Expert recommendation method and device based on knowledge graph and storage medium

Abstract

The disclosure provides a knowledge graph-based expert recommendation method and device and a storage medium, and relates to the technical field of computers. The method comprises the following steps: setting a corresponding path weight for a connection path in a joint expert knowledge graph, setting a plurality of dimension weights corresponding to a second keyword based on a plurality of weight settings, searching in the joint expert knowledge graph based on a search keyword and using a shortest path search algorithm to determine a path between the search keyword and an expert node, and calculating the similarity between the expert node and enterprise demand information according to the corresponding path weight and the comprehensive weight to determine a recommended expert that matches the enterprise demand information. The method, device and medium of the disclosure make the recommended expert information more accurate and perfect, can more comprehensively and accurately recommend technical experts by finding the shortest weighted path in the knowledge graph through the shortest path algorithm, and improve the user experience.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a knowledge graph-based expert recommendation method, apparatus, and storage medium. Background Technology

[0002] When undertaking technological innovation projects, enterprises often face challenges such as insufficient internal expert reserves and a lack of adequate scientific and technological expert resources. Furthermore, existing expert matching methods suffer from incomplete expert databases, a lack of information on successfully completed projects by experts and the enterprise's historical project requirements, and simplistic processing methods for enterprise project needs. These methods fail to differentiate between the requirements of different enterprises and thus cannot effectively measure the compatibility between enterprises and experts. Therefore, a new technical solution for expert recommendation is needed. Summary of the Invention

[0003] To address the aforementioned technical problems, this disclosure is proposed. Embodiments of this disclosure provide a knowledge graph-based expert recommendation method, apparatus, and storage medium.

[0004] According to a first aspect of the present disclosure, a knowledge graph-based expert recommendation method is provided, comprising: extracting a first entity and a first association between the first entities from multiple expert information; constructing a joint expert knowledge graph based on the first entity and the first association; wherein the joint expert knowledge graph includes multiple first nodes for representing the first entity; connections between the first nodes are used to represent corresponding first associations; the first entity includes: expert name and first keyword; setting corresponding path weights for connection paths in the joint expert knowledge graph; extracting a second entity and a second association between the second entities from enterprise demand information; constructing an enterprise demand graph based on the second entity and the second association; wherein the enterprise demand graph includes multiple second nodes for representing the second entity; the second... The connections between nodes are used to represent the corresponding second association relationship; the second entity includes: enterprise name and second keyword; multiple weights are set for the second keyword based on multiple weight settings; the comprehensive weight of the second keyword is determined according to the multiple weights, and the second keyword is determined in the enterprise demand graph based on the comprehensive weight as the search keyword; based on the search keyword and using the shortest path search algorithm, a search process is performed in the joint expert knowledge graph to determine the path between the search keyword and the expert node, wherein the expert node is the first node representing the expert name; based on the path and according to the corresponding path weight and the comprehensive weight, the similarity between the expert node and the enterprise demand information is calculated to determine the recommended expert matching the enterprise demand information based on the similarity.

[0005] Optionally, setting corresponding path weights for the connection paths in the joint expert knowledge graph includes: determining the connection between the expert node and the keyword node as a direct path, and setting corresponding direct path weights for the direct path based on the weight information of the first keyword; wherein, the keyword node is a first node representing the first keyword; determining the connection between the key nodes as an indirect path, and setting corresponding indirect path weights for the indirect path based on the relationship information between the first keywords.

[0006] Optionally, the multiple weight setting dimensions include: type weight setting dimension and keyword weight setting dimension; the setting of multiple dimension weights corresponding to the second keyword based on the multiple weight setting dimensions includes: setting a corresponding first dimension weight for the second keyword based on the type weight setting dimension and the type corresponding to the second keyword; wherein, the type includes: industry, business, product characteristics, technical information of demand; and setting a corresponding second dimension weight for the second keyword based on the keyword weight setting dimension.

[0007] Optionally, determining the comprehensive weight of the second keyword based on the multiple dimension weights includes: calculating the product of the first dimension weight and the second dimension weight of the second keyword, and using the product as the comprehensive weight of the second keyword.

[0008] Optionally, determining the second keyword as a search keyword in the enterprise demand map based on the comprehensive weight includes: sorting all the second keywords based on the comprehensive weight; and selecting a preset number of second keywords as search keywords according to the sorting results.

[0009] Optionally, the step of performing search processing in the joint expert knowledge graph based on the search keywords and using a shortest path search algorithm to determine the path between the search keywords and expert nodes includes: sorting the search keywords according to the comprehensive weights; sequentially selecting search keywords in the joint expert knowledge graph based on the sorting order to search for whether there is a first node matching the search keywords; if so, using this first node as the source node and using a shortest path search algorithm to perform search processing in the joint expert knowledge graph according to the rule of shortest path length to determine the shortest path between the search keywords and each expert node; wherein, the path length is represented by path weights, and the shortest path search algorithm includes: Dijkstra's algorithm.

[0010] Optionally, the path weight is

[0011]

[0012] Where Dis[v] is the path weight from the source to the Vth first node, θi is the direct path weight of the i-th direct path between the source and the Vth first node, and λj is the indirect path weight of the j-th indirect path between the source and the Vth first node.

[0013] Optionally, calculating the similarity between the expert node and the enterprise demand information based on the path and according to the corresponding path weight and the comprehensive weight includes: calculating the similarity as follows:

[0014]

[0015] Among them, A k The comprehensive weight of the k-th search keyword is the first matching node in the joint expert knowledge graph. k It is the path weight of the shortest path between the k-th search keyword and the expert node.

[0016] Optionally, determining the recommended experts matching the enterprise's needs based on the similarity includes: obtaining the similarity corresponding to all expert nodes, sorting all expert nodes according to the similarity; determining the predicted availability time of each expert based on the expert information; and determining a predetermined number of recommended experts based on the sorting result of all expert nodes and the predicted availability time.

[0017] According to a second aspect of the present disclosure, a knowledge graph-based expert recommendation device is provided, comprising: an expert graph creation module, configured to extract a first entity and a first association between the first entities from multiple expert information, and construct a joint expert knowledge graph based on the first entity and the first association; wherein the joint expert knowledge graph includes multiple first nodes for representing the first entity; the connections between the first nodes are used to represent corresponding first associations; the first entity includes: expert name and first keyword; a path weight setting module, configured to set corresponding path weights for connection paths in the joint expert knowledge graph; and a demand graph creation module, configured to extract a second entity and a second association between the second entities from enterprise demand information, and construct an enterprise demand graph based on the second entity and the second association; wherein the enterprise demand graph includes multiple second nodes for representing the second entity; the connections between the second ... the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: expert name and first keyword; the first entity includes: The system is used to characterize the corresponding second association relationship; the second entity includes: enterprise name and second keyword; a dimension weight setting module is used to set multiple dimension weights for the second keyword based on multiple weight setting dimensions; a search term determination module is used to determine the comprehensive weight of the second keyword based on the multiple dimension weights, and determine the second keyword in the enterprise demand graph based on the comprehensive weight as a search keyword; a path search module is used to perform search processing in the joint expert knowledge graph based on the search keyword and using the shortest path search algorithm to determine the path between the search keyword and the expert node, wherein the expert node is the first node representing the expert name; a similarity determination module is used to calculate the similarity between the expert node and the enterprise demand information based on the path and according to the corresponding path weight and the comprehensive weight; and a recommendation expert determination module is used to determine the recommendation expert that matches the enterprise demand information based on the similarity.

