A bidirectional matching method, device, medium and equipment based on canonical correlation analysis

Abstract

The present invention relates to the field of information matching technology, and in particular to a two-way matching method, device, medium and equipment based on canonical correlation analysis. The text data of experts and projects are converted into numerical data through a cosine similarity algorithm, and multiple feature numerical values ​​are constructed to form the matching similarity between experts and projects in multiple aspects, and the similarity is obtained as a calculated value with a value range of [0,1]. The corresponding coefficient weight can be obtained through a canonical correlation analysis algorithm, so that all cosine similarities are integrated into a numerical value, and the numerical value can be calculated by the status of the expert at the same time. The finally calculated numerical value is the final recommendation score of the expert matching.

Description

Technical Field

[0001] The present invention relates to the field of information matching technology, and in particular to a two-way matching method, device, medium and equipment based on typical correlation analysis. Background Art

[0002] With the continuous development of human society, the number of various projects to be built is also increasing rapidly, which requires a large number of relevant talents to participate in the construction of the projects. Before the construction of the project, there is a lot of preparation work, such as confirming the construction type of the project and selecting suitable experts to participate in the project. But sometimes there is a situation that the project is not determined and there are only a few text descriptions. In this case, it is difficult to select the corresponding experts to participate in the project. One of the most common methods is to select experts who have a certain degree of overlap with the project description to participate in the project, but this method has certain defects, and it is impossible to calculate which of these experts is most suitable to participate in the project.

[0003] Therefore, there is an urgent need for a suitable expert recommendation method that can quickly calculate the matching degree between experts and projects based on the existing data, and select experts who can ensure the smooth progress of the project to participate in the project to the greatest extent possible. Summary of the invention

[0004] In order to solve the deficiencies of the prior art, the present invention provides a two-way matching method based on canonical correlation analysis, comprising the following steps:

[0005] S100, extracting keywords of the target project, determining the target field and target subject classification of the target project, and determining the number of target fields and target subjects;

[0006] S200, obtaining historical projects, determining project characteristic parameters of the historical projects and expert characteristic parameters of corresponding experts participating in the historical projects according to the target field and the target discipline, combining the project characteristic parameters to generate a plurality of historical project characteristic vectors, and generating a plurality of first expert characteristic vectors according to the expert characteristic parameter combination of the corresponding historical project characteristic vectors;

[0007] S300, substituting a number of historical project feature vectors and a number of corresponding first expert feature vectors into a cosine similarity algorithm to calculate a number of expert project similarity values;

[0008] S400, calculating the time correlation of experts' participation in historical projects to obtain a time correlation value;

[0009] S500, multiplying a plurality of expert project similarity values ​​by time correlation values ​​to obtain a plurality of expert project similarity value vectors, organizing the plurality of expert project similarity value vectors into a one-dimensional expert project feature array, and using the one-dimensional expert project feature array of historical projects to establish a two-dimensional expert project feature array related to the target project;

[0010] S600, organizing the expert project similarity value vectors corresponding to the historical projects into a one-dimensional expert feature array according to the first expert feature vector of the expert, and using the one-dimensional expert feature array to organize the two-dimensional expert feature array corresponding to the two-dimensional expert project feature array;

[0011] S700, the two-dimensional expert project feature array and the two-dimensional expert feature array are substituted into the canonical correlation analysis model to perform canonical correlation analysis, a recommendation score is calculated according to the result of the canonical correlation analysis, and a two-way matching is performed between the target project and the expert according to the recommendation score.

[0012] Furthermore, the historical item feature vector is:

[0013]

[0014] in, is the number of historical project feature vectors of experts participating in historical projects, is the number of item characteristic parameters, is the project characteristic parameter, For the Project characteristic parameters, is the historical item feature vector, No. historical item feature vectors;

[0015] The first expert feature vector is:

[0016]

[0017] in, is the number of the first expert feature vectors of the experts who participated in the historical projects, and the number of feature vectors of the historical projects and the first expert feature vector The values ​​are equal; is the number of expert feature parameters, is the expert feature parameter, is the t-th expert feature parameter, is the first expert feature vector, The i-th first expert feature vector;

[0018] The calculation formula of expert project similarity value is:

[0019]

[0020] in, is the expert item similarity value, For the Expert project similarity values.

[0021] Furthermore, the step of calculating the time correlation value in S400 is:

[0022] S410, calculating the interval year based on the date when the expert participated in the historical project and the current date when the target project is being carried out;

[0023] The calculation formula for the interval year is:

[0024]

[0025] in, For the interval year, The current date for the target project, Dates of involvement of the expert in the historical project;

[0026] S420. Calculate the time correlation value of the historical project according to the interval year. The calculation formula of the time correlation value is:

[0027]

[0028] in, is the number of experts, is the number of experts involved in different historical projects, For the jth expert to participate in The time correlation value of each historical item.

[0029] Furthermore, the one-dimensional expert item feature array is:

[0030]

[0031] in, is the number of experts who participated in History Project Expert item similarity values, is a one-dimensional array of expert item features, For the jth expert to participate in One-dimensional expert item feature array of historical items;

[0032] The two-dimensional expert item feature array is:

[0033] ,in, is a two-dimensional array of expert item features.

