Character recommendation method and device, electronic equipment and storage medium

Abstract

The invention provides a figure recommendation method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining a to-be-audited item; performing vectorization processing on the to-be-audited item to obtain a to-be-audited item vector; performing vectorization processing on a plurality of pre-stored figures to obtain a plurality of figure vectors; according to the to-be-audited item vector and the plurality of character vectors, to-be-selected characters consistent with the to-be-audited item in category are determined from a plurality of characters; and determining a target person from the to-be-selected persons according to a preset screening standard, and generating a person recommendation result according to the target person. In this way, it is ensured that the to-be-selected person and the to-be-audited item belong to the same technical field, and the target person is more accurate by further screening the to-be-selected person. By generating the character recommendation result, the target character matched with the to-be-audited item can be clearly displayed in the form of the character recommendation result.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to a character recommendation method, device, electronic device and storage medium. Background Art

[0002] In the process of science and technology project management, people in related fields are needed to review science and technology projects. However, the matching methods of science and technology project reviewers are usually manual matching, keyword matching and random matching, which is difficult to meet the matching needs of people in interdisciplinary projects.

[0003] In view of this, how to accurately match people in related fields of scientific and technological projects for project review has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, the purpose of the present disclosure is to propose a character recommendation method, device, electronic device and storage medium to solve or partially solve the above technical problems.

[0005] Based on the above purpose, the first aspect of the present disclosure proposes a character recommendation method, the method comprising:

[0006] Get the items to be reviewed;

[0007] Performing vectorization processing on the items to be reviewed to obtain a vector of items to be reviewed;

[0008] Vectorizing the pre-stored multiple characters to obtain multiple character vectors;

[0009] Determining, from the plurality of persons, a person to be selected that is consistent with the category of the item to be reviewed according to the vector of the item to be reviewed and the plurality of person vectors;

[0010] A target person is determined from the candidate persons according to a preset screening criterion, and a person recommendation result is generated according to the target person.

[0011] Based on the same inventive concept, the second aspect of the present disclosure proposes a character recommendation device, comprising:

[0012] An acquisition module, configured to acquire items to be reviewed;

[0013] A first vectorization processing module is configured to perform vectorization processing on the item to be reviewed to obtain an item to be reviewed vector;

[0014] A second vectorization processing module is configured to perform vectorization processing on a plurality of pre-stored characters to obtain a plurality of character vectors;

[0015] A first determination module is configured to determine a candidate person who is consistent with the category of the item to be reviewed from a plurality of persons according to the item to be reviewed vector and the plurality of person vectors;

[0016] The second determination module is configured to determine a target person from the candidate persons according to a preset screening criterion, and generate a person recommendation result according to the target person.

[0017] Based on the same inventive concept, the third aspect of the present disclosure proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the method as described above when executing the computer program.

[0018] Based on the same inventive concept, a fourth aspect of the present disclosure proposes a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method as described above.

[0019] From the above, it can be seen that the character recommendation method, device, electronic device and storage medium provided by the present disclosure. Obtain the project to be reviewed. Vectorize the project to be reviewed to obtain the project vector to be reviewed. Vectorize multiple pre-stored characters to obtain multiple character vectors. According to the project vector to be reviewed and the multiple character vectors, it is possible to accurately determine the candidate character that is consistent with the category of the project to be reviewed from multiple characters, and ensure that the candidate character and the project to be reviewed belong to the same technical field, so that the candidate character is more accurate. Determine the target person from the candidate characters according to the preset screening criteria, and generate a character recommendation result based on the target person. In this way, by further screening the candidate character, the target person belongs to the same technical field as the project to be reviewed and meets the preset screening criteria, so that the target person is more accurate. By generating the character recommendation results, the target person that matches the project to be reviewed can be clearly displayed in the form of character recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present disclosure or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 A flowchart of a character recommendation method according to an embodiment of the present disclosure;

[0022] Figure 2 A schematic diagram of a character information relationship diagram of an embodiment of the present disclosure;