[0018] Optionally, the path weight setting module is used to determine the connection between the expert node and the keyword node as a direct path, and set a corresponding direct path weight for the direct path based on the weight information of the first keyword; wherein, the keyword node is a first node representing the first keyword; and to determine the connection between the key nodes as an indirect path, and set a corresponding indirect path weight for the indirect path based on the relationship information between the first keywords.

[0019] Optionally, the plurality of weight setting dimensions include: type weight setting dimension and keyword weight setting dimension; wherein, the dimension weight setting module is used to set a corresponding first dimension weight for the second keyword based on the type weight setting dimension and the type corresponding to the second keyword; wherein, the type includes: industry, business, product characteristics, technical information of demand; and to set a corresponding second dimension weight for the second keyword based on the keyword weight setting dimension.

[0020] Optionally, the search term determination module is used to calculate the product of the first dimension weight and the second dimension weight of the second keyword, and use the product as the comprehensive weight of the second keyword.

[0021] Optionally, the search term determination module is used to sort all the second keywords based on the comprehensive weight; and select a preset number of second keywords as search keywords according to the sorting result.

[0022] Optionally, the path search module is used to sort the search keywords according to the comprehensive weight; based on the sorting order, sequentially select the search keywords in the joint expert knowledge graph to search for whether there is a first node that matches the search keywords; if so, use this first node as the source node and use the shortest path search algorithm to search the joint expert knowledge graph according to the rule of the shortest path length to determine the shortest path between the search keywords and each expert node; wherein, the path length is represented by the path weight, and the shortest path search algorithm includes: Dijkstra's algorithm.

[0023] Optionally, the expert recommendation module is configured to obtain the similarity to all expert nodes, sort all expert nodes according to the similarity, determine the prediction availability time of each expert according to the expert information, and determine a predetermined number of expert recommendations based on the sorting result of all expert nodes and the prediction availability time.

[0024] According to a third aspect of the present disclosure, a knowledge graph-based expert recommendation apparatus is provided, comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the method described above.

[0025] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, the storage medium storing a computer program for performing the above-described method.

[0026] Based on the knowledge graph-based expert recommendation method, apparatus, and storage medium provided in the above embodiments of this disclosure, the recommended expert information is more accurate and complete, providing a more comprehensive description of the expert's capabilities and better measuring the compatibility between experts and enterprises. By using the shortest path algorithm to find the shortest weighted path in the knowledge graph, technical expert recommendations can be made more comprehensively and accurately, improving the compatibility between experts and enterprises and enhancing the user experience.

[0027] The technical solutions of this disclosure will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0028] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0029] Figure 1 This is a flowchart illustrating an embodiment of the knowledge graph-based expert recommendation method disclosed herein.

[0030] Figure 2 This is a schematic diagram of a joint expert knowledge graph. Figure 3 A schematic diagram of a knowledge graph representing enterprise needs;

[0031] Figure 4 This is a flowchart illustrating the search path in one embodiment of the knowledge graph-based expert recommendation method disclosed herein.

[0032] Figure 5 This is a schematic diagram illustrating the process of determining a recommendation expert in one embodiment of the knowledge graph-based expert recommendation method disclosed herein; Figure 6A-6O This is a schematic diagram illustrating path search processing within a joint expert knowledge graph.

[0033] Figure 7 This is a schematic diagram of one embodiment of the knowledge graph-based expert recommendation device disclosed herein;

[0034] Figure 8 This is a schematic diagram of another embodiment of the knowledge graph-based expert recommendation device disclosed herein. Detailed Implementation

[0035] Those skilled in the art will understand that the terms "first," "second," etc., in the embodiments of this disclosure are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they indicate a necessary logical order between them.

[0036] It should also be understood that in the embodiments of this disclosure, "multiple" can refer to two or more, and "at least one" can refer to one, two, or more. It should also be understood that any component, data, or structure mentioned in the embodiments of this disclosure can generally be understood as one or more unless explicitly defined or otherwise indicated in the context.

[0037] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it need not be discussed further in subsequent figures. Embodiments of this disclosure can be applied to electronic devices such as terminal devices, computer systems, and servers, and can operate with numerous other general-purpose or special-purpose computing system environments or configurations.

[0038] Exemplary methods

[0039] Figure 1 This is a flowchart illustrating an embodiment of the knowledge graph-based expert recommendation method disclosed herein, as shown below. Figure 1 The method shown includes steps S101-S107. Each step is explained below.

[0040] S101, extract the first entity and the first relationship between the first entities from multiple expert information, and construct a joint expert knowledge graph based on the first entity and the first relationship.

[0041] In one embodiment, the joint expert knowledge graph includes multiple first nodes for representing first entities; the connections between the first nodes are used to represent corresponding first relationships; the first entity includes expert names, first keywords, etc.

[0042] S102, set the corresponding path weights for the connection paths in the joint expert knowledge graph.

[0043] In one embodiment, the expert node is a first node representing the expert's name, and the keyword node is a first node representing the first keyword. The connection between the expert node and the keyword node is determined as a direct path, and a corresponding direct path weight is assigned to the direct path based on the weight information of the first keyword. The connection between the key nodes is determined as an indirect path, and a corresponding indirect path weight is assigned to the indirect path based on the relationship information between the first keywords.

[0044] S103, extract the second entity and the second relationship between the second entities from the enterprise demand information, and construct the enterprise demand graph based on the second entity and the second relationship.

[0045] In one embodiment, the enterprise demand graph includes multiple second nodes for representing second entities; the lines between the second nodes are used to represent corresponding second relationships; the second entities include enterprise names, second keywords, etc.

[0046] S104, based on multiple weight settings, sets corresponding weights for multiple dimensions of the second keyword.

[0047] S105. Determine the comprehensive weight of the second keyword based on the weight of multiple dimensions, and determine the second keyword in the enterprise demand graph based on the comprehensive weight, which will then be used as the search keyword.