[0034] Furthermore, the one-dimensional expert feature array is:

[0035]

[0036] in, For the Experts participated in The first historical project expert feature parameters, For the Experts participated in A one-dimensional array of expert features for each historical item.

[0037] The two-dimensional expert feature array is:

[0038] ,in, is a two-dimensional expert feature array.

[0039] Furthermore, the specific steps of S700 are:

[0040] S710, using the two-dimensional expert project feature array and the two-dimensional expert feature array to construct a covariance matrix and substitute it into the canonical correlation analysis model to obtain the canonical correlation coefficient, the expert feature canonical correlation coefficient, and the project similarity canonical correlation coefficient;

[0041] S720, establishing a second expert feature vector;

[0042] S730, multiplying the canonical correlation coefficient, the expert characteristic canonical correlation coefficient corresponding to each expert, and the second expert characteristic vector to obtain the recommendation score of each expert; the calculation formula is:

[0043]

[0044] in, is the typical correlation coefficient, is the expert characteristic canonical correlation coefficient, is the second expert feature vector.

[0045] Furthermore, the first expert feature vector and the one-dimensional expert feature array are composed of expert feature parameters before the date when the expert joined the historical project, and the second expert feature vector is composed of various expert feature parameters before the current date when the expert is performing the target project.

[0046] The present invention provides a bidirectional matching device based on canonical correlation analysis, comprising:

[0047] A keyword extraction module is used to extract keywords of the target project, determine the target field and target subject classification of the target project, and determine the number of target fields and target subjects;

[0048] A historical project module is used to obtain historical projects, determine the project characteristic parameters of the historical projects and the expert characteristic parameters of the corresponding experts participating in the historical projects according to the target field and the target discipline, generate a number of historical project characteristic vectors after combining the project characteristic parameters, generate a number of first expert characteristic vectors according to the expert characteristic parameter combination of the corresponding historical project characteristic vectors, and assign characteristic values ​​to the historical project characteristic vectors and the first expert characteristic vectors;

[0049] An expert project similarity value calculation module, used to calculate a plurality of expert project similarity values ​​using the historical project feature vector and the first expert feature vector;

[0050] The time correlation analysis module is used to calculate the time correlation of experts’ participation in historical projects and obtain the time correlation value;

[0051] A project similarity array establishment module is used to obtain several expert project similarity value vectors by multiplying several expert project similarity values ​​of historical projects by time correlation values, arranging several expert project similarity value vectors into a one-dimensional expert project feature array, and using the one-dimensional expert project feature array of historical projects to establish a two-dimensional expert project feature array related to the target project;

[0052] An expert feature array establishment module is used to organize the expert project similarity value vectors corresponding to the historical projects into a one-dimensional expert feature array according to the first expert feature vector of the expert, and use the one-dimensional expert feature array to organize the two-dimensional expert feature array corresponding to the two-dimensional expert project feature array;

[0053] The matching module is used to substitute the two-dimensional expert project feature array and the two-dimensional expert feature array into the canonical correlation analysis model for canonical correlation analysis, calculate the recommendation score according to the result of the canonical correlation analysis, and perform bidirectional matching between the target project and the expert according to the recommendation score.

[0054] The present invention also provides a computer-readable storage medium, which stores computer instructions. When the computer is executed by a processor, the computer implements the two-way matching method based on typical correlation analysis as in the above embodiment.

[0055] The present invention also provides an electronic device, comprising at least one processor and a memory communicatively connected to the processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the processor performs a two-way matching method based on typical correlation analysis as in the above-mentioned embodiment.

[0056] Compared with the prior art, the bidirectional matching method based on canonical correlation analysis provided by the present invention solves the difficulty of converting text data into numerical data through the cosine similarity algorithm, and forms the matching similarity between experts and projects in multiple aspects by constructing multiple feature values. However, since the similarity is a plurality of calculated values ​​with a value range of [0,1], it is impossible to calculate the weight value of each calculated value. Then, the corresponding coefficient weight can be obtained through the canonical correlation analysis algorithm, thereby integrating all cosine similarities into a numerical value, which can also be calculated by the status of the expert. Therefore, the final calculated value is the final expert matching recommendation score. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 A flowchart of a bidirectional matching method based on canonical correlation analysis provided in Embodiment 1 of the present invention;

[0059] Figure 2 A flowchart of calculating a time correlation value provided in the first embodiment of the present invention;

[0060] Figure 3 A flowchart of calculating a recommendation score provided in the first embodiment of the present invention;

[0061] Figure 4 A structural block diagram of a bidirectional matching device based on canonical correlation analysis provided in Embodiment 3 of the present invention;

[0062] Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] In the description of the present invention, it should be noted that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0065] The present invention provides a two-way matching method based on canonical correlation analysis, such as Figure 1 As shown, the following steps are included:

[0066] S100, extracting keywords of the target project, determining the target field and target subject classification of the target project, and determining the number of target fields and target subjects.

[0067] Specifically, taking the project selection of scientific research and policy-making as an example, firstly, by analyzing the description of the target project and extracting the keywords of the target project, we can obtain the target field a and target subject classification b corresponding to the project selection. Correspondingly, a=1 field and b=2 subjects. It should be noted that the field and subject are not limited to 1, but can also be multiple, and the number is determined according to the target project information.

