Multifunctional random extraction method, system and device for NQI field expert database

Abstract

The invention discloses an expert matching method based on structured data, and the method comprises the following steps: firstly, calculating the objective weight and subjective weight of each result index through information and a Delphi method based on the structured data in a database, and generating an expert comprehensive academic score; then, keyword information input by a front end is obtained, the matching degree of keywords and expert fields is calculated by adopting a word frequency-inverse document frequency algorithm, and experts are grouped according to the matching degree; and finally, according to the scale of the expert database and the number of persons to be extracted, adaptively selecting a random extraction or hierarchical distribution extraction mode to determine a target expert. According to the method, through a dynamic weight calculation model and an intelligent stratified sampling mechanism, the accuracy and flexibility of expert matching are realized, the score is continuously optimized in combination with the historical participation performance data of the experts, and the scientificity and practicability of expert extraction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer data processing and screening, and in particular to an expert matching method based on structured data. Background Art

[0002] Existing techniques for extracting experts primarily rely on single random sampling or fixed-weight scoring methods, which struggle to simultaneously meet the requirements for both expertise and coverage. For example, traditional systems typically perform simple sorting or grouping based on keyword matching, ignoring the impact of academic proficiency scores on the final results, resulting in potentially unauthorized experts. Furthermore, fixed-threshold grouping methods are unable to adapt to the fluctuating popularity of different fields, and are prone to under-sampling or extreme sampling when the expert database is small.

[0003] Currently, mainstream expert extraction schemes mostly use static scoring models, assigning weights solely based on objective indicators such as the number of papers and patent achievements, without fully considering the influence of field characteristics and subjective experience. This single weighting method performs poorly in cross-disciplinary project reviews, with low matching accuracy and weak anti-bias capabilities. Furthermore, existing sampling strategies lack flexibility, making it impossible to dynamically adjust grouping rules and sampling ratios based on the size of the expert database, the number of samples to be extracted, and user needs. This is especially true when the number of experts to be extracted is small, and overly detailed grouping can easily lead to missing target samples.

[0004] Furthermore, traditional methods have significant shortcomings in feedback optimization. The performance data from expert meetings is not effectively integrated into the scoring update mechanism, leading to significant cumulative scoring errors during the system's long-term operation. While some technologies attempt to incorporate machine learning to improve sampling decisions, their judgments are limited to historical data statistics and fail to integrate large-scale model analysis of matching distribution characteristics, making it difficult to achieve intelligent selection of extraction methods. Therefore, the applicability and accuracy of existing technologies in complex scenarios still have significant room for improvement. Summary of the Invention

[0005] Based on this, in order to solve the technical problems of low matching accuracy, low efficiency and inability to fully consider the complex correlation between user needs and expert expertise in existing expert matching methods, an expert matching method based on structured data is proposed.

[0006] This invention protects a method for expert matching based on structured data. This method, applied to a server, includes the following steps: calculating and determining comprehensive expert academic scores based on structured data in a database, and storing the calculated results in the database; obtaining keyword information input from the front-end, determining the degree of match between the keywords and the expert's field, and grouping the experts based on the matching degree; adaptively matching the optimal sampling method based on the expert groupings, calculating and determining target experts, and feeding the target expert information back to the front-end.

[0007] Furthermore, the following steps are also included: obtaining expert performance data from front-end feedback, updating the expert comprehensive academic score through an exponential smoothing model; and updating the updated expert comprehensive academic score to the expert stratification table.

[0008] Furthermore, based on the structured data in the database, the steps of calculating and determining the expert scores include: obtaining each outcome indicator of the structured data in the database; calculating the objective weight of each outcome indicator based on information entropy theory; and adjusting the subjective weight of each outcome indicator through the Delphi method.

[0009] Furthermore, each achievement indicator includes an expert information database of professional fields, titles, academic achievements and review history; based on the structured data in the expert information database, a dynamic weight calculation model is used to generate an expert comprehensive academic score based on objective weights and subjective weights.