[0048] S106, based on search keywords and using the shortest path search algorithm, performs search processing in the joint expert knowledge graph to determine the path between search keywords and expert nodes.

[0049] S107. Based on the path and according to the corresponding path weight and comprehensive weight, calculate the similarity between the expert node and the enterprise demand information, so as to determine the recommended expert that matches the enterprise demand information based on the similarity.

[0050] In one embodiment, expert information is collected from multiple information channels. This information includes details about expert projects, research, and professional backgrounds, and can be structured, semi-structured, or unstructured. Various methods are employed to parse the expert information, such as using a trained neural network model to extract keywords and relationships. Multiple preprocessing steps can be performed on the expert information, such as word segmentation of expert project information, followed by removal of punctuation and invisible characters, Chinese word segmentation, removal of stop words, and removal of low-frequency words.

[0051] The connection paths in the joint expert knowledge graph are divided into two types: direct paths and indirect paths. Direct paths are those where the expert's name is directly associated with the first keyword (the first association includes research field, journal, intellectual property, supervisor, etc.); indirect paths are those where the first keyword extends from the expert's work unit, such as national-level laboratories, first-class disciplines, and major projects and research achievements that the expert participated in at their work unit.

[0052] Following the principle of "unique entity, multiple source paths," a unified joint expert knowledge graph is constructed for multiple experts. Appropriate weights are assigned to the paths within this joint expert knowledge graph. Weights can be set based on the frequency or type of the first keyword in the expert information, and path weights are then determined accordingly. For example, a direct path weight θ can be set based on the assigned weight. i 'i' represents the direct path type number, such as research field, journal, etc.; the indirect path weight is represented by λ. iThe symbol 'i' represents the indirect path type number. The weight of the indirect path indicates the degree of closeness between the different paths extending from the first keyword such as 'work unit' and the first keyword such as 'work unit'. The first keyword such as 'work unit' itself is connected to the expert node through the direct path, representing the degree of closeness between the first keyword such as 'work unit' and the expert.

[0053] For example, the extended content of expert Xiaofang's work unit is as follows: During her time at a certain pharmaceutical university (University A), Xiaofang's research and development work and achievements mainly included: Based on the major special project on fish vaccine research and development, within the context of the first-class discipline of pharmaceutical science, she successfully revealed the important role of molecule 000 in X cell-dependent humoral immune responses in small fish, providing a new theoretical basis for the development of fish vaccines and possessing significant theoretical innovation significance. She also conducted market value research on ornamental fish variants, including some basic research.

[0054] From the above information, we can extract that Xiaofang's workplace is a pharmaceutical university A, Xiaofang's research achievements during her time at the pharmaceutical university are humoral immunity and pathogenesis, the first-class discipline of the pharmaceutical university A is, and the major project that Xiaofang participated in during her time at the pharmaceutical university is vaccine research and development.

[0055] Extracting the primary keywords from the expert information of another expert, Xiao Li, the keywords for research fields are: gram-level protein, biomedicine, reagent kit development; the keywords for academic papers and journals are: animal subunit vaccines; and the keywords for invention patents are: biological products. This expert's main research work at a certain ocean university (B level) involves: animal vaccine research and development based on a major national special project, within the research background of first-class disciplines in biological products, and supported by a national-level laboratory centered on vaccine development. Research achievements include: novel targeted antibody drugs for small cell lung cancer, intracellular vesicle transport mechanisms, and research on the industrialization of vaccine applications.

[0056] Based on the expert names, primary keywords, and primary relationships of the two experts mentioned above, different entity-relationship-keyword triples are constructed and organized under the same knowledge graph, resulting in the following: Figure 2 The fused knowledge graph shown. Figure 2 As shown, the entity-relationship-entity relationship in the joint expert knowledge graph is a directed weighted graph. Starting from experts and organizations, path weights are set according to the attributes of the paths. The corresponding weights are assigned to the paths in the joint expert knowledge graph, as shown in Table 1 below.

[0057]

[0058] Table 1 - Path Weights in the Joint Expert Knowledge Graph

[0059] In one embodiment, multiple weight setting dimensions include type weight setting dimensions and keyword weight setting dimensions. Based on the type weight setting dimension and the type corresponding to the second keyword, a corresponding first-dimensional weight is assigned to the second keyword. Types include industry, business, product characteristics, and technical information related to requirements. A corresponding second-dimensional weight is assigned to the second keyword based on the keyword weight setting dimension. The product of the first-dimensional weight and the second-dimensional weight of the second keyword is calculated, and this product is used as the overall weight of the second keyword.

[0060] Enterprise demand information can be obtained from various channels, such as enterprise demand information submitted on platforms. Secondary keywords and corresponding secondary relationships can be extracted from this demand information (e.g., demand documents). Enterprise technical demands are described in two dimensions. The first dimension (type weight setting dimension) is the keyword type, such as the technology industry to which the demand belongs, technical keywords in the enterprise's main products, and the enterprise's core technology areas. The second dimension (keyword weight setting dimension) is the related keywords within the specific keyword type, such as the specific industry to which the demand belongs, such as intelligent manufacturing or biomedicine.

[0061] The first dimension uses ω to assign weights to keyword types. k This indicates that 'k' represents a specific type number, such as industry, main business, product characteristic words, specific technical content of the requirement, etc. The second dimension, keyword weight, uses... Let 'i' represent different keywords, such as clinical medicine, biomedical technology, reagent kits, and in vitro non-invasive diagnostics. Calculate the comprehensive weight (weight) of the second keyword.

[0062] The first dimension of keyword types includes demand keywords, industry, main business, product feature words, etc. The second dimension of keywords are related keywords in specific keyword types, including infectious diseases, in vitro non-invasive diagnostics, rapid testing, reagent kits, clinical medicine-basic, biomedical technology and services, biological products, respiratory medicine, etc.

[0063] For example, a company might input its technical requirements through the platform as follows: Develop a rapid, non-invasive in vitro detection technology for common bacterial infectious diseases using high-throughput sequencing, immunofluorescence detection, quantitative PCR, and nucleic acid hybridization techniques, and guide clinical trials. This includes a rapid detection technology primarily focused on common aquatic bacterial pathogens and a reagent kit. The technical specifications for this technology include the ability to simultaneously detect at least 70 common aquatic bacterial pathogens, with at least 6 molecular markers detected for each pathogen, a detection time not exceeding 8 hours, and an accuracy >85%. It is primarily intended for the rapid detection of common bacterial pathogens in aquariums, swimming pools, and natural water bodies.