[0068] S200, obtaining historical projects, determining project characteristic parameters of the historical projects and expert characteristic parameters of corresponding experts participating in the historical projects according to the target field and target discipline, assigning characteristic values ​​to the project characteristic parameters and the expert characteristic parameters, combining the project characteristic parameters to generate several historical project characteristic vectors, and corresponding to the historical project characteristic vectors generating several first expert characteristic vectors according to the expert characteristic parameter combination.

[0069] Specifically, by querying the database storing the project participation data, historical projects that have intersections with field a and subject classification b, as well as expert data participating in historical projects in the expert knowledge base, are obtained. Based on the historical project data, project feature parameters and expert feature parameters of historical projects and experts based on target fields and target subjects are constructed. If the project feature parameters corresponding to the historical projects and the expert feature parameters corresponding to the experts have the corresponding target fields or target subject classifications, the corresponding project feature parameters and expert feature parameters have values ​​of 1, otherwise they are 0.

[0070] In one embodiment, the project feature parameters are aggregated to form a historical project feature vector, which is:

[0071]

[0072] in, is the number of historical project feature vectors of experts participating in historical projects, is the number of item characteristic parameters, is the project characteristic parameter, For the Project characteristic parameters, is the historical item feature vector, For the The feature vector of historical items.

[0073] The expert feature parameters are gathered to form the first expert feature vector, which is:

[0074] in, is the number of the first expert feature vectors of the experts who participated in the historical projects, and the number of feature vectors of the historical projects and the first expert feature vector The values ​​are equal; is the number of expert feature parameters, is the expert feature parameter, For the expert feature parameters, is the first expert feature vector, For the First expert feature vectors;

[0075] S300 , substituting the historical project feature vector and the first expert feature vector into a cosine similarity algorithm to calculate several expert project similarity values.

[0076] Specifically, when the historical project feature vector and the first expert feature vector have been constructed, the expert project similarity value can be obtained by the cosine similarity algorithm. The calculation formula of the expert project similarity value is:

[0077]

[0078] in, For the The expert project similarity value between the feature vector of the historical project and the feature vector of the first expert. A project can have multiple experts, and an expert can also participate in multiple projects. The expert project similarity value here is The project feature vector and the first expert feature vector of the same historical project of the same expert are calculated. By using the cosine similarity algorithm to calculate the historical project feature vector and the first expert feature vector, the historical project and expert data are merged to facilitate the later matching, taking into account the factors of both historical projects and experts for matching, and the matching accuracy is higher.

[0079] Specifically, in the actual calculation process, multiple project feature parameters related to the project and expert feature parameters related to the expert are collected. The project feature parameters can be combined in multiple different ways from different angles to form multiple historical project feature vectors; the expert feature parameters can be combined in multiple different ways to form multiple first expert feature vectors. The first expert feature vector corresponds to the combination of the historical project feature vector.

[0080] In one embodiment, the first historical project feature vector and the first first expert feature vector can construct a feature vector of the target project based on the field and discipline according to the data of a related historical project. If the historical project has the corresponding field or discipline classification, the feature value corresponding to the historical project feature vector and the first expert feature vector is 1, otherwise it is 0. The project feature parameters can be whether a historical project meets the field of the target project, whether a project meets the first discipline of the target project, and whether a project meets the second discipline of the target project.

[0081]

[0082] in, =1, It is determined based on the selected project characteristic parameters.

[0083] Furthermore, based on the experts who participated in the relevant history project, expert feature parameters based on the target field and target discipline are constructed and aggregated into a first expert feature vector. If the expert is familiar with the corresponding target field or target discipline, the expert feature parameter is 1, otherwise it is 0. The expert feature parameter can be whether the experts who participated in a certain history project are scholars in the target field, whether the experts who participated in a certain history project are scholars in the first target discipline, and whether the experts who participated in a certain history project are scholars in the second target discipline.

[0084]

[0085] in, =1, It is determined based on the selected expert feature parameters.

[0086] In the constructed history project feature vector and the first expert feature vector In the case of , the expert item similarity value can be obtained by the cosine similarity algorithm .

[0087]

[0088] In one embodiment, the second historical project feature vector and the second first expert feature vector can be obtained based on the weight values ​​of each target field and target subject classification in the historical project data. If there is no weight value, the default value is 1. and Multiplying them together can construct the second historical project feature vector ;at this time =2.

[0089] )

[0090] Determine the expert feature parameters according to the project feature parameters of the second historical project feature vector, so as to generate the second first expert feature vector by combining the expert feature parameters If one has experience in participating in projects in the target field or discipline, the characteristic value of the corresponding dimension is the total number of projects participated in, otherwise it is 0.

[0091]

[0092] After constructing the second historical project feature vector and the first expert feature vector In the case of , the second expert item similarity value can be obtained by the cosine similarity algorithm .

[0093]

[0094] Preferably, the number of projects, number of papers published, number of paper citations, etc. based on the target project level classification can also be constructed to construct other types of historical project feature vectors. and the first expert feature vector , thereby calculating the expert project similarity value .

