate the demand value of each user for each knowledge entry, and determine the initial user-knowledge score matrix according to the historical knowledge behavior data; then, use the matrix decomposition model and the alternating least squares method to perform matrix decomposition and score prediction on the initial user-knowledge score matrix to obtain the user-knowledge score prediction matrix, and perform weighted summation on the user-knowledge demand matrix and the user-knowledge score prediction matrix to obtain the user-knowledge hybrid matrix. The method of matrix decomposition reduces the amount of calculation; finally, according to the final score ranking of each knowledge entry in the user-knowledge hybrid matrix, recommend the knowledge entries with the final score greater than the set threshold to the user in real time, and can efficiently and accurately recommend the most appropriate process knowledge to the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0055] Figure 1 It is a flowchart of the process knowledge recommendation method of the present invention;

[0056] Figure 2 It is a schematic diagram of the process knowledge recommendation process;

[0057] Figure 3 It is a user field-role classification diagram;

[0058] Figure 4 It is a knowledge entry classification diagram;

[0059] Figure 5 It is a schematic diagram of the initial user-knowledge score matrix;

[0060] Figure 6 This is a schematic diagram of the modules of the process knowledge recommendation system of the present invention.

[0061] Symbol Explanation:

[0062] Domain role acquisition unit - 1, knowledge category acquisition unit - 2, demand calculation unit - 3, historical behavior acquisition unit - 4, initial score calculation unit - 5, initial score matrix determination unit - 6, score prediction unit - 7, final score calculation unit - 8, recommendation unit - 9. Detailed Implementation Manner

[0063] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0064] The purpose of the present invention is to provide a process knowledge recommendation method, system, electronic device and storage medium. By using a matrix decomposition model and the alternating least squares method to perform matrix decomposition and score prediction on the initial user - knowledge score matrix, the amount of calculation is reduced, and the most appropriate process knowledge can be efficiently and accurately recommended to users.

[0065] To make the above - mentioned objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0066] The present invention is applied to a specific process knowledge field, such as the knowledge recommendation process for designers in aircraft part process design.

[0067] Embodiment 1

[0068] As Figure 1 and Figure 2 shown, the process knowledge recommendation method provided in this embodiment includes:

[0069] S1: Obtain the field to which each user belongs and the role of the user in the field to obtain the field - role vector of each user. Specifically, obtain the field and role of the user from the existing user personal information. The field - role vector of each user includes a field vector and a role vector. The field - role vector of user i is represented by (a i , b i ), where a i represents the field vector of user i, and b i represents the role vector of user i.

[0070] In this embodiment, asFigure 3 As shown in the figure, the user's fields of expertise include turning, milling, grinding, and drilling. The roles in the turning field, milling field, grinding field, and drilling field all include tool management, equipment management, programming, and process operation.

[0071] As a specific implementation, domain vectors a i and role vectors b i are generated respectively according to the user's domain and role. The user's domain is divided into four domains, so the domain vector is a four-dimensional vector. There are also four roles within the domain, so the role vector is also a four-dimensional vector. If the user is responsible for a certain role within a certain domain, the domain dimension and role dimension coordinates corresponding to the user's domain vector and role vector are 1, and the rest are 0. For example, if user i is a programmer responsible for turning, then the user's domain vector a i is (1, 0, 0, 0), and the role vector b i is (0, 0, 1, 0). The domain-role vector of this user is (1, 0, 0, 0, 0, 0, 1, 0).

[0072] S2: Obtain the knowledge entries in the knowledge base and the categories of each knowledge entry, and determine the category vectors of each knowledge entry. The knowledge base includes multiple knowledge entries and the categories of each knowledge entry. The categories of each knowledge entry include a major category and a minor category. The major category corresponds to the user's domain, and the minor category corresponds to the user's role. The category vectors of each knowledge entry include a major category vector and a minor category vector. The category vector of knowledge entry j is represented by (c j , d j ), where c j represents the major category vector of knowledge entry j, and d j represents the minor category vector of knowledge entry j.

[0073] In this embodiment, as Figure 4 shown, the major categories of each knowledge entry include turning knowledge, milling knowledge, grinding knowledge, and drilling knowledge. The minor categories under the major category of turning knowledge, the minor categories under the major category of milling knowledge, the minor categories under the major category of grinding knowledge, and the minor categories under the major category of drilling knowledge all include tool knowledge, equipment knowledge, programming knowledge, and process knowledge.