[0023] Figure 3 A schematic diagram of matching characters of scientific and technological projects according to an embodiment of the present disclosure;

[0024] Figure 4 is a structural schematic diagram of a character recommendation device according to an embodiment of the present disclosure;

[0025] Figure 5 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0027] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present disclosure should be understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0028] Based on the description of background technology, in the whole process of science and technology project management, science and technology project consultants participate in the whole process. In different links such as guideline compilation, scientific research review, project acceptance, etc., different types of science and technology consultants are required to participate. Personnel review is a key link, but the current review person matching methods (such as manual matching, keyword matching, random matching, etc.) are difficult to meet the needs of review person matching tasks in interdisciplinary projects. Specifically, manual matching consumes a lot of resources, and it is difficult to avoid the intervention of human subjective factors, which affects the fairness and accuracy of the review results; keyword matching lacks in-depth mining of semantic features, and the matching accuracy is not satisfactory; random matching lacks systematic organization, which may lead to confusion in the review process and fail to form an effective review system. With the continuous cross-integration of science and technology and the expansion of disciplinary differences, the disadvantages of the current matching method are becoming more and more obvious, and it is difficult to cope with the current complex and changeable review needs. Researching the intelligent matching method of science and technology project reviewers in response to the above problems has practical significance and theoretical research value, which not only helps to promote the optimization of science and technology resource allocation, but also can further improve the professionalism and correctness of project review.

[0029] Currently, character recommendations mostly use the following methods:

[0030] (1) Content-based recommendation. Content-based recommendation refers to recommending other objects with similar attributes based on the object information selected by the user. It is a continuation and development of information filtering technology. It is a recommendation based on the content text information of the project, and does not need to be based on the user's evaluation of the project. It is more about obtaining the user's interest information from the text information about the feature description of the content by segmenting the user's text information and establishing a vector space model.

[0031] (2) Collaborative filtering recommendation. Collaborative filtering recommendation is based on a hypothesis: the best way to find content that a user is really interested in is to first find other user groups with similar hobbies or habits to the user, and then recommend the content that they are interested in or think is good to the user. Collaborative filtering recommendation is the most widely studied and applied personalized recommendation technology, and the most common implementation algorithm is the nearest neighbor method.

[0032] (3) Recommendation based on association rules. To some extent, the recommendation algorithm based on association rules can be regarded as a kind of reasoning technology. It is not based on user needs and preferences to generate recommendations, but uses specific rules formulated for specific fields to perform reasoning based on association rules and examples. The recommendation of this method is based on association rules, with the purchased products as the rule head and the recommended objects as the rule body.

[0033] The terms used in this disclosure are explained as follows:

[0034] 1. Knowledge graph: Knowledge graph is a form of graph organization that links various entities through semantic associations, also known as semantic web. Knowledge graph extracts and integrates structured and unstructured data, embodies the idea of ​​data governance and semantic connection, and is conducive to the use and migration of large-scale data.

[0035] 2. K-Means algorithm: K-Means algorithm is a clustering algorithm that clusters a set of data into k groups. The algorithm flow is as follows: Step 1: Randomly select k center points in the data set; Step 2: Calculate the distance from each data point to the k center points respectively, and classify the data points according to the distance; Step 3: Calculate the midpoint of the same type of data points as the position of the center point of the class to be updated; Step 4: Update the center point and repeat step 2. If the sum of the distances between each data point and the center point of the class to which it belongs no longer changes, the algorithm ends.

[0036] 3. BERT model: The BERT model is a bidirectional language representation model based on the Transformer architecture. Through pre-training on massive text data, it learns rich contextual representations and then fine-tunes them on downstream tasks to achieve extremely high performance.

[0037] 4. Text CNN model: Text CNN is a convolutional neural network (CNN) for text classification. It extracts text features through structures such as word embedding, convolutional layer, pooling layer, and fully connected layer. It is often used for short text classification, such as news and comment classification.

[0038] 5. CoSENT (Cosine Sentence) model: CoSENT is a text embedding model. In the loss function, CoSENT uses cosine similarity ranking loss to replace the classification cross entropy loss of Sentence-BERT, making the training process more suitable for application scenarios and accelerating the convergence of the model during the training phase.