[0064] Based on the above information regarding enterprise technology needs, a keyword extraction method was used to extract secondary keywords representing the specific technical content of the needs from the enterprise's input. The extracted secondary keywords included high-throughput sequencing, immunofluorescence detection, nucleic acid hybridization, and molecular markers. For each enterprise, each secondary keyword corresponds to a different weight (second-dimensional weight). Each keyword type also has a certain weight (first dimension weight) ω k The overall weight of the second keyword is:

[0065]

[0066] Where, ω k The keyword type weight of Company A,

[0067]

[0068] To assign weights to keywords collected from different fields, where i represents different keywords, different weights are assigned based on the closeness of the keyword's relationship with the company's needs. The corresponding values ​​are shown in Table 2 below:

[0069]

[0070]

[0071] Table 2 - Multi-dimensional weighting of Company A's second keyword

[0072] Based on the second keyword, the second association, and the corresponding weight information, a corresponding enterprise demand knowledge graph is constructed, such as... Figure 3 As shown.

[0073] Figure 4 This is a flowchart illustrating the search path in one embodiment of the knowledge graph-based expert recommendation method disclosed herein, as shown below. Figure 4 The method shown includes steps S401-S405. Each step is explained below.

[0074] S401, sort all secondary keywords based on comprehensive weight.

[0075] S402, Select a preset number of second keywords based on the sorting results as search keywords.

[0076] In one embodiment, the preset number can be set, for example, to 4, 5, 6, etc. All secondary keywords are sorted from highest to lowest based on a comprehensive weight, and 4, 5, 6, etc., secondary keywords are selected from the top of the sorted list as search keywords.

[0077] S403, sort search keywords based on comprehensive weight.

[0078] S404: Select search keywords sequentially based on the sorting order and search in the joint expert knowledge graph for a first node that matches the search keywords.

[0079] In one embodiment, the existence of a first node matching the search keyword is searched in the joint expert knowledge graph. If the first node represents a first keyword that is the search keyword or a synonym of the search keyword, then it is determined that a first node matching the search keyword exists.

[0080] S405, if so, then take this first node as the source and use the shortest path search algorithm to search in the joint expert knowledge graph according to the rule of the shortest path length to determine the shortest path between the search keyword and each expert node.

[0081] In one embodiment, path length is represented by path weight, and the shortest path search algorithm includes Dijkstra's algorithm, etc.

[0082] Path weight is

[0083]

[0084] Where Dis[v] is the path weight from the source to the Vth first node, θi is the direct path weight of the i-th direct path between the source and the Vth first node, λj is the indirect path weight of the j-th indirect path between the source and the Vth first node, n is the number of direct paths, and m is the number of indirect paths.

[0085] The similarity is calculated as follows:

[0086]

[0087] Among them, A k The comprehensive weight of the k-th search keyword, where the k-th search keyword has a matching first node in the joint expert knowledge graph, is Dis. k Let L be the path weight of the shortest path between the k-th search keyword and the expert node, and L be the number of search keywords that have a matching first node in the joint expert knowledge graph.

[0088] Figure 5 This is a schematic diagram illustrating the process of determining a recommendation expert in one embodiment of the knowledge graph-based expert recommendation method disclosed herein, as shown below. Figure 5 The method shown includes steps S501-S505. Each step is explained below.

[0089] S501, obtain the similarity to all expert nodes, and sort all expert nodes according to the similarity.

[0090] S502, determine the available time for each expert's prediction based on expert information.

[0091] In one embodiment, all experts are sorted according to similarity. After selecting one or more experts with the highest similarity, the information on ongoing projects in the expert information is obtained. Based on the information on ongoing projects, the available working time information of the expert is determined. For example, the number of ongoing projects of the expert and the cycle of each ongoing project are obtained. The available working time of the expert is then used as the predicted available time.

[0092] S503, based on the ranking results of all expert nodes and the predicted available time, determines a predetermined number of recommended experts.

[0093] For example, all experts are sorted based on similarity, and one or more experts with the highest similarity are selected. Information on ongoing projects from the expert information is retrieved. Based on this information, the predicted availability time for each expert is determined. Then, based on this predicted availability time, it is assessed whether the expert has time to fulfill the company's requirements. Based on this assessment, a preset number of experts with the time to fulfill the company's requirements are selected from the one or more experts with the highest similarity and sent to the company. The preset number can be 2, 3, etc.

[0094] In one embodiment, a search is performed in the joint expert knowledge graph based on search keywords and using a shortest path search algorithm. The shortest path search algorithm finds the vertex closest to the source vertex (the first node in the joint expert knowledge graph that matches the search keyword; the source vertex can be vertex 0) each time, and then expands from that vertex to obtain the shortest paths from the source vertex to all other nodes. The basic steps are as follows:

[0095] Step 1: Divide all vertices into two parts: the set P of vertices with known shortest paths and the set Q of vertices with unknown shortest paths. Initially, the set P of vertices with known shortest paths contains only the source vertex. Use a passed[i] array to record which vertices are in set P, and an unpass[j] array to record which vertices are in set Q. Define e[u][v] as the edge weight from node uv. When uv is a direct path, then... When uv is an indirect path, then The corresponding calculation formula is:

[0096]

[0097] Step 2: Set the shortest path from the source node s to itself to 0, i.e., dis = 0. If there exists a vertex i that the source node can directly reach, then set dis[i] to the weight from the source node s to node i, and set the shortest path of all other vertices (that the source node cannot directly reach) to ∞.

[0098] Step 3: Select a vertex u from all vertices in set Q that is closest to the source vertex s (i.e., has the smallest dis[u]) and add it to set P. Then examine all edges originating from vertex u and perform a relaxation operation on each edge.

[0099] For example, given an edge from u to v, a path from s to v can be extended by adding the edge u->v to the tail. The length of this path is dis[u] + e[u][v]. If this value is smaller than the currently known value of dis[v], the new value can replace the current value in dis[v].

[0100] Step four: Repeat step three. If set Q is empty, the algorithm ends. Finally, the values ​​in the dis[] array are the shortest paths from the source vertex to all vertices.

[0101] Step 5: Sort the shortest weighted paths from the search keyword to different experts, and find the expert with the shortest path weight (shortest path) 1.

[0102] Based on the weight of the search keywords, the weighted path length from each search keyword to the corresponding expert (node) is calculated sequentially. The calculation method involves iteratively executing steps two and three. Since expert 1 is already determined, in step three, it's unnecessary to iterate through all nodes. Instead, the program terminates by checking if set P contains a node associated with expert 1. The loop stops when this condition is met. The path weights dis[v] from each other search keyword to expert 1 (numbered v) are obtained sequentially and multiplied by the reciprocal of the overall weight to obtain the shortest weighted path Bi from each search keyword to expert 1, where i is the number of the different search keywords. The sum of the products Bi is used as the similarity X1 between the enterprise's technical needs and expert 1.