[0074] As a specific implementation, the major category vector c j and the minor category vector d j of the knowledge entry j are generated respectively according to the category labels of the knowledge entry j. Process knowledge is a certain minor category of knowledge within a certain major category of process, so the domain dimension and role dimension coordinates corresponding to the major category vector and the minor category vector are 1, and the rest are 0. For example, if knowledge entry j is the tool knowledge of milling, then the major category vector c j of this knowledge entry is (0, 1, 0, 0), and the minor category vector d jis (1, 0, 0, 0), and the category vector is (0, 1, 0, 0, 1, 0, 0, 0).

[0075] S3: According to the domain-role vectors of each user and the category vectors of each knowledge item, calculate the demand values of each user for each knowledge item to obtain a user-knowledge demand matrix. Specifically, generate an x×y user-knowledge demand matrix based on the demand values of each user for each knowledge item, where x is the number of users and y is the number of knowledge items.

[0076] Specifically, use the following formula to calculate the demand value of user i for knowledge item j:

[0077]

[0078] where A ij is the demand value of user i for knowledge item j, a i is the domain vector of user i, b i is the role vector of user i, c j is the major category vector of knowledge item j, and d j is the minor category vector of knowledge item j.

[0079] S4: Obtain the historical knowledge behavior data of each user. The historical knowledge behavior data includes the browsing, liking, collecting, and blocking (not interested) behaviors of users for each knowledge item. Generally speaking, when a knowledge item is extremely important to a user, the user will collect it; when it is relatively important, the user will like it; when it may be useful, the user will browse it (the more times of browsing, the more useful it is); and when it is determined to be of no help, the user will choose not to be interested. The historical knowledge behavior data of users is the statistical data of the number of times users browse each knowledge item and whether they have liked, collected, or blocked each knowledge item.

[0080] S5: For any user, calculate the initial score of the user for each knowledge item according to the historical knowledge behavior data of the user.

[0081] Specifically, the higher the initial score, the more important the knowledge item is. In terms of importance, collection > like > browse. Therefore, different weights are added to different knowledge behaviors, and blocking is completely negative and conflicts with the above three behaviors. Use the following formula to calculate the initial score of user i for knowledge item j:

[0082] G ij = (1 - e)(1×f + 2×g + 4×h);

[0083] where G ijdenotes the initial score of user i for knowledge item j, e represents whether user i has blocked knowledge item j, where e = 1 indicates that user i has blocked knowledge item j, e = 0 indicates that user i has not blocked knowledge item j, f represents the number of times user i has browsed knowledge item j, f ≥ 0, g represents whether user i has liked knowledge item j, where g = 1 indicates that user i has liked knowledge item j, g = 0 indicates that user i has not liked knowledge item j, and h represents whether user i has collected knowledge item j, where h = 1 indicates that user i has collected knowledge item j, h = 0 indicates that user i has not collected knowledge item j.

[0084] It should be noted that for knowledge items that users have not browsed, liked, collected, or blocked, the initial score of such knowledge items is empty rather than 0.

[0085] S6: Determine the initial user-knowledge score matrix based on the initial scores of each user for each knowledge item. That is, generate an x×y initial user-knowledge score matrix according to the initial scores of each user for each knowledge item.

[0086] S7: Use the matrix factorization model and the alternating least squares method to perform matrix factorization and score prediction on the initial user-knowledge score matrix to obtain the user-knowledge score prediction matrix.

[0087] Furthermore, S7 specifically includes:

[0088] S71: Use the matrix factorization algorithm to factorize the initial user-knowledge score matrix into two matrices: the first matrix and the second matrix. Specifically, split the x×y initial user-knowledge score matrix into the product of an x×k first matrix P and a k×y second matrix Q, where k is a parameter.

[0089] S72: Determine the loss function of the matrix factorization model based on the non-empty values in the initial user-knowledge score matrix, the first matrix, and the second matrix.

[0090] Specifically, the loss function of the matrix factorization model is:

[0091]

[0092] where LOSS(x,y) is the value of the loss function, r ij is the non-empty value in the initial user-knowledge score matrix, that is, the initial score of each user for each knowledge item, p i is the i-th row of the first matrix P, and q j is the j-th column of the second matrix Q. In this embodiment, the i-th row of the first matrix P represents user i, and the j-th column of the second matrix Q represents knowledge item j.

[0093] S73: According to the loss function, the first matrix and the second matrix are optimized and calculated by using the alternating least squares method until the loss function converges, and then the first optimal matrix and the second optimal matrix are obtained.