[0095]

[0096] It should be noted that the project characteristic parameters and expert characteristic parameters can be specifically determined based on the target project and the information of the collected historical projects, and the examples given here are not complete.

[0097] S400: Calculate the time correlation of historical items to obtain a time correlation value.

[0098] Specifically, Figure 2 As shown, the steps of calculating the time correlation value in S400 are:

[0099] S410, calculating the interval year based on the date when the expert participated in the historical project and the current date when the target project is being carried out;

[0100] The calculation formula for the interval year is:

[0101]

[0102] in, For the interval year, is the current date of the target project, Dates on which the expert became involved in the historical project.

[0103] S420, calculating the time correlation value of the historical project according to the interval year;

[0104] The time correlation value is calculated as: ,in, is the number of experts, is the number of experts involved in different historical projects, For the Experts participated in The time correlation value of each historical item.

[0105] Specifically, a historical project may be completed by multiple experts, and an expert may also participate in multiple historical projects. Here, the time correlation value of each expert's participation in each project is calculated, and the time when the expert participated in the historical project and the current time of the target project are used for calculation. The time factor is taken into account to make the matching result more accurate. Preferably, the time when the expert participated in the historical project refers to the earliest time when the expert participated in the historical project; the current time when the target project is carried out is based on the start date of the target project. It should be noted that it is not limited to the above dates, and can also be determined specifically according to actual conditions.

[0106] S500, multiplying several expert project similarity values ​​of historical projects by time correlation values ​​to obtain several expert project similarity value vectors, organizing the several expert project similarity value vectors into a one-dimensional expert project feature array, and using the one-dimensional expert project feature array of historical projects to establish a two-dimensional expert project feature array related to the target project.

[0107] Specifically, the calculation formula of the expert item similarity value vector is:

[0108]

[0109] in, For the Experts participated in Time correlation value of historical items, For the Experts participated in History Project Expert item similarity values, is the expert-item similarity value vector, For the Experts participated in History Project Expert item similarity value variables.

[0110] The one-dimensional expert item feature array is:

[0111]

[0112] in, is a one-dimensional array of expert item features, For the Experts participated in One-dimensional expert item feature array of historical items;

[0113] The two-dimensional expert item feature array is:

[0114] ,in, is a two-dimensional array of expert item features.

[0115] S600 , organizing the expert project similarity value vectors corresponding to the historical projects into a one-dimensional expert feature array according to the first expert feature vector of the expert, and using the one-dimensional expert feature array to organize the two-dimensional expert feature array corresponding to the two-dimensional expert project feature array.

[0116] Specifically, the one-dimensional expert feature array is:

[0117]

[0118] in, For the Experts participated in The first historical project expert feature parameters, No. Experts participated in A one-dimensional array of expert features for each historical item.

[0119] The two-dimensional expert feature array is:

[0120] ,in, Two-dimensional array of expert features.

[0121] S700, the two-dimensional expert project feature array and the two-dimensional expert feature array are substituted into the canonical correlation analysis model to perform canonical correlation analysis, a recommendation score is calculated according to the result of the canonical correlation analysis, and a two-way matching is performed between the target project and the expert according to the recommendation score.

[0122] Specifically, Figure 3 As shown, the specific steps of S700 are:

[0123] S710, using the two-dimensional expert project feature array and the two-dimensional expert feature array to construct a covariance matrix and substitute it into the canonical correlation analysis model to obtain the canonical correlation coefficient , expert characteristic canonical correlation coefficient , Canonical Correlation Coefficient of Item Similarity ;

[0124] Specifically, using a two-dimensional expert item feature array and a two-dimensional array of expert features Construct a covariance matrix for canonical correlation analysis and obtain a canonical correlation coefficient , and the corresponding eigenvector roots , . There is an equivalent relationship to the calculation results:

[0125]

[0126] Based on the status data of the experts in the expert database at the beginning of the project to be matched, according to the one-dimensional expert feature array The construction method constructs the second expert feature vector of the expert at the beginning of the target project , and the canonical correlation coefficient and characteristic coefficient The final recommendation score is obtained by multiplying The higher the score, the higher the recommendation level.

[0127]

[0128] in, is the typical correlation coefficient, is the expert characteristic canonical correlation coefficient, is the second expert feature vector of the expert at the beginning of the project, is the number of expert features selected.

[0129] The first expert feature vector and one expert feature array are composed of expert feature parameters before the date when the expert joined the historical project, and the second expert feature vector is composed of various expert feature parameters before the current date when the expert was engaged in the target project.

[0130] The present invention mainly uses the project experience in which the experts participated and the number of features adopted by the experts at that time. First, the project data that overlaps with the project description of the experts to be recommended are screened, and then the cosine similarity algorithm is used to calculate the expert-project similarity value of the relevance between the experts in the project experience and the projects.

[0131] The characteristic parameters involved in the calculation are then used as expert characteristic data, and the expert-project similarity value items are used as project similarity characteristic data. The typical correlation analysis method is used for analysis to obtain two typical correlation coefficients and the correlation equation of the above characteristics. The equation coefficient on the expert characteristic data side of the characteristic equation corresponding to the maximum typical correlation coefficient is taken, and multiplied by the second expert characteristic vector corresponding to the current expert to obtain the recommendation score for the match between the expert and the project.