[0103] Dis[v] represents the shortest weighted path from source node s to node v after traversal, and the corresponding calculation formula is:

[0104]

[0105] Where i represents the direct path number from source node s to node v; j represents the indirect path number from source node s to node v. Source node s represents the keyword number, node v represents the expert node number, n is the number of direct paths, and m is the number of indirect paths. The similarity is calculated as follows:

[0106]

[0107] Where i represents the number of the different search keywords matched with the expert, such as clinical medicine, biomedical technology, etc.

[0108] The similarity X2 between expert 2 and enterprise demand information is obtained through iteration. The specific steps are as follows: Repeat steps three to five. In step four, select expert 2, which has the second shortest distance from the initial keyword obtained in step four of the previous iteration, as the fixed expert in step four. After the iteration is completed, X2 is obtained.

[0109] The similarity X3 between expert 3 and the enterprise's demand information is obtained by traversing through the graph. The traversal process is as follows: repeat steps 3-5 to obtain X3 (experts not in the subgraph are set to infinity).

[0110] The number of experts (X) to be calculated can be selected according to the required matching accuracy. Experts are sorted according to their Xi values, and the Xi values ​​are compared. The smaller the Xi value, the higher the matching degree. Finally, the top k experts with high matching degree are obtained. Based on the experts' ongoing research projects, the time when the experts can provide assistance is predicted to determine whether the experts have time to complete the company's technical needs.

[0111] In one embodiment, the shortest path search algorithm can be Dijkstra's algorithm, defining four arrays: Nopass, Passed, Dis, Path, and an edge weight matrix e[][]. Nopass is used to store points whose shortest distance is not yet determined. Passed is used to store points whose shortest distance is already determined. e[i][j] records the edge weights of nodes ij before they are searched; the array Dis records the minimum path between vertices.

[0112] For example, Dis[0][3] = 2.17, which means that the shortest path from vertex 0 to 3 is 2.17. Dis[i][j] = min(e[i][j], e[i][k] + e[k][j]). The array Path records the intermediate points in the shortest path between vertices. For example, Path[0][3] = 1, path[0][1] = 1, which means that the shortest path from 0 to 3 is: 0->1->3.

[0113] Step 1: Find words that match the search keywords in the knowledge graph. For example... Figure 6AAs shown, multiple secondary keywords, such as "biomedical technology" and "clinical medicine," are identified in the enterprise demand graph as search keywords. Multiple first nodes matching (identical or synonymous with) the search keywords are searched in the joint expert knowledge graph. These first nodes represent primary keywords such as "biological products," "infectious disease detection," "high-throughput drugs," "nucleic acid hybridization," and "biomedicine." The subgraphs containing these primary keywords are then located.

[0114] The combined weight (joint weight) of the search keywords corresponding to the first keyword is obtained from Table 2 above. The combined weight of the first keyword is the combined weight of the search keywords that match it, as shown in Table 3.

[0115]

[0116] Table 3 - Overall Weighting of the Primary Keyword and its Matching Search Keywords

[0117] Step 2: Update the paths between directly related entities in the subgraph of the expert knowledge graph to the calculated path weights. Update the direct path from the first keyword to the expert and the indirect path from the first keyword to the work unit. The list of calculated path weights for direct and indirect paths is shown in Table 4 below.

[0118]

[0119] Table 4 - Path Weights for Direct and Indirect Paths

[0120] The calculated edge weights are generated into a matrix as shown in Table 5 below:

[0121]

[0122]

[0123] Table 5 - List of Matrix Values ​​for Edge Weights

[0124] Step 3: Starting from the keyword with the highest overall weight among the unqueried search keywords, find out if there is a path between the keyword (node) that matches the search keyword and other experts (nodes). If there is, update the path length of the keyword-expert to the minimum of the known path length and the updated path length, and find the expert 1 with the shortest weighted path between the keyword and the expert.

[0125] For example, let's take the keyword "biological products" that matches the search keywords as an example. Figure 6BAs shown, a connected subgraph starting with "biological products" and including several experts and other keywords is selected. Taking path length update as an example, this illustrates how to find the shortest weighted path from the biological keyword to each expert.

[0126] The connections between the vertices in the initial subgraph of the second step are shown in Table 6 below. For ease of subsequent description, different node names are numbered 1-10, such as... Figure 6C As shown.

[0127]

[0128] Table 6 - Vertex Numbering Table

[0129] exist Figure 6C In the above, the node numbers and their representative information are as follows, and the formula for the shortest path update process in step three is as follows:

[0130] Dis[i][j]=min(e[i][j],e[i][k]+e[k][j]); Starting with biological products: we have; biological product 1-Xiao Li 2, 1.67; biological product 1-Xiao Fang 5, ∞; biological product 1-Xiao Zhang 7, ∞; biological product 1-Xiao Liu 9, ∞; it is known that the shortest path of biological product-Xiao Li is 1, and the path of biological product-Xiao Fang is to be found below. Dis[1][5]=min(e[1][5],e[1][k]+e[k][5]), where K takes the value of the node number directly connected to 1.

[0131] like Figure 6D As shown, node 2 is taken as the starting node. Nodes 1, 8, 6, and 4 are the nodes with the shortest path from the starting node to this node that have been determined. The remaining nodes 9, 7, and 5 are the nodes with the shortest weighted path that has not been determined and need to wait for the next iteration.

[0132] Nopass: Stores points whose shortest distance is not yet determined; Passed: Stores points whose shortest distance is already determined; Dis: An array where each element represents the shortest distance from the starting point to the point at the corresponding index. The initial value ∞ = 10086 is assigned to nodes with no initial path, representing infinity. Path: An array where each element represents the midpoint of the shortest distance from the starting point to the point at the corresponding index.

[0133] Step 1, as follows Figure 6EAs shown, first take node 0 as the starting node, and update the shortest path connecting other nodes through node 0: nopass:[1,2,3,4,5,6,7,8,9]; passed:[0]; dis:[0,1.67,2.0,10086,10086,10086,10086,10086,10086,10086]; path:[0,0,0,0,0,0,0,0,0,0].

[0134] Step 2, as follows Figure 6F As shown, taking node 1 as the intermediate point, update the shortest path from node 0 to other nodes through node 1: nopass:[2,3,4,5,6,7,8,9]; passed:[0,1]; dis:[0,1.67,2.0,2.17,10086,10086,10086,10086,10086,10086]; path:[0,0,0,1,0,0,0,0,0,0].

[0135] Step 3, as follows Figure 6G As shown, with node 2 as the intermediate node, update the shortest path from node 0 to other nodes through node 2: nopass:[3,4,5,6,7,8,9]; passed:[0,1,2]; dis:[0,1.67,2.0,2.17,10086,10086,10086,3.0,10086,2.85]; path:[0,0,0,1,0,0,0,2,0,2].