[0039] As mentioned above, how to accurately match people in relevant fields of scientific and technological projects for project review has become an important research issue.

[0040] Based on the above description, if Figure 1 As shown, the character recommendation method proposed in this embodiment includes:

[0041] Step 101, obtaining the project to be reviewed.

[0042] During the specific implementation, the projects to be reviewed are scientific and technological projects that need to be reviewed.

[0043] Determine the information of the project to be reviewed. The information of the project to be reviewed includes: the name of the project to be reviewed and the scenario of the project to be reviewed. The scenario of the project to be reviewed refers to the review process of the project department, and the scenario of the project to be reviewed includes: guideline preparation, scientific research review and project acceptance.

[0044] Step 102: vectorize the items to be reviewed to obtain a vector of the items to be reviewed.

[0045] In specific implementation, the vector of the item to be reviewed is the text vector of the item to be reviewed. The vector of the item to be reviewed is generally short text data within 20 characters, and the text granularity is between the phrase level and the sentence level.

[0046] Specifically, a project name field and a project scenario field are extracted from the project to be reviewed; the project name field and the project scenario field are matched to obtain a vector of the project to be reviewed.

[0047] Step 103 , vectorizing the pre-stored multiple characters to obtain multiple character vectors.

[0048] In specific implementation, the person is a person who reviews the project to be reviewed, including researchers or experts. The person vector is the text vector of the person.

[0049] Each of the pre-stored multiple characters is stored in the form of a character information relationship diagram, wherein the character information relationship diagram represents the relationship between the character and the scientific research project.

[0050] Specifically, the scientific research fields of the person are extracted from the person information relationship diagram; the person is matched with the corresponding scientific research fields to obtain a person vector.

[0051] Step 104 : determining a candidate person whose category is consistent with the item to be reviewed from among multiple persons according to the item to be reviewed vector and the multiple person vectors.

[0052] In a specific implementation, the similarity between the item to be reviewed and each person is determined based on the vector of the item to be reviewed and multiple person vectors, and a candidate person with high similarity is determined from the multiple persons.

[0053] For example, based on the vector of the item to be reviewed and multiple character vectors, it is determined that the similarity between the item to be reviewed and character A is 70%, the similarity between the item to be reviewed and character B is 90%, and the similarity between the item to be reviewed and character C is 80%, then characters B and C are selected as candidates.

[0054] Step 105 , determining a target person from the candidate persons according to a preset screening criterion, and generating a person recommendation result according to the target person.

[0055] In specific implementation, the preset screening criteria are pre-set criteria for further screening of the candidates, wherein the preset screening criteria include avoidance screening criteria.

[0056] Specifically, the candidate person information is obtained, and whether the candidate person meets the avoidance screening criteria is determined based on the candidate person information; in response to determining that the candidate person meets the avoidance screening criteria, the candidate person is used as a target person.

[0057] Through the above embodiments, the project to be reviewed is obtained. The project to be reviewed is vectorized to obtain the project to be reviewed vector. Multiple pre-stored characters are vectorized to obtain multiple character vectors. Based on the project to be reviewed vector and the multiple character vectors, a candidate character that is consistent with the category of the project to be reviewed can be accurately determined from multiple characters, ensuring that the candidate character and the project to be reviewed belong to the same technical field, making the candidate character more accurate. A target person is determined from the candidate characters according to preset screening criteria, and a character recommendation result is generated based on the target person. In this way, by further screening the candidate characters, the target person belongs to the same technical field as the project to be reviewed and meets the preset screening criteria, making the target person more accurate. By generating character recommendation results, the target person that matches the project to be reviewed can be clearly displayed in the form of character recommendation results.

[0058] In some embodiments, before step 103, the pre-storage process of each character in the plurality of characters includes:

[0059] Step 103A, obtaining the character information of the character; wherein the character information includes identity information and scientific research information.