[0132] The present invention solves the difficulty of converting text data into numerical data through the cosine similarity algorithm, and forms the matching similarity between experts and projects in multiple aspects by constructing multiple feature values. However, since the similarity is a plurality of calculated values ​​with a value range of [0,1], it is impossible to calculate the weight value of each calculated value. Then, the corresponding coefficient weight can be obtained by the canonical correlation analysis algorithm, thereby integrating all cosine similarities into a numerical value, which can also be calculated by the status of the expert. Therefore, the final calculated value is the final expert matching recommendation score.

[0133] Embodiment 2

[0134] S100, extracting keywords of the target project, determining the target field and target subject classification of the target project, and determining the number of target fields and target subjects.

[0135] The topic field extracted from this topic by manual or other means is "a1", and the extracted disciplines are "b1" and "b2". Corresponding to a=1 field and b=2 disciplines in the patent.

[0136] S200, obtaining historical projects, determining project characteristic parameters of the historical projects and expert characteristic parameters of corresponding experts participating in the historical projects according to the target field and target discipline, combining the project characteristic parameters to generate several historical project characteristic vectors, and correspondingly generating several first expert characteristic vectors according to the expert characteristic parameter combination of the historical project characteristic vectors.

[0137] It is assumed that the data contained in the database include historical projects, basic information of experts, expert papers, monographs, etc.; among which, the basic information of experts includes expert number, expert date of birth, administrative division code of expert's native place, administrative division code of expert's work place, expert's academic qualifications, expert's length of experience, expert's field, and expert's discipline.

[0138] According to the target fields and disciplines obtained in step 1, extract a batch of data that have intersections with the fields and disciplines of the current topic from the expert topic selection project situation data table. This data can be expert data or historical project data that have intersections with historical projects. And associate other data tables, including but not limited to the above data tables, construct a batch of data based on the participating projects and add other field information. Through screening, find the expert topic selection project situation table related to this project as shown in Table 1 below.

[0139] Table 1 Experts’ participation in topic selection projects

[0140]

[0141] According to Table 1, a batch of data based on participating projects is constructed, and other field information is added to form a data table as shown in Table 2:

[0142] Table 2

[0143]

[0144] Then the historical project feature vector of the historical project with project number 2 is , the first expert feature vector , the corresponding first expert project similarity value is:

[0145]

[0146] If the project feature parameters corresponding to the second historical project feature vector and the first expert feature vector are shown in Table 3, and the corresponding expert feature parameters are shown in Table 4, the weight values ​​of each field and subject classification in the historical project data are =(1,1.2,1,0.9,1,1).

[0147] Table 3

[0148]

[0149] Table 4

[0150]

[0151] Then the second historical project feature vector corresponding to project number 2 is:

[0152]

[0153] According to Table 6, the second first expert feature vector is:

[0154]

[0155] The expert item similarity value of this data is:

[0156]

[0157] If the project feature parameters corresponding to the third historical project feature vector and the first expert feature vector are as shown in Table 5 below, the corresponding expert feature parameters are as shown in Table 6 below.

[0158] Table 5

[0159]

[0160] Table 6

[0161]

[0162] The satisfaction rate is calculated based on the number of published papers as follows:

[0163] The third historical item feature vector is:

[0164] The third first expert feature vector is:

[0165] The number of published papers meets the similarity

[0166] Extract all project feature parameters with project number 2 and expert feature parameters with expert number 4 corresponding to project number 2. The project feature parameters are shown in Table 7 below, and the expert feature parameters are shown in Table 8 below:

[0167] Table 7

[0168]

[0169] Table 8

[0170]

[0171] It should be noted that the number of dimensions of the historical project feature vector and the first expert feature vector must be the same, so multiple similarity results can be calculated again in the following way:

[0172]

[0173] S300: Calculate several expert project similarity values ​​using the historical project feature vector and the first expert feature vector.

[0174] In this embodiment, i is 9, and there are 9 expert item similarity values, which are:

[0175] S400: Calculate the time correlation of historical items to obtain a time correlation value.

[0176] Assuming that the target project to be matched starts on December 4, 2024, and the expert No. 4 participated in the historical project No. 2 on April 23, 2024, the interval year is:

[0177]

[0178] Specifically, the time correlation value between the expert No. 4 who participated in the historical project No. 2 and the target project is 0.919959. It should be noted that the expert No. 4 is not necessarily the fourth expert, because it is possible that the experts No. 1, 2, and 3 are screened out. The fourth expert's order is 1, and its j corresponds to 1. This is just an example, and the specific number needs to be determined based on the actual data. And the expert No. 4 may also participate in multiple historical projects at the same time. The historical project No. 2 may be one of the projects he participated in, or it may be the only one. The specific number also needs to be determined based on the data of the historical projects in which each expert participated.

[0179] Here, assuming that the expert numbered 4 is ranked 1, and expert numbered 2 is the historical project ranked 1 by the expert numbered 4, then the time correlation value of the first historical project of the first expert is:

[0180]

[0181] Then the time correlation value between the first expert in the first historical project and the target project is 0.919959.