[0010] Furthermore, the step of determining the matching degree between the keyword and the expert field includes: using word frequency-inverse document frequency combined with similarity to calculate the matching degree between the input keyword and the expert field; generating hierarchical labels based on the matching degree, and the hierarchical labels include complete match, partial match and mismatch.

[0011] Furthermore, the step of grouping experts based on matching degree includes: obtaining the stratification threshold configured by the front end, which is used to divide the matching degree interval between keywords and expert fields; assigning a label to each expert based on the stratification threshold, loading the stratification label into the expert stratification table and updating the expert stratification table.

[0012] Furthermore, the steps of adaptively matching the best sampling method include: obtaining a preset expert threshold and a threshold of the number of experts to be extracted configured by the front end, and when the size of the expert database is less than the preset expert threshold or the number of experts to be extracted is less than the threshold of the number of experts to be extracted, randomly extracting only from the group with the highest matching degree; When the size of the expert database within the group is at least the preset expert threshold and the number of experts to be extracted is at least the threshold number of experts to be extracted, stratified distribution extraction is adopted.

[0013] Furthermore, the use of stratified distribution extraction includes: adjusting the proportion of each layer through the front-end visualization page, obtaining the proportion data of each layer, and realizing random extraction within each layer according to the proportion; or adjusting and determining the normal distribution parameters through the front-end visualization page, and realizing normal extraction within each layer according to the normal distribution parameters.

[0014] Furthermore, achieving normal extraction within each stratum according to normal distribution parameters includes: constructing a normal distribution model based on the experts' comprehensive academic level scores; according to the normal distribution model, preferentially extracting samples from the medium matching group, followed by extracting samples from the high matching group, and finally extracting samples from the low matching group.

[0015] Furthermore, the big model automatically optimizes the expert stratification weight distribution ratio and the intra-stratification extraction method based on historical data and project requirements, thereby achieving consistency between the in-person matching method and the historical matching method.

[0016] The present invention protects an expert matching method based on structured data. This method calculates the comprehensive academic scores of experts and stores them in a database. It then determines the degree of expert field matching based on keyword information input from the front end. The method further groups the experts and adaptively selects the best sampling method to determine the target experts for feedback to the front end. Its core innovation lies in combining structured data with adaptive sampling to achieve a precise and efficient expert matching process, thereby significantly improving the accuracy and efficiency of expert matching while reducing the need for manual intervention. Furthermore, in the expert matching process, the normal distribution of front-end parameters combined with an adjusted probability density function can ensure the breadth of expert matching domains on the one hand, and prioritize recommendations for highly rated experts on the other, thereby optimizing both the depth and breadth of expert recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 A flowchart of an expert matching method based on structured data provided in an embodiment of the present application; Figure 2 A schematic diagram of calculating an expert comprehensive academic score provided in an embodiment of the present application; Figure 3 A schematic diagram of expert domain matching provided in an embodiment of the present application; Figure 4 A schematic diagram of target expert determination provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of an expert matching device based on structured data provided in an embodiment of the present application; Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for which protection is claimed, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work falls within the scope of protection of the present application.

[0019] Research has found that current expert selection methods primarily rely on completely random sampling, which often fails to meet practical needs. This is particularly true when cross-disciplinary or interdisciplinary reviews are required, where the results may be overly concentrated or extreme. Furthermore, existing technologies lack a comprehensive quantitative assessment of experts' academic proficiency, making it difficult to fully consider their actual capabilities during the selection process. Furthermore, for expert databases in the National Quality Infrastructure (NQI) field, there is no method that can implement feature keyword matching and screening, sorting by academic ability, and combining completely random sampling with normal distribution sampling. This limits the flexibility and scientific nature of expert selection.