[0060] Step 103B, determining the annotation content according to the identity information and the scientific research information, and annotating the character information based on the annotation content to obtain pre-annotated character information.

[0061] Step 103C, extracting the scientific research information to obtain scientific research entity information and scientific research relationship information, and annotating the pre-annotated person information based on the scientific research entity information and the scientific research relationship information to obtain target annotated person information.

[0062] Step 103D: construct a character information relationship diagram based on the character information annotated by the target, store the character information relationship diagram, and display the character information relationship diagram on a character information storage interface.

[0063] In specific implementation, identity information is information related to the identity of a person. The identity information includes the name of the person and the educational background of the person. Scientific research information is information related to the scientific research results of the person. The scientific research information includes the name of the scientific research and the type of scientific research project.

[0064] The annotation content is determined according to the identity information and scientific research information, and the character information is annotated based on the annotation content to obtain pre-annotated character information. The annotation content includes entities, attributes, relations, and attribute values. Part-of-speech tagging includes verbs and nouns.

[0065] The scientific research information is extracted and processed to obtain scientific research entity information and scientific research relationship information, and the pre-annotated person information is annotated based on the scientific research entity information and scientific research relationship information to obtain the target annotated person information. Among them, the scientific research entity information includes the person's name, paper name, patent name, software title, thesis title and project name. The scientific research relationship information includes the paper relationship, patent relationship, software relationship, thesis relationship and project participant relationship.

[0066] Utilize the knowledge graph to build a character information relationship graph based on the target labeled character information, store the character information relationship graph, and display the character information relationship graph on the character information storage interface.

[0067] Figure 2 Schematic diagram of the character information relationship diagram of the embodiment of the present disclosure. Figure 2 As shown in the figure, in the character information relationship diagram, the character's name is in the middle, and the corresponding scientific research information is around the character's name. The scientific research information is represented in the form of scientific research entity information and scientific research relationship information. Based on the character information relationship diagram, it is convenient and intuitive to understand the character's scientific research information. For example, the character's scientific research information includes: paper XXX, patent XXX, software XXX, monograph XXX, and project XXX completed with project participant XX.

[0068] Through the above scheme, by constructing a character information relationship graph, it is convenient and intuitive to understand the scientific research information of each character, and the character information relationship graph can clearly represent the character's scientific research type and the corresponding scientific research name. In addition, according to the constructed character information relationship graph, the scientific research field can also be accurately extracted for vectorization processing.

[0069] In some embodiments, step 102 includes:

[0070] Step 1021: extracting a project name field and a project scenario field from the project to be reviewed based on a language representation model.

[0071] Step 1022: Based on the convolutional neural network for text, the project name field and the project scenario field are matched to obtain a vector of the project to be reviewed.

[0072] In specific implementation, the Bidirectional Encoder Representations from Transformers (BERT model) is a pre-trained model based on the Transformer structure. The Text Convolutional Neural Network (CNN model) is a model that applies convolutional neural networks to text classification tasks.

[0073] Based on the language representation model, the project name field and the project scenario field are extracted from the project to be reviewed. Based on the convolutional neural network for text, the project name field and the project scenario field are matched to obtain the project vector to be reviewed.

[0074] Among them, the project scenario fields include: guideline preparation, scientific research review and project acceptance. The project name field and project scenario field are generally short text data within 20 characters, and the text granularity is between the phrase level and the sentence level. Therefore, the BERT model combined with the Text CNN model is used to extract the project vector to be reviewed.

[0075] Through the above solution, based on the BERT model combined with the Text CNN model, the project to be reviewed is vectorized to obtain the project vector to be reviewed. In this way, based on the BERT model, the project name field and project scenario field can be accurately extracted from the project to be reviewed. Based on the Text CNN model, the project name field and project scenario field can be accurately matched to obtain the project vector to be reviewed.

[0076] In some embodiments, step 103 includes:

[0077] Step 1031 , determining a character information relationship diagram corresponding to each character according to the pre-stored multiple characters.

[0078] Step 1032: extract the scientific research field of each character from the character information relationship diagram based on the language representation model.