[0094] In this embodiment, the alternating least squares method used is to first initialize P and Q, take any initial value for Q and fix it, take the partial derivative of the loss function LOSS(x, y) with respect to P, and set this partial derivative to 0 to obtain P; then fix P and take the partial derivative of the loss function with respect to Q, and set this partial derivative to 0 to obtain Q; repeat the above steps until the loss function converges, and P and Q when the loss function converges are the first optimal matrix and the second optimal matrix.

[0095] S74: Multiply the first optimal matrix and the second optimal matrix to obtain an initial score prediction matrix. The initial score prediction matrix includes the predicted score values of each user for each knowledge item.

[0096] S75: Normalize each predicted score value in the initial score prediction matrix to obtain a user-knowledge score prediction matrix. Specifically, each predicted score value in the initial score prediction matrix is divided by the maximum predicted score value in the initial score prediction matrix to achieve normalization.

[0097] S8: Weightedly sum the user-knowledge demand matrix and the user-knowledge score prediction matrix to obtain a user-knowledge hybrid matrix. The user-knowledge hybrid matrix includes the final scores of each user for each knowledge item.

[0098] Specifically, the following formula is used to determine the user-knowledge hybrid matrix:

[0099]

[0100] where R is the user-knowledge hybrid matrix, n is the number of non-empty values in the initial user-knowledge score matrix, x is the number of users, y is the number of knowledge items, A is the user-knowledge demand matrix, and B is the user-knowledge score prediction matrix.

[0101] For example Figure 5 , u represents the user, k represents the knowledge item, there are a total of 5 users and 6 knowledge items, then x is 5, y is 6, and there are a total of 20 non-empty elements in the initial user-knowledge score matrix, that is, n is 20.

[0102] S9: For any user, sort the final scores of the user for each knowledge item in the user-knowledge hybrid matrix, and recommend the knowledge items with the final scores greater than the set threshold to the user.

[0103] The present invention utilizes the personal information of users and the knowledge entries stored by category in the knowledge base to generate the domain-role vectors of users and the category vectors of knowledge entries respectively. By calculating the similarity between the two, the user-knowledge demand values are obtained and a user-knowledge demand matrix is formed. At the same time, the historical knowledge behaviors of users in the past are statistically analyzed to form an initial user-knowledge scoring matrix, and this matrix is decomposed and trained using a matrix decomposition model and the alternating least squares method. The model is used to predict the scores of users for the knowledge that has not been browsed yet, and a user-knowledge scoring prediction matrix is obtained. Finally, through dynamic weight calculation, the user-knowledge demand matrix and the user-knowledge scoring prediction matrix are fused to obtain a user-knowledge hybrid matrix, and the knowledge entries are recommended to users according to the scores in the user-knowledge hybrid matrix. This solves the shortcoming of the cold start of general knowledge recommendation methods, reduces the calculation amount by using the matrix decomposition method, and realizes the real-time recommendation of process knowledge.

[0104] Embodiment 2

[0105] In order to execute the method corresponding to the above Embodiment 1 to achieve the corresponding functions and technical effects, a process knowledge recommendation system is provided below.

[0106] As Figure 6 shown, the process knowledge recommendation system provided in this embodiment includes: a domain role acquisition unit 1, a knowledge category acquisition unit 2, a demand calculation unit 3, a historical behavior acquisition unit 4, an initial scoring calculation unit 5, an initial scoring matrix determination unit 6, a scoring prediction unit 7, a final scoring calculation unit 8, and a recommendation unit 9.

[0107] Among them, the domain role acquisition unit 1 is used to obtain the domain to which each user belongs and the role of the user in the domain to obtain the domain-role vector of each user.

[0108] The knowledge category acquisition unit 2 is used to obtain the knowledge entries in the knowledge base and the categories of each knowledge entry to determine the category vector of each knowledge entry. The knowledge base includes a plurality of knowledge entries and the categories of each knowledge entry. The categories of each knowledge entry include a large category and a small category, the large category corresponds to the domain of the user, and the small category corresponds to the role of the user.

[0109] The demand calculation unit 3 is respectively connected to the domain role acquisition unit 1 and the knowledge category acquisition unit 2. The demand calculation unit 3 is used to calculate the demand value of each user for each knowledge entry according to the domain-role vector of each user and the category vector of each knowledge entry to obtain a user-knowledge demand matrix.

[0110] The historical behavior acquisition unit 4 is used to obtain the historical knowledge behavior data of each user. The historical knowledge behavior data includes the browsing, liking, collecting, and blocking behaviors of users for each knowledge entry.