[0020] Based on this, see Figure 1 , the embodiment of the present application provides an expert matching method based on structured data, which calculates the expert comprehensive academic score ( Figure 2 ), determine the matching degree between keywords and expert fields and group them ( Figure 3 ), and adaptively matching the best sampling method to determine the target expert ( Figure 4 The specific steps include: calculating and determining the comprehensive academic score of experts based on the structured data in the database, and storing the calculation results in the database; obtaining the keyword information input by the front-end, determining the matching degree between the keywords and the expert fields, and grouping the experts based on the matching degree; based on the expert grouping, adaptively matching the best sampling method, calculating and determining the target experts, and feeding the target expert information back to the front-end.

[0021] The embodiment of the present application provides an expert matching method based on structured data, which can effectively solve the problems of single expert extraction method and incomplete academic level assessment in the existing technology. First, by comprehensively and quantitatively scoring the academic level of experts, the scientific assessment of expert capabilities is ensured; second, through keyword matching and grouping operations, accurate screening of expert fields is achieved; finally, based on the best sampling method of expert grouping adaptive matching, it not only supports completely random extraction, but also can be combined with normal distribution proportional sampling, thereby taking into account the professionalism and diversity of the extraction results. This method significantly improves the flexibility and applicability of expert extraction, and is particularly suitable for the complex review needs in the field of national quality infrastructure (NQI).

[0022] The present invention will be described in further detail below with reference to specific examples, but this is not intended to limit the present invention.

[0023] S101 calculates the expert's comprehensive academic score based on the structured data in the database and stores the results in the database. During implementation, it first needs to obtain the expert's structured data from the database. This structured data includes, but is not limited to, information such as professional field, professional title, academic achievements, and review history. By collecting this multi-dimensional data, it provides comprehensive foundational support for subsequent expert comprehensive academic scoring.

[0024] S1011 obtains various achievement indicators of the structured data in the database. Specifically, the various achievement indicators cover the expert's professional field, professional title level, number of published papers, number of patents, standard setting status, and other quantitative indicators that can reflect academic level. For example, for a certain expert, he or she may have academic achievement records in multiple dimensions, such as 30 SCI papers published, 15 invention patents applied for, and presided over the formulation of 3 national standards. By classifying and extracting these achievement indicators, a clear data basis is provided for subsequent weight calculations.

[0025] S1012 determines the objective weight of each outcome indicator based on information entropy. To ensure the scientific and rational nature of the weight distribution, information entropy is used to calculate the objective weight of each outcome indicator. During the objective weight determination process, the degree of fluctuation of the objective indicator data is determined. The greater the fluctuation, the greater the information content of that dimension, and thus a higher weight value is assigned. For example, in an expert database, if a certain outcome indicator (such as the number of published papers) varies significantly among all experts, then that indicator will be assigned a higher weight; conversely, if the data difference of a certain indicator is small, then the weight will be lower.

[0026] S1013 adjusts the subjective weights of each outcome indicator. In addition to weight calculations based on objective data, subjective weights need to be introduced to compensate for the shortcomings of purely data-driven methods. In specific implementation, authoritative experts in relevant fields are invited to form a review panel to score the importance of each outcome indicator and iteratively optimize it. For example, for the indicator "publishing papers", experts in different fields may give different weight recommendations. Ultimately, a unified opinion is formed through multiple rounds of feedback, making the subjective weights more in line with actual needs.

[0027] S1014 uses a dynamic weight calculation model to generate the expert's comprehensive academic score based on the structured data in the expert information database, as well as the objective weights and supervisory weights. After completing the above-mentioned objective and subjective weight calculations, the dynamic weight calculation model is used to quantitatively evaluate the comprehensive academic level of each expert. The advantage of the dynamic weight model is that it can flexibly adjust the weight distribution ratio according to actual conditions. For example, in certain specific projects, when more attention is paid to the academic achievements of experts rather than their titles, the key points can be highlighted by adjusting the model parameters. The final generated comprehensive academic score of the expert not only reflects his or her scientific research ability, but also provides an important reference basis for subsequent expert matching.