[0079] Step 1033, based on the text embedding model, each character is matched with the corresponding scientific research field to obtain each character vector.

[0080] Step 1034 , aggregating each character vector to obtain multiple character vectors.

[0081] In specific implementation, the speech representation model (Bidirectional Encoder Representations from Transformers, referred to as the BERT model) is a pre-trained model based on the Transformer structure. The text embedding model (Cosine Sentence, referred to as the CoSENT model) is a model for text matching.

[0082] Based on the language representation model, the scientific research fields of each character are extracted from the character information relationship diagram. Based on the text embedding model, each character is matched with the corresponding scientific research field to obtain each character vector.

[0083] Among them, the scientific research fields include: paper name, patent name, software title, thesis title and project name. For example, a person vector can be represented as "person name; paper name; patent name; software title; thesis title; project name". The person vector contains multiple short texts of words, and the text length generally reaches the sentence level. Therefore, a sentence-level text representation model (BERT model combined with CoSENT model) is used to generate the person vector.

[0084] When training, the BERT model combined with the CoSENT model only needs to consider that the similarity of positive samples is greater than the similarity of negative samples, without relying on specific values. Experimental results show that the BERT model combined with the CoSENT model has significantly better training effects in most text datasets.

[0085] Through the above solution, based on the BERT model combined with the CoSENT model, the character information is vectorized to obtain the character vector. In this way, based on the BERT model, the scientific research field can be accurately extracted from the character information. Based on the CoSENT model, the scientific research field can be accurately matched to obtain the character vector.

[0086] In some embodiments, step 104 includes:

[0087] Step 1041 : Determine the similarity between the item to be reviewed and each person according to the item to be reviewed vector and the multiple person vectors.

[0088] Step 1042: Based on the fuzzy clustering algorithm, determine a candidate person with high similarity from the multiple persons according to the similarity.

[0089] In specific implementation, the fuzzy clustering algorithm (K-Means algorithm) is an iterative clustering analysis algorithm whose goal is to divide the data set into k clusters so that each data point belongs to its nearest mean (cluster center), so that the points within the cluster are as similar as possible, and the points between clusters are as different as possible.

[0090] According to the vector of the project to be reviewed and the vectors of multiple characters, the similarity between the project to be reviewed and each character is determined. Based on the fuzzy clustering algorithm, the candidate character with high similarity is determined from the multiple characters after multiple iterations according to the similarity.

[0091] In addition, multiple preset character attributes and the weight corresponding to each character attribute are obtained, and based on the multiple preset character attributes and the weight corresponding to each character attribute, the similarity between the item to be reviewed and each character is determined according to the item to be reviewed vector and multiple character vectors, and a candidate character with high similarity is determined from the multiple characters.

[0092] For example, character attributes include educational qualifications and scientific research attributes. The weight corresponding to educational qualifications is 90%, and the weight corresponding to scientific research attributes is 10%. Based on the 90% educational qualifications weight and the 10% scientific research weight, the similarity between the project to be reviewed and each character is calculated, and the candidate with high similarity is determined from multiple characters.

[0093] The characters whose similarity among the multiple characters is greater than the preset similarity threshold are selected as the characters to be selected. For example, if the preset similarity threshold is 85%, the similarity between the item to be reviewed and character A is 70%, the similarity between the item to be reviewed and character B is 90%, and the similarity between the item to be reviewed and character C is 80%, then character B is selected as the character to be selected.

[0094] Alternatively, multiple characters are sorted in descending order of similarity, and a preset number of candidates with high similarity are determined from the sorted multiple characters. For example, if the preset number is 2, the similarity between the item to be reviewed and character A is 70%, the similarity between the item to be reviewed and character B is 90%, and the similarity between the item to be reviewed and character C is 80%, then characters B and C are selected as candidates.

[0095] Through the above scheme, the similarity between the project to be reviewed and each person can be accurately determined based on the vector of the project to be reviewed and multiple person vectors. Based on the fuzzy clustering algorithm, the candidate with high similarity is determined from multiple persons according to the similarity, so that the candidate and the project to be reviewed can be more matched.