[0111] The initial score calculation unit 5 is connected to the historical behavior acquisition unit 4. The initial score calculation unit 5 is configured to calculate, for any user, the initial scores of the user for each knowledge item according to the historical knowledge behavior data of the user.

[0112] The initial score matrix determination unit 6 is connected to the initial score calculation unit 5. The initial score matrix determination unit 6 is configured to determine an initial user-knowledge score matrix according to the initial scores of each user for each knowledge item.

[0113] The score prediction unit 7 is connected to the initial score matrix determination unit 6. The score prediction unit 7 is configured to perform matrix decomposition and score prediction on the initial user-knowledge score matrix by using a matrix decomposition model and the alternating least squares method to obtain a user-knowledge score prediction matrix.

[0114] The final score calculation unit 8 is respectively connected to the requirement calculation unit 3 and the score prediction unit 7. The final score calculation unit 8 is configured to perform weighted summation of the user-knowledge requirement matrix and the user-knowledge score prediction matrix to obtain a user-knowledge hybrid matrix. The user-knowledge hybrid matrix includes the final scores of each user for each knowledge item.

[0115] The recommendation unit 9 is connected to the final score calculation unit 8. The recommendation unit 9 is configured to, for any user, sort the final scores of the user for each knowledge item in the user-knowledge hybrid matrix, and recommend to the user the knowledge items whose final scores are greater than a set threshold.

[0116] Embodiment III

[0117] This embodiment provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the process knowledge recommendation method of Embodiment I.

[0118] Optionally, the above electronic device may be a server.

[0119] In addition, an embodiment of the present invention further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the process knowledge recommendation method of Embodiment I is implemented.

[0120] In this specification, the various embodiments are described in a progressive manner. The key points of each embodiment are the differences from other embodiments. The same or similar parts among the various embodiments may be referred to each other.

[0121] In this article, specific examples are used to illustrate the principles and implementation modes of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation modes and application scopes. In summary, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. A process knowledge recommendation method, characterized in that The described process knowledge recommendation method includes: Obtain the field to which each user belongs and the role of the user in the field to obtain the field-role vector of each user; Obtain the knowledge entries in the knowledge base and the categories of each knowledge entry, and determine the category vector of each knowledge entry; the knowledge base includes multiple knowledge entries and the categories of each knowledge entry; the categories of each knowledge entry include a major category and a minor category, the major category corresponds to the field of the user, and the minor category corresponds to the role of the user; According to the field-role vector of each user and the category vector of each knowledge entry, calculate the demand value of each user for each knowledge entry to obtain a user-knowledge demand matrix; Obtain the historical knowledge behavior data of each user; the historical knowledge behavior data includes the browsing, liking, collecting, and blocking behaviors of the user for each knowledge entry; For any user, calculate the initial score of the user for each knowledge entry according to the historical knowledge behavior data of the user; According to the initial scores of each user for each knowledge entry, determine the initial user-knowledge score matrix; Use a matrix factorization model and the alternating least squares method to perform matrix factorization and score prediction on the initial user-knowledge score matrix to obtain a user-knowledge score prediction matrix; Weightedly sum the user-knowledge demand matrix and the user-knowledge score prediction matrix to obtain a user-knowledge hybrid matrix; the user-knowledge hybrid matrix includes the final scores of each user for each knowledge entry; For any user, sort the final scores of the user for each knowledge entry in the user-knowledge hybrid matrix, and recommend knowledge entries with a final score greater than a set threshold to the user.

2. The process knowledge recommendation method according to claim 1, characterized in that The fields to which users belong include turning, milling, grinding, and drilling; The roles in the turning field, the milling field, the grinding field, and the drilling field all include tool management, equipment management, programming, and process operation; The major categories of each knowledge entry include turning knowledge, milling knowledge, grinding knowledge, and drilling knowledge; The minor categories under the major category of turning knowledge, the minor categories under the major category of milling knowledge, the minor categories under the major category of grinding knowledge, and the minor categories under the major category of drilling knowledge all include tool knowledge, equipment knowledge, programming knowledge, and process knowledge.

3. The process knowledge recommendation method according to claim 1, wherein The field-role vector of each user includes a field vector and a role vector; the category vector of each knowledge entry includes a major category vector and a minor category vector.

4. The process knowledge recommendation method according to claim 3, wherein Use the following formula to calculate the demand value of user i for knowledge entry j: Among them, A ij is the demand value of user i for knowledge item j, a i is the domain vector of user i, b i is the role vector of user i, c j is the major category vector of knowledge item j, d j is the minor category vector of knowledge item j.