[0028] In summary, the experts comprehensively evaluate the academic level of each expert, including obtaining the level and number of awards, published papers, patents and standards from the database as the basis for random sampling or selection of review experts. To this end, in the evaluation process, according to the number of various academic achievements of multiple experts, appropriate weights are assigned to each achievement, and then their weighted sum is calculated, and the final result is used as the comprehensive academic level evaluation indicator. Represents the experts in the expert database, where i=1,…,n, indicating that there are a total of n experts’ data. Each is a p-dimensional vector, representing p different scientific research results, namely:

[0029] Therefore, all data are collected and processed in the back-end server of the entire system, forming an n×p dimensional matrix. , calculate a corresponding weight:

[0030] Regarding the methods for determining the function f(•) in the above formula, they can be generally divided into two categories: subjective and objective. The subjective method is to consult experts in related fields and assign appropriate weights to each dimension. The objective method is to calculate the corresponding weights based on the actual data of the column through statistical methods. The greater the discrimination or fluctuation of the data, the greater the amount of information it contains, and the greater the weight should be assigned. The subjective method can take into account more factors other than data, but it is easy to lead to unreasonable weight assignment, so that the final evaluation index does not have enough discrimination; the objective method relies entirely on data. In the example, the data of scientific research results that are easier to obtain are often very rich and have sufficient information, while those very important academic results are usually difficult to obtain, and their data are likely to contain many zeros. From the data point of view alone, the amount of information is approximately equal to zero. Therefore, a combination of the two methods is used to overcome the determination of the two methods as much as possible, and calculate a relatively reasonable weight for each column of data. In this way, for each expert , through back-end calculations, we can get a weighted sum of all scientific research results data:

[0031] Put this It serves as an evaluation indicator for the comprehensive academic level of each expert and is used as a reference when selecting review experts.

[0032] Furthermore, to ensure the traceability and accuracy of the scoring results, the calculated comprehensive academic scores of the experts are stored in a database for easy access and updating. This completes a closed-loop management process from data collection to score generation, ensuring the scientific and reliable nature of the expert matching method.

[0033] The above is the calculation of expert scores in advance in the back-end, and the calculated data is stored in the corresponding table of the database, which makes it easy to track the entire calculation process and logic. After that, it is necessary to match the expert field, which is explained in detail in S102.

[0034] S102 obtains keyword information input from the front-end, determines the degree of match between the keywords and the expert's field, and groups the experts based on this match. In specific implementations, the front-end first obtains the user-entered keyword information (e.g., the professional direction or field of project review) and calculates the degree of match between the two, combined with the expert's field information stored in the database. This step allows the selection of experts with a high correlation with the input keyword, providing a basis for subsequent grouping.

[0035] S1021 uses a combination of term frequency and inverse document frequency (TF-IDF) similarity to calculate the match between the input keyword and the expert's domain. To achieve precise matching, this embodiment uses the term frequency and inverse document frequency (TF-IDF) algorithm combined with cosine similarity to measure the match between the input keyword and the expert's domain. Specifically, for each expert's domain description, the keyword feature vector is first extracted using TF-IDF, and then the cosine similarity between this vector and the user-entered keyword vector is calculated. For example, when a user enters "artificial intelligence" as a keyword, the system vectorizes the keyword "artificial intelligence" to form vector data with 768 dimensions. Before entering the keyword "artificial intelligence," the system has already vectorized the technical fields of all experts and stored the processed data in a vector database. This 768-dimensional artificial intelligence vector data is then matched with the content in the vector database, and the match between the two is calculated. This gives a matching score for all experts in the "artificial intelligence" field. The higher the score, the more closely the expert fits the field.

[0036] S1022 generates hierarchical labels based on the matching degree, and the hierarchical labels include complete match, partial match and mismatch. According to the calculated matching degree, experts are divided into different matching levels. The hierarchical labels mainly include three categories: complete match (matching degree is higher than the preset high threshold), partial match (matching degree is between the high and low thresholds) and mismatch (matching degree is lower than the low threshold). For example, if the threshold for complete match is set to 0.9 and the threshold range for partial match is 0.6 to 0.9, then an expert with a matching degree of 0.95 will be marked as "complete match", while an expert with a matching degree of 0.7 will be marked as "partial match".