[0182] S500, multiplying several expert project similarity values ​​of historical projects by time correlation values ​​to obtain several expert project similarity value vectors, organizing the several expert project similarity value vectors into a one-dimensional expert project feature array, and using the one-dimensional expert project feature array of historical projects to establish a two-dimensional expert project feature array related to the target project.

[0183]

[0184] The two-dimensional expert project feature array obtained by arranging the one-dimensional expert project feature arrays of multiple historical projects of multiple experts is as follows:

[0185]

[0186] The one-dimensional expert feature arrays corresponding to the historical projects participated by multiple experts are sorted into a two-dimensional expert feature array as follows:

[0187]

[0188] S700, the two-dimensional expert project feature array and the two-dimensional expert feature array are substituted into the canonical correlation analysis model to perform canonical correlation analysis to obtain recommendation scores, and bidirectional matching is performed on the projects and experts according to the recommendation scores.

[0189] Canonical correlation analysis is demonstrated using Python code. The version is Python 3.79+. Install the third-party package: pip install scikit-learn numpy

[0190] Execution code:

[0191] import numpy as np

[0192] from sklearn.cross_decomposition import CCA

[0193] # Create a canonical correlation analysis model and specify to retain 1 typical variable

[0194] cca = CCA(n_components=1)

[0195] P = np.array([

[0196] [0.6505,0.7264,0.8132,0.5311,0.9199,0.5133,0.5211,0.1804,0.7511]

[0197] ,[0.0488,0.4717,0.0580,0.6110,0.6333,0.8569,0.6350,0.2456,0.1309]

[0198] ,[0.7970,0.0776,0.3496,0.5519,0.6994,0.1418,0.8885,0.1550,0.8861]

[0199] ,[0.6637,0.2031,0.5157,0.9468,0.3288,0.3154,0.9715,0.5760,0.5119]

[0200] ,[0.1227,0.7394,0.6166,0.9033,0.0242,0.3206,0.3917,0.3049,0.2650] ])

[0202] Q = np.array([

[0203] 0.0834, 0.0834], [1, 0, 1, 1, 0, 1, 0.9984, 0.2287, 0.5158] ])

[0205] # Fitting data

[0206] cca.fit(Q, P)

[0207] P_w,Q_w = cca.x_weights_,cca.y_weights_

[0208] # Get Canonical Correlation Coefficients

[0209] P_c,Q_c = cca.fit_transform(P, Q)

[0210] corr = np.corrcoef(Q_c[:,0],P_c[:,0])[0,1]

[0211] # Output canonical correlation coefficient

[0212] print("\nTypical correlation coefficient:", corr)

[0213] print("\nExpert characteristic typical correlation coefficient:", Q_w)

[0214] print("\nProject similarity typical correlation coefficient:",P_w)

[0215] Execution Result:

[0216] Typical correlation coefficient: 1.0

[0217] Expert characteristic canonical correlation coefficient:

[0218] [[-0.56805641] [ 0.16977395]

[0220] [-0.36911077] [ 0.06680077]

[0222] [-0.31833616] [ 0.20460992]

[0224] [-0.32868154]

[0225] [-0.17739906]

[0226] [-0.4744244 ]]

[0227] Canonical correlation coefficient of item similarity: [[ 0.10229848]

[0229] [-0.34397557]

[0230] [-0.20637639] [ 0.3888367 ] [ 0.01540759] [ 0.34397557] [ 0.38749687]

[0235] [-0.44225087] [ 0.46166481] ]

[0238] Therefore, the final calculated result is: Expert feature typical correlation coefficient = [-0.56805641, 0.16977395, -0.36911077, 0.06680077, -0.31833616, 0.20460992, -0.32868154, -0.17739906, -0.4744244]

[0239] As of the current time of the target project, the second expert feature vectors of each expert are shown in Table 9 below:

[0240] Table 9

[0241]

[0242] Then the similarity of the topic selected by the expert with serial number 13 is:

[0243]

[0244] Similarly, the similarities of the topics selected by the other three experts can be calculated. The similarity of expert No. 23 is -0.7487649569994355, the similarity of expert No. 15 is -1.9994336199999998, and the similarity of expert No. 9 is -1.9809680312794284.

[0245] Therefore, the expert numbers that best match the project are 23, 13, 9, and 15 respectively.

[0246] Embodiment 3

[0247] The present invention also provides a bidirectional matching device based on canonical correlation analysis, such as Figure 4 As shown, including:

[0248] A keyword extraction module is used to extract keywords of the target project, determine the target field and target subject classification of the target project, and determine the number of target fields and target subjects;

[0249] A historical project module is used to obtain historical projects, determine the project characteristic parameters of the historical projects and the expert characteristic parameters of the corresponding experts participating in the historical projects according to the target field and the target discipline, generate a number of historical project characteristic vectors after combining the project characteristic parameters, generate a number of first expert characteristic vectors according to the expert characteristic parameter combination of the corresponding historical project characteristic vectors, and assign characteristic values ​​to the historical project characteristic vectors and the first expert characteristic vectors;

[0250] An expert project similarity value calculation module, used to calculate a plurality of expert project similarity values ​​using the historical project feature vector and the first expert feature vector;