[0096] In some embodiments, step 105 includes:

[0097] Step 1051, obtaining the unit to which the candidate person belongs and the unit to which the project to be reviewed belongs.

[0098] Step 1052, in response to determining that the unit to which the person belongs is inconsistent with the unit to which the project belongs, taking the candidate person as a target person.

[0099] In specific implementation, the unit to which the person belongs refers to the scientific research unit where the person is located, and the unit to which the project belongs refers to the scientific research unit where the project members are located.

[0100] The preset screening criteria may be an avoidance screening criteria. When the unit to which the candidate belongs is consistent with the unit to which the project belongs, there is a risk of unfair review of the project to be reviewed. Therefore, when the unit to which the candidate belongs is consistent with the unit to which the project belongs, the candidate is eliminated. When the unit to which the candidate belongs is inconsistent with the unit to which the project belongs, the candidate is used as the target person.

[0101] Through the above scheme, by judging whether the unit to which the candidate belongs is consistent with the unit to which the project belongs, it is possible to accurately judge whether the candidate has the risk of causing unfair evaluation of the project to be reviewed. When the unit to which the candidate belongs is inconsistent with the unit to which the project belongs, the candidate is taken as the target person, so that the problem of the target person affecting the unfair evaluation of the project to be reviewed can be avoided to the greatest extent.

[0102] In some embodiments, step 105 includes:

[0103] Step 1053, determining the identity information, scientific research information and recommendation score information of the target person.

[0104] Step 1054, generating a character recommendation result based on the identity information, the scientific research information and the recommendation score information, and displaying the character recommendation result on a character recommendation interface.

[0105] In specific implementation, the target person's identity information includes: the target person's name and the target person's education background. The target person's scientific research information includes: thesis name, patent name, software name, thesis name and project name. The recommendation score information can be the similarity between the project to be reviewed and the target person.

[0106] Determine the target person's identity information, scientific research information, and recommendation score information. Generate person recommendation results based on the identity information, scientific research information, and recommendation score information, and display the person recommendation results on the person recommendation interface.

[0107] Through the above scheme, by displaying the character recommendation results on the character recommendation interface, the identity information, scientific research information and recommendation score information of the target person matching the project to be reviewed can be intuitively displayed. The educational background of the target person can be clearly determined based on the identity information, the scientific research results of the target person can be clearly determined based on the scientific research information, and the degree of match between the target person and the project to be reviewed can be clearly determined based on the recommendation score information.

[0108] Through the above embodiments, the project to be reviewed is obtained. The project to be reviewed is vectorized to obtain the project to be reviewed vector. Multiple pre-stored characters are vectorized to obtain multiple character vectors. Based on the project to be reviewed vector and the multiple character vectors, a candidate character that is consistent with the category of the project to be reviewed can be accurately determined from multiple characters, ensuring that the candidate character and the project to be reviewed belong to the same technical field, making the candidate character more accurate. A target person is determined from the candidate characters according to preset screening criteria, and a character recommendation result is generated based on the target person. In this way, by further screening the candidate characters, the target person belongs to the same technical field as the project to be reviewed and meets the preset screening criteria, making the target person more accurate. By generating character recommendation results, the target person that matches the project to be reviewed can be clearly displayed in the form of character recommendation results.

[0109] It should be noted that the embodiments of the present disclosure may be further described in the following manner:

[0110] Figure 3 FIG. 1 is a schematic diagram of matching scientific and technological projects and characters according to an embodiment of the present disclosure. Figure 3 As shown in the figure, the technology project character matching process includes:

[0111] (1) Expert portrait construction model based on knowledge graph

[0112] The expert portrait architecture based on knowledge graph mainly includes four parts: data preprocessing, information extraction, information storage, and interface display. Data preprocessing mainly includes two parts: pre-labeling and document supplementation. Under the condition of meeting the privacy protection standards, relevant scientific research personnel information is obtained through different materials or systems, and the text is accurately labeled. The labeled data is used to train the natural language processing algorithm (Natural Language Processing, referred to as NLP). The annotation content includes: entities, attributes, relationships and attribute values. Part-of-speech tagging includes verbs, nouns, etc. For document supplementation, due to the professionalism of the project and the diversity of the data involved, the current NLP annotation is often unable to directly understand the document data, so as to perform data labeling. Information extraction includes: entity extraction and relationship extraction. Entities include expert names, names of participating projects, patent names, paper names, software titles, etc. Relationships include papers, patents, software works, etc. Information storage uses Neo4j graph database to store knowledge graph data according to actual conditions. Such as Figure 2 As shown, the interface display shows the stored entities and relationships in an interface.

[0113] (2) Text Topic Vectorization

[0114] The text topic vectorization representation includes three parts: review experts, scientific and technological projects, and project scenario vectorization. First, the review expert vectorization process: the model is built based on the expert portrait, and the relationship between the expert and the project can be obtained. In order to ensure that the expert can be represented effectively and consistently, the words of the participating projects, patents, software works, and papers are concatenated into a unique string, such as "participating in projects; patents; papers; software works". This method concatenates multiple short texts of words, and the text length generally reaches the sentence level. Therefore, the sentence-level text representation model BERT+CoSENT is used to generate text vectors for scientific and technological projects and experts. When training, the BERT+CoSENT model only needs to consider that the similarity of positive samples is greater than the similarity of negative samples, and does not depend on specific values. Experimental results show that the training effect of the BERT+CoSENT method is significantly better than Sentence-BERT in most Chinese data sets. Vectorized representation process of project scenarios: In the whole process of science and technology project management, science and technology consulting experts participate in the whole process, including guideline preparation, scientific research review, project acceptance and other different links. These scenarios are generally short text data within 20 characters, and the text granularity is between the phrase level and the sentence level. Therefore, the BERT+TEXTCNN method is used to generate text vectors.

[0115] (3) Expert Intelligent Recommendation Algorithm Model

[0116] A strategy of avoiding first and then dynamically extracting is adopted to determine the final list of matching experts. First, the DPK-HFC clustering algorithm is used to obtain the list of experts belonging to the same cluster as the project to be reviewed. The DPK-HFC algorithm is an adaptive method for determining local density and relative distance thresholds that takes into account the characteristics of data distribution. The sample points that meet the threshold conditions are used as the initial cluster centers. At the same time, the K-Means iteration based on the information entropy weighted fuzzy membership completes the clustering optimization; then avoidance screening is performed based on unit information to remove the candidates who belong to the same unit as the applicant; finally, the candidate expert set that has completed the avoidance screening is randomly sampled (the random.sampl function in Python can be used), and the list of review experts is dynamically obtained from the screened expert list. This ensures the accuracy of expert matching while increasing the diversity and flexibility of the extracted experts and avoiding the fixed review mode. When the project needs to undergo two or more rounds of review, this strategy effectively alleviates the problem of fixed review experts that may be caused by the TOPN idea, which helps to improve the objectivity of the review.

[0117] (4) Interface display

[0118] Users can configure indicator configuration items and weights according to different scenarios of scientific and technological projects, and the system will return expert recommendation results in different scenarios to users. The interface display content includes: scenario configuration interface of pending projects, expert recommendation list interface, scientific researcher configuration item management interface, and scientific researcher attribute weight configuration interface in different scenarios.

[0119] It should be noted that the method of the embodiment of the present disclosure can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present disclosure, and the multiple devices will interact with each other to complete the described method.

[0120] It should be noted that the above describes some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0121] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present disclosure also provides a character recommendation device.

[0122] refer to Figure 4 , the character recommendation device comprises:

[0123] The acquisition module 301 is configured to acquire the items to be reviewed;

[0124] A first vectorization processing module 302 is configured to perform vectorization processing on the item to be reviewed to obtain a vector of the item to be reviewed;

[0125] The second vectorization processing module 303 is configured to perform vectorization processing on a plurality of pre-stored characters to obtain a plurality of character vectors;

[0126] A first determination module 304 is configured to determine a candidate person who is consistent with the category of the item to be reviewed from a plurality of persons according to the item to be reviewed vector and the plurality of person vectors;

[0127] The second determination module 305 is configured to determine a target person from the candidate persons according to a preset screening criterion, and generate a person recommendation result according to the target person.