[0037] The threshold here can be configured on the front end or the system default partition. If it is front-end configuration data, the stratification threshold configured on the front end is obtained. The stratification threshold is used to divide the matching range between keywords and expert fields. In actual applications, users are allowed to customize the stratification threshold according to their needs to flexibly adjust the grouping strategy. For example, in some scenarios, it may be more inclined to strictly distinguish the matching degree of experts. In this case, a higher stratification threshold can be set; in other scenarios, a loose threshold standard can be selected to expand the range of candidate experts. In this way, the system can better adapt to the needs of different business scenarios. During the calculation process of the back-end server, it is first necessary to determine whether the threshold is limited by the front end. If it is limited, the threshold limited by the front end will be used for partitioning; if it is not limited, the system default data will be used for distinction.

[0038] S1023 assigns a label to each expert based on the stratification threshold, loads the stratification label into the expert stratification table, and updates the expert stratification table. After completing the matching calculation and stratification label generation, the system assigns a corresponding stratification label to each expert and records it in the expert stratification table. Simultaneously, the expert stratification table is updated regularly or in real time to ensure the timeliness and accuracy of the data. This process not only facilitates the subsequent extraction of target experts on demand but also provides basic support for the dynamic optimization of the system.

[0039] In summary, the above steps complete the entire process from keyword input to expert matching and grouping, ensuring the scientific and targeted nature of the expert selection process. This method not only considers the accuracy of matching keywords and expert fields, but also takes into account flexibility and scalability, laying a solid foundation for the subsequent selection of the optimal sampling method.

[0040] S103 adaptively matches the best sampling method based on expert grouping, calculates and determines the target expert, and feeds the target expert information back to the front end.

[0041] During implementation, the system first performs further processing based on the previously generated expert stratification table (e.g., complete match, partial match, and mismatch). It then selects the most appropriate sampling method by analyzing the current size of the expert pool and the required number of experts to be sampled. The following are the detailed implementation steps.

[0042] S1031 obtains the preset expert threshold and the threshold for the number of experts to be extracted configured by the front end. When the size of the expert database is smaller than the preset expert threshold or the number of experts to be extracted is smaller than the threshold for the number of experts to be extracted, random extraction is performed only within the group with the highest matching degree.

[0043] In practice, users can enter or configure two key parameters through the front-end interface: the preset expert threshold and the threshold for the number of experts to be sampled. These parameters determine whether the current expert pool is large enough to support complex sampling strategies. For example, if the total number of experts in a given field is small (less than the preset expert threshold) or the number of experts to be sampled is small (less than the threshold for the number of experts to be sampled), the system will automatically select a random sample from the "perfect match" group to simplify operations and improve efficiency. This simple random sampling method ensures fair results while preventing complex sampling methods from failing due to insufficient sample size.

[0044] S1032 When the size of the expert database in the group is at least the preset expert threshold and the number of experts to be extracted is at least the threshold number of experts to be extracted, stratified distribution extraction is adopted.

[0045] If both the size of the expert pool and the number of people to be selected meet the above conditions, the system will enter the more complex stratified distribution extraction stage. This stage allows users to flexibly adjust the stratification ratio or normal distribution parameters according to their needs, thereby achieving more refined expert selection.

[0046] Adjust the proportion of each layer through the front-end visualization page, obtain the proportion data of each layer, and implement random sampling within each layer according to the proportion.

[0047] In this sub-step, users can manually set the proportions of each stratum through the front-end visual interface (e.g., 60% for the fully matched group, 30% for the partially matched group, and 10% for the mismatched group). The system then performs random sampling independently within each stratum based on these proportions. For example, if the target is to sample five experts, three might be selected from the fully matched group, two from the partially matched group, and no mismatched group would be considered.