5. The process knowledge recommendation method according to claim 1, characterized in that Use the following formula to calculate the initial score of user i for knowledge entry j: G ij = (1 - e)(1×f + 2×g + 4×h); Among them, G ij is the initial score of user i for knowledge item j. e represents whether user i has blocked knowledge item j. e = 1 means that user i has blocked knowledge item j, and e = 0 means that user i has not blocked knowledge item j. f is the number of times user i has browsed knowledge item j, f ≥ 0. g represents whether user i has liked knowledge item j. g = 1 means that user i has liked knowledge item j, and g = 0 means that user i has not liked knowledge item j. h represents whether user i has collected knowledge item j. h = 1 means that user i has collected knowledge item j, and h = 0 means that user i has not collected knowledge item j.

6. The process knowledge recommendation method according to claim 1, wherein The use of a matrix factorization model and the alternating least squares method to perform matrix factorization and score prediction on the initial user-knowledge score matrix to obtain a user-knowledge score prediction matrix specifically includes: Use a matrix factorization algorithm to decompose the initial user-knowledge score matrix into two matrices: a first matrix and a second matrix; According to the non-empty values in the initial user-knowledge score matrix, the first matrix, and the second matrix, determine the loss function of the matrix factorization model; According to the loss function, use the alternating least squares method to perform optimization calculations on the first matrix and the second matrix until the loss function converges, and then obtain a first optimal matrix and a second optimal matrix; Multiply the first optimal matrix and the second optimal matrix to obtain an initial score prediction matrix; the initial score prediction matrix includes the predicted score values of each user for each knowledge item. Normalize each predicted score value in the initial score prediction matrix to obtain a user-knowledge score prediction matrix.

7. The process knowledge recommendation method according to claim 1, wherein Use the following formula to perform a weighted sum of the user-knowledge demand matrix and the user-knowledge score prediction matrix to obtain a user-knowledge hybrid matrix: Where R is the user-knowledge hybrid matrix, n is the number of non-empty values in the initial user-knowledge score matrix, x is the number of users, y is the number of knowledge items, A is the user-knowledge demand matrix, and B is the user-knowledge score prediction matrix.

8. A process knowledge recommendation system, characterized in that, The process knowledge recommendation system includes: A domain role acquisition unit, configured to acquire the domain to which each user belongs and the role of the user in the belonging domain, and obtain the domain-role vector of each user. A knowledge category acquisition unit, configured to acquire the knowledge items in the knowledge base and the categories of each knowledge item, and determine the category vector of each knowledge item; the knowledge base includes a plurality of knowledge items and the categories of each knowledge item; the categories of each knowledge item include a major category and a minor category, the major category corresponds to the domain of the user, and the minor category corresponds to the role of the user. A demand calculation unit, respectively connected to the domain role acquisition unit and the knowledge category acquisition unit, configured to calculate the demand value of each user for each knowledge item according to the domain-role vector of each user and the category vector of each knowledge item, and obtain a user-knowledge demand matrix. A historical behavior acquisition unit, configured to acquire the historical knowledge behavior data of each user; the historical knowledge behavior data includes the browsing, liking, collecting, and blocking behaviors of the user for each knowledge item. An initial score calculation unit, connected to the historical behavior acquisition unit, configured to calculate, for any user, the initial score of the user for each knowledge item according to the historical knowledge behavior data of the user. An initial score matrix determination unit, connected to the initial score calculation unit, configured to determine an initial user-knowledge score matrix according to the initial scores of each user for each knowledge item. A score prediction unit, connected to the initial score matrix determination unit, configured to perform matrix decomposition and score prediction on the initial user-knowledge score matrix by using a matrix decomposition model and the alternating least squares method, and obtain a user-knowledge score prediction matrix. A final score calculation unit, respectively connected to the demand calculation unit and the score prediction unit, configured to perform a weighted sum of the user-knowledge demand matrix and the user-knowledge score prediction matrix to obtain a user-knowledge hybrid matrix; the user-knowledge hybrid matrix includes the final scores of each user for each knowledge item. A recommendation unit, connected to the final score calculation unit, configured to, for any user, sort the final scores of the user for each knowledge item in the user-knowledge hybrid matrix, and recommend to the user the knowledge items whose final scores are greater than a set threshold.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the process knowledge recommendation method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the process knowledge recommendation method according to any one of claims 1 to 7.

Images