[0136] Step 4, as follows Figure 6H As shown, taking node 3 as the intermediate node, update the shortest path from node 0 to other nodes through node 3: nopass:[4,5,6,7,8,9]; passed:[0,1,2,3]; dis:[0,1.67,2.0,2.17,2.67,10086,2.67,3.0,10086,2.85]; path:[0,0,0,1,3,0,3,2,0,2].

[0137] Step 5, as follows Figure 6I As shown, with node 4 as the intermediate node, update the shortest path from node 0 to other nodes through node 4: nopass:[5,6,7,8,9]; passed:[0,1,2,3,4]; dis:[0,1.67,2.0,2.17,2.67,4.34,2.67,3.0,10086,2.85]; path:[0,0,0,1,3,4,3,2,0,2].

[0138] Step 6, as follows Figure 6J As shown, with node 6 as the intermediate node, update the shortest path from node 0 to other nodes through node 6: nopass:[5,7,8,9]; passed:[0,1,2,3,4,6]; dis:[0,1.67,2.0,2.17,2.67,4.34,2.67,3.0,4.97,2.85]; path:[0,0,0,1,3,4,3,2,6,2].

[0139] Step 7, as follows Figure 6K As shown, with node 9 as the intermediate node, update the shortest path from node 0 to other nodes through node 9: nopass:[5,7,8]; passed:[0,1,2,3,4,6,9]; dis:[0,1.67,2.0,2.17,2.67,4.34,2.67,3.0,4.52,2.85]; path:[0,0,0,1,3,4,3,2,9,2].

[0140] Step 8, as follows Figure 6L As shown, with node 7 as the intermediate node, update the shortest path from node 0 to other nodes through node 7: nopass:[5,8]; passed:[0,1,2,3,4,6,9,7]; dis:[0,1.67,2.0,2.17,2.67,4.34,2.67,3.0,4.52,2.85]; path:[0,0,0,1,3,4,3,2,9,2].

[0141] Step 9, as follows Figure 6M As shown, with node 5 as the intermediate node, update the shortest path from node 0 to other nodes through node 5: nopass:[8]; passed:[0,1,2,3,4,6,9,7,5]; dis:[0,1.67,2.0,2.17,2.67,4.34,2.67,3.0,4.52,2.85]; path:[0,0,0,1,3,4,3,2,9,2].

[0142] Step 10, as follows Figure 6N As shown, with node 8 as the intermediate node, update the shortest path from node 0 to other nodes through node 5: nopass:[]; passed:[0,1,2,3,4,6,9,7,5,8]; dis:[0,1.67,2.0,2.17,2.67,4.34,2.67,3.0,4.52,2.85]; path:[0,0,0,1,3,4,3,2,9,2].

[0143] Step 11, after all updates are complete, the shortest weighted path array from keyword 0 to other nodes is: dis:[0,1.67,2.0,2.17,2.67,4.34,2.67,3.0,4.52,2.85]. The corresponding shortest path table for different nodes is shown in Table 7 below:

[0144]

[0145]

[0146] Table 7 - Shortest Path Table

[0147] Based on the above process, the path weight update process between each node is completed. The path starting from the biological product and ending at different experts can be obtained. The array of transit nodes from the starting point to the ending point is: path:[0,0,0,1,3,4,3,2,9,2]. This is organized into the corresponding transit node table for different nodes as shown in Table 8 below:

[0148]

[0149] Table 8 - Transit Nodes

[0150] By changing the weights of the 10 nodes in the aforementioned subgraph, Dijkstra's algorithm is applied to the knowledge graph to find the shortest weighted path from a given keyword to different experts. Other keywords matching the search keyword are searched using the same method to find their corresponding shortest paths. The resulting two path weight tables and transit node tables show the shortest path lengths and transit nodes from the keyword "biological products" matching the search keyword to expert nodes Xiao Li, Xiao Fang, Xiao Zhang, and Xiao Liu, as shown in Table 9 below.

[0151] Product manufacturing - experts Path length path Biological Products 0 - Xiao Li 1 1.67 Biological Products - Xiao Li 0-1 Biological product 0-Xiaofang 4 2.67 Biological Products - Xiao Li - Biopharmaceuticals - Xiao Li 0-1-3-4 Biological products 0-Xiao Zhang 6 2.67 Biological Products - Xiao Li - Biopharmaceuticals - Xiao Zhang 0-1-3-6 Biological products 0-Xiao Liu 8 4.52 Biological Products - A Certain Ocean University B - Vaccine Development - Xiao Liu 0-2-9-8

[0152] Table 9 - Shortest Path Length and Transfer Nodes

[0153] like Figure 6O As shown, taking biological product 0 to Xiao Liu 8 as an example, the shortest weighted path value corresponding to column 9 of Xiao Liu in row 0 of biological product can be found in the shortest path table, which is 4.97. In the transit node table, the transit node for 0-8 is 6, for 0-6 it is 3, for 0-3 it is 1, and for 0-1 it is 1. Therefore, the shortest weighted path for biological product 0 to Xiao Liu 8 is 1-2-4-7-9. For keywords: infectious disease detection, high-throughput drugs, nucleic acid hybridization, and biomedicine, the shortest path length from different keywords to the expert is updated sequentially, resulting in the path weights from each matched keyword to the expert, as shown in Table 10 below.

[0154]

[0155] Table 10 - Path weight table from each matching keyword to the expert

[0156] Based on the aforementioned method, the shortest path lengths from different keywords to all experts and their sums are shown in Table 11 below:

[0157] <![CDATA[Path weight sum ∑α i > bioproducts Biomedicine Nucleic acid hybridization High-throughput drugs Infectious disease detection Xiaofang 2.67 0.5 6.85 3.5 1.5 Xiao Li 1.67 0.5 6.52 4.5 2.5 Xiao Zhang 2.67 0.5 3.5 2.35 4.84

[0158] Table 11 - Shortest Path Length from Keywords to Experts

[0159] The comprehensive weight calculation results of enterprise demand technology keywords (search keywords) are shown in Table 12 below:

[0160]

[0161] Table 12 - Comprehensive Weight Table of Search Keywords: Formula for Calculating Similarity Between Enterprise Needs Information and Experts:

[0162]

[0163] Where, β i Bi represents the reciprocal of the weight of the technical keywords (search keywords) related to the enterprise's technical needs, where i is the keyword number of the different technical needs that match the expert, such as clinical medicine or biomedical technology. Summing Bi yields the matching degree X between the expert and the enterprise's technical needs, as shown in Table 13 below:

[0164] expert Xiaofang Xiao Li Xiao Zhang Matching degree weight with enterprises 27.12 27.96 29.30

[0165] Table 13 - Similarity Table between Expert Nodes and Enterprise Needs Information

[0166] Based on the calculation results, Xiaofang, the expert, has a low matching score with the company, making her the most matched expert. Following her are Xiaoli and Xiaozhang, whose matching scores are relatively high. The calculation also predicts the time an expert can provide assistance based on their ongoing research projects, determining whether the expert has the time to fulfill the company's technical needs.