[0048] Alternatively, the normal distribution parameters can be adjusted through the front-end visualization interface, and normal sampling within each stratum can be achieved based on these parameters. The motivation for this sampling method is similar to stratified sampling, aiming to cover a wider range of expert samples with varying degrees of matching. However, stratified sampling still considers the proportion of the population. If the samples with the highest matching degree have a numerical advantage, the largest number of samples will be drawn from them. Normal proportion sampling, on the other hand, does not consider the proportion of different matching degrees, ensuring that the largest number of samples is drawn from samples with moderate matching degrees. Consider the standard normal distribution N(0,1) (where μ=0, σ=1). According to the 3σ principle, most of its density is concentrated in the interval [-3σ, 3σ], so this interval is primarily considered. Since there are q different matching scores, we divide this interval into q smaller intervals: [-3+6k / q, -3+6(k+1) / q], where k=0,…,q-1. Let Φ(•) be the distribution function of the standard normal distribution. Then for the sample group with matching score , we need to extract samples from it.

[0049] As another preferred option, users can also choose a sampling method based on normal distribution. This method uses the front-end page to input the relevant parameters of the normal distribution (such as mean μ and standard deviation σ) to build a normal distribution model for different matching groups. The following are the specific implementation details: S10321 constructs a normal distribution model based on the expert comprehensive academic proficiency scores: The system uses the calculated expert comprehensive academic proficiency scores (see S101) and the normal distribution parameters provided by the user to generate a probability density function related to academic ability. For example, higher academic scores correspond to higher extraction probabilities.

[0050] The system normalizes the expert scores to make them conform to the normal distribution and calculates the extraction probability of each expertP ( Si ) can correct the skewness of the probability density function (e.g., right-skewed distribution), ensuring a higher probability of high-scoring experts. The specific implementation process will be described later.

[0051] S10322 Based on the normal distribution model, samples are drawn from the medium-match group first, followed by samples from the high-match group, and finally from the low-match group: Based on the above normal distribution model, the system will give priority to drawing samples from the medium-match group, and according to the probability density function, give priority to experts with higher comprehensive scores among the medium-match experts. Subsequently, the remaining samples are drawn from the high-match group and the low-match group in turn until the required total number of people is met. The extraction of the high-match group and the low-match group can also be processed using the above-mentioned probability density function that has been corrected for skewness. This approach can not only ensure wide coverage of professional fields, but also take into account the high and low distribution of academic abilities.

[0052] The following example introduces the implementation logic of the above two steps: In the specific implementation method, through step S102, the system divides the experts into three groups according to the keyword matching degree: high matching group (complete matching keywords, such as matching score ≥ 80%); medium matching group (partial matching keywords, such as 40% ≤ matching score < 80%); low matching group (weak matching or no matching, such as matching score < 40%).

[0053] Then, the normal distribution driven priority extraction strategy of this part is adopted. The first step is to extract the medium matching group. This group of experts is usually concentrated in the middle area of ​​the normal distribution. The system selects the experts according to the modified P ( Si ) selection process, prioritizing experts with above-average academic scores but moderate fit to ensure a balance between diversity of opinion and expertise. For example, if 10 individuals are to be sampled, 7 individuals will be drawn from the moderate fit group in the first round (accounting for 70% of the total, consistent with a normal distribution).

[0054] Step 2: Selection of the remaining places from the high-match group will be carried out in descending order of academic scores, with priority given to experts with top academic abilities and complete matches.

[0055] Step 3: Supplementary sampling of the low-match group, and finally randomly select a small number of samples (such as 10%) from the low-match group to introduce cross-disciplinary perspectives and stimulate innovative review opinions.

[0056] S1033 feeds back the target expert information to the front end.

[0057] Ultimately, regardless of the sampling method used, the system will aggregate the calculated target expert information (including name, field, academic rating, etc.) and return it to the user through the front-end interface. The user can then further confirm or adjust the expert selection results based on this information, thus completing the entire expert matching process.