[0251] The time correlation analysis module is used to calculate the time correlation of experts’ participation in historical projects and obtain the time correlation value;

[0252] A project similarity array establishment module is used to obtain several expert project similarity value vectors by multiplying several expert project similarity values ​​of historical projects by time correlation values, arranging several expert project similarity value vectors into a one-dimensional expert project feature array, and using the one-dimensional expert project feature array of historical projects to establish a two-dimensional expert project feature array related to the target project;

[0253] An expert feature array establishment module is used to organize the expert project similarity value vectors corresponding to the historical projects into a one-dimensional expert feature array according to the first expert feature vector of the expert, and use the one-dimensional expert feature array to organize the two-dimensional expert feature array corresponding to the two-dimensional expert project feature array;

[0254] The matching module is used to substitute the two-dimensional expert project feature array and the two-dimensional expert feature array into the canonical correlation analysis model for canonical correlation analysis, calculate the recommendation score according to the result of the canonical correlation analysis, and perform bidirectional matching between the target project and the expert according to the recommendation score.

[0255] The specific design of each module can be referred to the description of the first embodiment, which will not be elaborated here.

[0256] Embodiment 4

[0257] The present invention also provides a computer-readable storage medium, which stores computer instructions. When the computer is executed by a processor, the two-way matching method based on typical correlation analysis as described in Example 1 is implemented.

[0258] In a specific implementation, the computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (HDD) or a solid-state drive (SSD), etc.; the computer-readable storage medium may also include a combination of the above-mentioned types of memory.

[0259] Embodiment 5

[0260] The present invention also provides an electronic device, such as Figure 5 As shown, it includes at least one processor and a memory that is communicatively connected to the processor, wherein the memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to enable the processor to perform the two-way matching method based on typical correlation analysis as described in Example 1.

[0261] In specific implementation, the number of processors may be one or more, and the processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components and other chips, or a combination of the above-mentioned chips. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0262] The memory and the processor may be communicatively connected via a bus or other means, and the memory stores instructions executable by at least one processor, and the instructions are executed by at least one processor so that the processor executes the bidirectional matching method based on typical correlation analysis as described in the above method embodiment.

[0263] Although the terminology such as target project, target field, target discipline, historical project, project feature parameter, expert feature parameter, historical project feature vector, first expert feature vector, expert project similarity value, time correlation value, expert project similarity value vector, one-dimensional expert project feature array, two-dimensional expert project feature array, one-dimensional expert feature array, two-dimensional expert feature array and recommendation score is used more frequently in this article, the possibility of using other terms is not excluded. The use of these terms is only for more convenient description and explanation of the essence of the present invention; interpreting them as any additional limitation is contrary to the spirit of the present invention.

[0264] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A two-way matching method based on canonical correlation analysis, characterized in that: The steps include: S100, extracting keywords of the target project, determining the target field and target subject classification of the target project, and determining the number of the target fields and the target subjects; S200, obtaining a historical project, determining project characteristic parameters of the historical project and expert characteristic parameters of corresponding experts participating in the historical project according to the target field and the target discipline, combining the project characteristic parameters to generate a plurality of historical project characteristic vectors, and generating a plurality of first expert characteristic vectors corresponding to the historical project characteristic vectors according to the expert characteristic parameter combination; The historical item feature vector is: Where i is the number of historical project feature vectors of the experts participating in the historical project, is the number of characteristic parameters of the project, is the characteristic parameter of the project, For the characteristic parameters of the project, is the feature vector of the historical project, is the feature vector of the i-th historical item; The first expert feature vector is: in, is the number of the first expert feature vectors of the experts who participated in the historical project, and the number of the historical project feature vectors and the first expert feature vector The values ​​are equal; is the number of expert feature parameters, is the expert feature parameter, is the t-th expert feature parameter, is the first expert feature vector, For the the first expert feature vector; S300, substituting a number of the historical project feature vectors and a corresponding number of the first expert feature vectors into a cosine similarity algorithm to obtain a number of expert project similarity values; The calculation formula of the expert project similarity value is: in, is the expert item similarity value, For the Similarity values ​​of the expert items; S400, calculating the time correlation of the expert's participation in the historical project to obtain a time correlation value; S500, multiplying a plurality of the expert project similarity values ​​by the time correlation value to obtain a plurality of expert project similarity value vectors, organizing the plurality of the expert project similarity value vectors into a one-dimensional expert project feature array, and using the one-dimensional expert project feature array of the historical project to establish a two-dimensional expert project feature array related to the target project; S600, organizing the expert project similarity value vectors corresponding to the historical projects into a one-dimensional expert feature array according to the first expert feature vector of the expert, and using the one-dimensional expert feature array to organize the two-dimensional expert feature array corresponding to the two-dimensional expert project feature array; S700, the two-dimensional expert project feature array and the two-dimensional expert feature array are substituted into a canonical correlation analysis model to perform canonical correlation analysis, a recommendation score is calculated according to the result of the canonical correlation analysis, and a bidirectional matching is performed between the target project and the expert according to the recommendation score.