[0167] Exemplary device

[0168] In one embodiment, such as Figure 7 As shown, this disclosure provides an expert recommendation device based on knowledge graph, including an expert graph creation module 71, a path weight setting module 72, a demand graph creation module 73, a dimension weight setting module 74, a search term determination module 75, a path search module 76, a similarity determination module 77, and a recommendation expert determination module 78.

[0169] The expert graph creation module 71 extracts first entities and first relationships between first entities from multiple expert information, and constructs a joint expert knowledge graph based on the first entities and first relationships. The joint expert knowledge graph includes multiple first nodes representing first entities; the connections between first nodes represent corresponding first relationships; and the first entities include expert names, first keywords, etc. The path weight setting module 72 sets corresponding path weights for the connection paths in the joint expert knowledge graph.

[0170] The demand graph creation module 73 extracts second entities and second relationships between second entities from enterprise demand information, and constructs an enterprise demand graph based on the second entities and second relationships. The enterprise demand graph includes multiple second nodes representing second entities; the connections between second nodes represent corresponding second relationships; second entities include enterprise names, second keywords, etc. The dimension weight setting module 74 sets multiple dimension weights for the second keywords based on multiple weight setting dimensions.

[0171] The search term determination module 75 determines the comprehensive weight of the second keyword based on multiple dimensions, and then determines the second keyword in the enterprise demand graph based on the comprehensive weight, which serves as the search keyword. The path search module 76 performs search processing in the joint expert knowledge graph based on the search keyword and using a shortest path search algorithm to determine the path between the search keyword and expert nodes, where the expert node is the first node representing the expert's name. The similarity determination module 77 calculates the similarity between the expert node and the enterprise demand information based on the path weight and the comprehensive weight. The recommended expert determination module 78 determines recommended experts that match the enterprise demand information based on similarity.

[0172] In one embodiment, the path weight setting module 72 determines the connection between expert nodes and keyword nodes as a direct path, and sets a corresponding direct path weight for the direct path based on the weight information of the first keyword; wherein, the keyword node is the first node representing the first keyword. The path weight setting module 72 determines the connection between key nodes as an indirect path, and sets a corresponding indirect path weight for the indirect path based on the relationship information between the first keywords.

[0173] Multiple weight setting dimensions include type weight setting dimensions and keyword weight setting dimensions. The dimension weight setting module 74 sets the first dimension weight for the second keyword based on the type of the second keyword; where the type includes: industry, business, product characteristics, technical information of the demand, etc. The dimension weight setting module 74 also sets the second dimension weight for the second keyword based on the keyword weight setting dimensions.

[0174] The search term determination module 75 calculates the product of the first-dimensional weight and the second-dimensional weight of the second keyword, and uses the product as the comprehensive weight of the second keyword. Based on the comprehensive weight, the search term determination module 75 sorts all the second keywords and selects a preset number of second keywords as search keywords according to the sorting results.

[0175] The path search module 76 sorts the search keywords according to their comprehensive weights. Based on the sorting order, the path search module 76 sequentially selects each search keyword and searches the joint expert knowledge graph for a first node that matches it. If a first node is found, the path search module 76 uses this first node as the source node and employs a shortest path search algorithm to search the joint expert knowledge graph according to the shortest path length rule, determining the shortest path between the search keyword and each expert node. The path length is represented by path weights, and the shortest path search algorithm includes Dijkstra's algorithm.

[0176] Similarity determination module 77 calculates the similarity as follows:

[0177]

[0178] The expert recommendation module 78 obtains the similarity scores corresponding to all expert nodes and sorts all expert nodes according to the similarity scores. Based on the expert information, the expert recommendation module 78 determines the prediction availability time for each expert, and based on the sorting results and prediction availability times of all expert nodes, determines a predetermined number of recommended experts.

[0179] Figure 8 This is a schematic diagram of another embodiment of the knowledge graph-based expert recommendation device disclosed herein, as shown below. Figure 8 As shown, the knowledge graph-based expert recommendation device 81 includes one or more processors 811 and a memory 812.

[0180] The processor 811 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the knowledge graph-based expert recommendation device 81 to perform the desired functions.

[0181] The memory 812 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, and flash memory. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 811 may execute the program instructions to implement the knowledge graph-based expert recommendation methods and / or other desired functions of the various embodiments of this disclosure described above. Various contents such as input signals, signal components, and noise components may also be stored in the computer-readable storage medium.

[0182] In one example, the knowledge graph-based expert recommendation device 81 may further include an input device 813 and an output device 814, etc., these components being interconnected via a bus system and / or other forms of connection mechanisms (not shown). Furthermore, the input device 813 may include, for example, a keyboard, a mouse, etc. The output device 814 can output various information to the outside. The output device 814 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0183] Of course, for the sake of simplicity, Figure 8 This document only shows some of the components of the knowledge graph-based expert recommendation device 81 that are relevant to this disclosure, omitting components such as buses, input / output interfaces, etc. In addition, the knowledge graph-based expert recommendation device 81 may include any other suitable components depending on the specific application.

[0184] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps in the knowledge graph-based expert recommendation method according to various embodiments of this disclosure described in the "Exemplary Methods" section above. The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof.

[0185] The knowledge graph-based expert recommendation method, apparatus, and storage medium in the above embodiments use a knowledge graph to construct an information system, making the recommended expert information more accurate and complete, providing a more comprehensive description of the expert's capabilities, and better measuring the compatibility between experts and enterprises. By using the shortest path algorithm to find the shortest weighted path in the knowledge graph, technical expert recommendations can be made more comprehensively and accurately, improving the compatibility between experts and enterprises and enhancing the user experience.