[0058] Through the above steps, this method realizes the automation of the entire process from expert grouping to sampling method selection, which not only ensures the scientificity and rationality of the results, but also fully considers the diverse needs in actual application scenarios.

[0059] After the experts complete their review or consultation tasks, the conference organizers submit feedback on the participants' performance through the system, including: professional level evaluation (1-5 points), contribution score (such as report quality, discussion participation), and other notes (such as whether they served as group leader).

[0060] Score Update Mechanism: Feedback data will be integrated into the expert's overall academic score using a weighted average method: New Score = Original Academic Score × 0.7 + Meeting Performance Score × 0.3. If the expert has participated in multiple meetings, the average of the three most recent feedback will be used to minimize the impact of single performance fluctuations.

[0061] When the expert rating is updated, the system automatically adjusts the expert table in the database, feeds the data determined by the update mechanism back into the table, and dynamically updates the calculated expert rating value.

[0062] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of an expert matching device based on structured data provided by an embodiment of the present application. The input interface of the method is provided through the graphical user interface of the terminal device; Figure 5 As shown in , the matching device 500 includes: Expert comprehensive academic score calculation module 510, used to calculate and determine the expert comprehensive academic score based on the structured data in the database, and store the calculation result in the database; Expert field matching module 520, which is used to obtain keyword information input by the front end, determine the matching degree between the keyword and the expert field, and group the experts based on the matching degree; The target expert determination module 530 , based on expert grouping, adaptively matches the best sampling method, calculates and determines the target expert, and feeds the target expert information back to the front end.

[0063] Furthermore, the expert comprehensive academic score calculation module 510 is used to obtain various achievement indicators of the structured data in the database; Calculate the objective weight of each outcome indicator based on information entropy theory; The subjective weight of each outcome indicator was adjusted through the Delphi method.

[0064] Furthermore, the expert domain matching module 520 uses word frequency-inverse document frequency combined with similarity to calculate the matching degree between the input keyword and the expert domain; A hierarchical label is generated based on the matching degree, wherein the hierarchical label includes a complete match, a partial match, and a non-match.

[0065] Furthermore, the target expert determination module 530 obtains a preset expert threshold and a threshold for the number of experts to be extracted configured by the front end, and when the size of the expert database is smaller than the preset expert threshold or the number of experts to be extracted is smaller than the threshold for the number of experts to be extracted, random extraction is performed only within the group with the highest matching degree; When the size of the expert pool within the group is at least the preset expert threshold and the number of experts to be extracted is at least the threshold number of experts to be extracted, stratified distribution extraction is adopted.

[0066] The present invention provides an expert matching device based on structured data. This device acquires configuration data through a terminal device's graphical user interface. By manipulating data in a database, it provides comprehensive expert academic scores. Based on expert field matching, it then groups experts by their technical fields, and then uses a normal distribution to identify expert candidates with both breadth and depth. This allows operators to configure data through a visual interface, enabling intelligent matching and recommendation of expert resources, significantly improving the efficiency and accuracy of expert screening.

[0067] See also Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown in FIG, the electronic device 600 includes a processor 610 , a memory 620 and a bus 630 .

[0068] The memory 620 stores machine-readable instructions executable by the processor 610. When the electronic device 600 is running, the processor 610 communicates with the memory 620 via the bus 630. When the machine-readable instructions are executed by the processor 610, the above-mentioned Figure 1 The specific implementation of the expert matching method based on structured data in the method embodiment shown can be found in the method embodiment, and will not be described in detail here.

[0069] The embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 The specific implementation of the steps of the expert matching method based on structured data in the illustrated method embodiment can be found in the method embodiment, and will not be repeated here.

[0070] 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 make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

[0071] Other embodiments of the present invention will readily occur to those skilled in the art after considering the specification and practicing the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of this application and incorporate common knowledge or customary techniques in the art disclosed herein. The description and examples are to be considered merely as exemplary, and the true scope of this application is indicated by the claims.