2. The two-way matching method based on canonical correlation analysis according to claim 1, characterized in that: The steps of calculating the time correlation value in S400 are: S410, calculating the gap year according to the date when the expert participated in the historical project and the current date when the target project is being carried out; the calculation formula for the gap year is: in, For the interval year, the current date on which the target project is being conducted, the date on which the expert participated in the historical project; S420, calculating the time correlation value of the historical project according to the interval year; the calculation formula of the time correlation value is: in, is the number of experts, is the number of experts involved in different historical projects, The jth expert participates in The time correlation value of each of the historical items.

3. The bidirectional matching method based on canonical correlation analysis according to claim 1, characterized in that: The one-dimensional expert item feature array is: in, is the number of experts who participated in of the historical projects similarity values ​​of the expert items, is the one-dimensional expert item feature array, The one-dimensional expert project feature array for the j-th expert participating in the h-th historical project; The two-dimensional expert item feature array is: ,in, is the two-dimensional expert item feature array.

4. The two-way matching method based on canonical correlation analysis according to claim 1, characterized in that: The one-dimensional expert feature array is: in, For the The experts participated in The first The expert feature parameters, For the Experts participated in One-dimensional expert feature array for each historical item; The two-dimensional expert feature array is: ,in, is the two-dimensional expert feature array.

5. The bidirectional matching method based on canonical correlation analysis according to claim 4, characterized in that: The specific steps of S700 are: S710, using the two-dimensional expert project feature array and the two-dimensional expert feature array to construct a covariance matrix and substitute it into a canonical correlation analysis model to obtain a canonical correlation coefficient, an expert feature canonical correlation coefficient, and a project similarity canonical correlation coefficient; S720, establishing a second expert feature vector; S730, multiplying the canonical correlation coefficient, the expert feature canonical correlation coefficient corresponding to each expert, and the second expert feature vector to obtain the recommendation score of each expert; The calculation formula is: in, is the typical correlation coefficient, is the expert characteristic canonical correlation coefficient, is the second expert feature vector.

6. The bidirectional matching method based on canonical correlation analysis according to claim 5, characterized in that: The first expert feature vector and the one-dimensional expert feature array are composed of the expert feature parameters before the date when the expert joined the historical project, and the second expert feature vector is composed of the expert feature parameters before the current date when the target project is being carried out.

7. A bidirectional matching device based on canonical correlation analysis, characterized in that: include: A keyword extraction module is used to extract keywords of the target project, determine the target field and target subject classification of the target project, and determine the number of target fields and target subjects; A historical project module, used to obtain historical projects, determine project characteristic parameters of the historical projects and expert characteristic parameters of corresponding experts participating in the historical projects according to the target field and the target discipline, generate a plurality of historical project characteristic vectors after combining the project characteristic parameters, generate a plurality of first expert characteristic vectors according to the expert characteristic parameter combination corresponding to the historical project characteristic vectors, and assign characteristic values ​​to the historical project characteristic vectors and the first expert characteristic vectors; The historical item feature vector is: Where i is the number of historical project feature vectors of the experts participating in the historical project, is the number of characteristic parameters of the project, is the characteristic parameter of the project, For the characteristic parameters of the project, is the feature vector of the historical project, is the feature vector of the i-th historical item; The first expert feature vector is: in, is the number of the first expert feature vectors of the experts who participated in the historical project, and the number of the historical project feature vectors and the first expert feature vector The values ​​are equal; is the number of expert feature parameters, is the expert feature parameter, is the t-th expert feature parameter, is the first expert feature vector, For the the first expert feature vector; An expert project similarity value calculation module, used to calculate a plurality of expert project similarity values ​​using the historical project feature vector and the first expert feature vector; The calculation formula of the expert project similarity value is: in, is the expert item similarity value, For the Similarity values ​​of the expert items; A time correlation analysis module, used to calculate the time correlation of the expert's participation in the historical project to obtain a time correlation value; a project similarity array establishment module, used for respectively multiplying a plurality of the expert project similarity values ​​of the historical projects by the time correlation value to obtain a plurality of expert project similarity value vectors, arranging the plurality of the expert project similarity value vectors into a one-dimensional expert project feature array, and using the one-dimensional expert project feature array of the historical projects to establish a two-dimensional expert project feature array related to the target project; An expert feature array establishment module, used to organize the expert project similarity value vector corresponding to the historical project into a one-dimensional expert feature array according to the first expert feature vector of the expert, and use the one-dimensional expert feature array to organize the two-dimensional expert feature array corresponding to the two-dimensional expert project feature array; A matching module is used to substitute the two-dimensional expert project feature array and the two-dimensional expert feature array into a canonical correlation analysis model to perform canonical correlation analysis, calculate a recommendation score according to the result of the canonical correlation analysis, and perform bidirectional matching between the target project and the expert according to the recommendation score.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer is executed by a processor, the computer implements the bidirectional matching method based on canonical correlation analysis as described in any one of claims 1 to 6.

9. An electronic device, characterized in that: It includes at least one processor and a memory communicatively connected to the processor, wherein the memory stores instructions executable by at least one processor, and the instructions are executed by at least one processor so that the processor performs the two-way matching method based on typical correlation analysis as described in any one of claims 1-6.

Images