[0186] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A knowledge graph-based expert recommendation method, comprising: extracting first entities and first association relationships between the first entities from a plurality of expert information, and constructing a joint expert knowledge graph according to the first entities and the first association relationships; wherein the joint expert knowledge graph comprises a plurality of first nodes for representing the first entities; and a connection between the first nodes represents a corresponding first association relationship; and the first entities comprise an expert name and a first keyword; setting a corresponding path weight for a connection path in the joint expert knowledge graph; extracting second entities and second association relationships between the second entities from enterprise demand information, and constructing an enterprise demand graph according to the second entities and the second association relationships; wherein the enterprise demand graph comprises a plurality of second nodes for representing the second entities; and a connection between the second nodes represents a corresponding second association relationship; and the second entities comprise an enterprise name and a second keyword; setting a corresponding plurality of dimension weights for the second keyword based on a plurality of weight setting dimensions; determining a comprehensive weight of the second keyword according to the plurality of dimension weights, and determining the second keyword as a search keyword in the enterprise demand graph based on the comprehensive weight; based on the search keyword and using a shortest path search algorithm, performing search processing in the joint expert knowledge graph to determine a path between the search keyword and an expert node, wherein the expert node is a first node representing the expert name; based on the path and according to the corresponding path weight and the comprehensive weight, calculating a similarity between the expert node and the enterprise demand information, to determine a recommended expert matching the enterprise demand information based on the similarity; wherein the setting of the corresponding path weight for the connection path in the joint expert knowledge graph comprises: determining a connection between the expert node and a keyword node as a direct path, and setting a corresponding direct path weight for the direct path based on weight information of the first keyword; wherein the keyword node is a first node representing the first keyword; and determining a connection between the keyword nodes as an indirect path, and setting a corresponding indirect path weight for the indirect path based on relationship information between the first keywords; the plurality of weight setting dimensions comprise a type weight setting dimension and a keyword weight setting dimension; and the setting of the corresponding plurality of dimension weights for the second keyword based on the plurality of weight setting dimensions comprises: setting a corresponding first dimension weight for the second keyword according to the type weight setting dimension and based on a type corresponding to the second keyword; and the type comprises an industry, a business, a product feature, and technical information of a demand; and setting a corresponding second dimension weight for the second keyword according to the keyword weight setting dimension; the determination of the comprehensive weight of the second keyword according to the plurality of dimension weights comprises: calculating a product of the first dimension weight and the second dimension weight of the second keyword, and taking the product as the comprehensive weight of the second keyword. The second keyword is determined as a search keyword in the enterprise demand graph based on the comprehensive weight, and the method comprises the following steps: All the second keywords are sorted based on the comprehensive weight, and a preset number of second keywords are selected as search keywords according to the sorting result; The search keyword is used to search in the joint expert knowledge graph based on the search keyword and using a shortest path search algorithm, so as to determine the path between the search keyword and the expert node, and the method comprises the following steps: The search keyword is sorted according to the comprehensive weight, and whether there is a first node matching the search keyword in the joint expert knowledge graph is searched according to the sorting order; if yes, the first node is taken as a source point, and a shortest path search algorithm is used to search in the joint expert knowledge graph according to the rule of shortest path length, so as to determine the shortest path between the search keyword and each expert node; wherein the path length is represented by a path weight, and the path weight is Wherein, Dis[v] is the path weight from the source point to the Vth first node, θi is the direct path weight of the ith direct path between the source point and the Vth first node, and λj is the indirect path weight of the jth indirect path between the source point and the Vth first node.

2. The method of claim 1, wherein the similarity between the expert node and the enterprise demand information is calculated based on the path and according to the corresponding path weight and the comprehensive weight, and the method comprises the following steps: The similarity is calculated as follows: wherein A k is a comprehensive weight of the kth search keyword, the kth search keyword having a matching first node in the joint expert knowledge graph, Dis k is a path weight of the shortest path between the kth search keyword and the expert node.

3. The method of claim 1, wherein the recommended expert matching the enterprise demand information is determined based on the similarity, and the method comprises the following steps: The similarity corresponding to all the expert nodes is obtained, and all the expert nodes are sorted according to the similarity; The predicted available time of each expert is determined based on the expert information; A predetermined number of recommended experts are determined based on the sorting result of all the expert nodes and the predicted available time.

4. An expert recommendation device based on a knowledge graph, comprising: An expert graph construction module configured to extract first entities and first association relationships between the first entities from a plurality of expert information, and construct a joint expert knowledge graph based on the first entities and the first association relationships; wherein the joint expert knowledge graph comprises a plurality of first nodes configured to represent the first entities; a connection between the first nodes is configured to represent a corresponding first association relationship; and the first entities comprise expert names and first keywords; A path weight setting module configured to set a corresponding path weight for a connection path in the joint expert knowledge graph; A demand graph construction module configured to extract second entities and second association relationships between the second entities from enterprise demand information, and construct an enterprise demand graph based on the second entities and the second association relationships; wherein the enterprise demand graph comprises a plurality of second nodes configured to represent the second entities; a connection between the second nodes is configured to represent a corresponding second association relationship; and the second entities comprise enterprise names and second keywords; The dimension weight setting module is used to set multiple dimension weights for the second keyword based on multiple weight setting dimensions. The search term determination module is used to determine the comprehensive weight of the second keyword based on the weights of the multiple dimensions, and to determine the second keyword in the enterprise demand graph based on the comprehensive weight, as the search keyword; A path search module is used to perform search processing in the joint expert knowledge graph based on the search keywords and using a shortest path search algorithm to determine the path between the search keywords and expert nodes, wherein the expert node is a first node representing the expert name; The similarity determination module calculates the similarity between the expert node and the enterprise demand information based on the path and according to the corresponding path weight and the comprehensive weight. The expert recommendation module is used to determine the expert recommendation that matches the enterprise's needs information based on the similarity. The path weight setting module is used to determine the connection between the expert node and the keyword node as a direct path, and set a corresponding direct path weight for the direct path based on the weight information of the first keyword; wherein, the keyword node is a first node representing the first keyword; and to determine the connection between the key nodes as an indirect path, and set a corresponding indirect path weight for the indirect path based on the relationship information between the first keywords. The multiple weight setting dimensions include: type weight setting dimension and keyword weight setting dimension; the dimension weight setting module is used to set a corresponding first dimension weight for the second keyword based on the type weight setting dimension and the type corresponding to the second keyword; the type includes: industry, business, product characteristics, technical information of demand; and to set a corresponding second dimension weight for the second keyword based on the keyword weight setting dimension. The search term determination module is used to calculate the product of the first dimension weight and the second dimension weight of the second keyword, and use the product as the comprehensive weight of the second keyword; sort all the second keywords based on the comprehensive weight; and select a preset number of second keywords as search keywords according to the sorting result. The path search module is used to sort the search keywords according to the comprehensive weight; based on the sorting order, it sequentially selects the search keywords and searches in the joint expert knowledge graph to see if there is a first node that matches the search keyword; if so, it uses this first node as the source node and uses a shortest path search algorithm to search in the joint expert knowledge graph according to the rule of shortest path length to determine the shortest path between the search keyword and each expert node; wherein, the path length is represented by the path weight, and the path weight is... Where Dis[v] is the path weight from the source to the Vth first node, θi is the direct path weight of the i-th direct path between the source and the Vth first node, and λj is the indirect path weight of the j-th indirect path between the source and the Vth first node.

5. A knowledge graph-based expert recommendation device, comprising: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1-3.

6. A computer-readable storage medium storing a computer program for performing the method according to any one of claims 1-3.

Images