Claims

1. An expert matching method based on structured data, applied to a server, comprising the following steps: Based on the structured data in the database, calculate and determine the expert's comprehensive academic score, and store the calculation results in the database; Obtain keyword information input by the front-end, determine the matching degree between the keyword and the expert field, and group the experts based on the matching degree; Based on expert grouping, the optimal sampling method is adaptively matched, the target expert is calculated and determined, and the target expert information is fed back to the front end.

2. The expert matching method based on structured data according to claim 1, characterized in that: The step of calculating and determining the expert score based on the structured data in the database includes: Obtaining various achievement indicators of the structured data in the database; Calculate the objective weight of each outcome indicator based on information entropy; The subjective weight of each outcome indicator was adjusted through the Delphi method.

3. The expert matching method based on structured data according to claim 2, characterized in that: The achievement indicators include professional fields, professional titles, academic achievements and expert information databases with participation in review history; Based on the structured data in the expert information database, combined with the objective weight and subjective weight, a dynamic weight calculation model is used to generate the expert comprehensive academic score; In the model Represents the experts in the expert database, each It is a p-dimensional vector, representing p different scientific research results. For each column of data, a corresponding weight is calculated: ； Using a combination of objective and subjective methods, we determine the weight of each column of data, and the weighted sum of scientific research results data is calculated to determine the comprehensive academic level evaluation index of each expert. , .

4. The expert matching method based on structured data according to claim 1, characterized in that: The step of determining the matching degree between the keyword and the expert field includes: The matching degree between input keywords and expert fields is calculated by combining word frequency-inverse document frequency with similarity; A hierarchical label is generated based on the matching degree, wherein the hierarchical label includes a complete match, a partial match, and a non-match.

5. The expert matching method based on structured data according to claim 4, characterized in that: The step of grouping experts based on matching degree includes: Get the stratification threshold configured on the front end. The stratification threshold is used to divide the matching interval between keywords and expert fields. Assign a label to each expert based on the stratification threshold, load the stratification label into the expert stratification table, and update the expert stratification table.

6. The expert matching method based on structured data according to claim 1, characterized in that: The step of adaptively matching the best sampling mode includes: Obtain the preset expert threshold and the threshold for the number of experts to be extracted configured by the front-end. When the size of the expert database is smaller than the preset expert threshold or the number of experts to be extracted is smaller than the threshold, only randomly extract from the group with the highest matching degree. When the size of the expert pool within the group is at least the preset expert threshold and the number of experts to be extracted is at least the threshold number of experts to be extracted, stratified distribution extraction is adopted.

7. The expert matching method based on structured data according to claim 6, characterized in that: The layered distribution extraction includes: adjusting the proportion of each layer through the front-end visualization page, obtaining the proportion data of each layer, and realizing random extraction within each layer according to the proportion; or adjusting and determining the normal distribution parameters through the front-end visualization page, and realizing normal extraction within each layer according to the normal distribution parameters; the normal distribution parameters include the mean μ and the standard deviation σ.

8. The expert matching method based on structured data according to claim 7, characterized in that: The method of realizing normal extraction in each layer according to normal distribution parameters includes: Constructing a normal distribution model based on the comprehensive academic level scores of the experts; According to the normal distribution model, samples are first drawn from the medium matching group, then from the high matching group, and finally from the low matching group.

9. The expert matching method based on structured data according to claim 8, characterized in that: The normal distribution model uses the normal distribution parameters to generate a probability density function related to academic ability. P ( Si ), and correct the probability density function to ensure that experts with high comprehensive academic level scores are more likely to be selected.

10. The expert matching method based on structured data according to claim 1, characterized in that: The following steps are also included: Obtain feedback on expert performance data and update comprehensive academic scores; The updated expert comprehensive academic score is updated to the expert stratification table.

Images