[0128] In some embodiments, the device further includes a character storage module; the character storage module includes:

[0129] A character information acquisition unit, configured to acquire character information of the character; wherein the character information includes identity information and scientific research information;

[0130] A first labeling unit is configured to determine labeling content according to the identity information and the scientific research information, and label the character information based on the labeling content to obtain pre-labeled character information;

[0131] A second labeling unit is configured to extract the scientific research information to obtain scientific research entity information and scientific research relationship information, and label the pre-labeled person information based on the scientific research entity information and the scientific research relationship information to obtain target labeled person information;

[0132] The character storage unit is configured to construct a character information relationship diagram based on the character information annotated by the target, store the character information relationship diagram, and display the character information relationship diagram on a character information storage interface.

[0133] In some embodiments, the first vectorization processing module 302 includes:

[0134] A first extraction unit is configured to extract a project name field and a project scenario field from the project to be reviewed based on a language representation model;

[0135] The first matching unit is configured to match the project name field and the project scenario field based on a convolutional neural network for text to obtain a vector of the project to be reviewed.

[0136] In some embodiments, the second vectorization processing module 303 includes:

[0137] A relationship diagram determining unit, configured to determine a character information relationship diagram corresponding to each character according to the pre-stored multiple characters;

[0138] A second extraction unit is configured to extract the scientific research field of each character from the character information relationship diagram based on a language representation model;

[0139] The second matching unit is configured to match each character with the corresponding scientific research field based on the text embedding model to obtain each character vector;

[0140] The summarizing processing unit is configured to perform summarizing processing on each character vector to obtain multiple character vectors.

[0141] In some embodiments, the first determining module 304 includes:

[0142] a similarity determination unit configured to determine the similarity between the item to be reviewed and each person according to the item to be reviewed vector and the plurality of person vectors;

[0143] The candidate person determination unit is configured to determine a candidate person with high similarity from the multiple persons according to the similarity based on a fuzzy clustering algorithm.

[0144] In some embodiments, the second determining module 305 includes:

[0145] An affiliation unit acquisition unit is configured to acquire the affiliation unit of the candidate person and the affiliation unit of the project to be reviewed;

[0146] The target person determination unit is configured to, in response to determining that the unit to which the person belongs is inconsistent with the unit to which the project belongs, select the candidate person as the target person.

[0147] In some embodiments, the second determining module 305 includes:

[0148] A target person information determination unit, configured to determine the identity information, scientific research information and recommendation score information of the target person;

[0149] The character recommendation result generating unit is configured to generate a character recommendation result based on the identity information, the scientific research information and the recommendation score information, and display the character recommendation result on the character recommendation interface.

[0150] For the convenience of description, the above device is described by dividing it into various modules according to its functions. Of course, when implementing the present disclosure, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0151] The device of the above embodiment is used to implement the corresponding character recommendation method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0152] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the character recommendation method described in any of the above embodiments is implemented.

[0153] Figure 5A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 in the device.

[0154] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0155] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.

[0156] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0157] The communication interface 1040 is used to connect a communication module (not shown in the figure) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB (Universal Serial Bus), network cable, etc.), or through a wireless mode (such as mobile network, WIFI (Wireless Fidelity), Bluetooth, etc.).

[0158] The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).

[0159] It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.

[0160] The electronic device of the above embodiment is used to implement the corresponding character recommendation method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0161] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present disclosure also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the character recommendation method described in any of the above embodiments.

[0162] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0163] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the character recommendation method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0164] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present disclosure is limited to these examples. Based on the concept of the present disclosure, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present disclosure as described above, which are not provided in detail for the sake of simplicity.

[0165] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present disclosure difficult to understand, the known power / ground connections to the integrated circuit (IC) chips and other components may or may not be shown in the provided figures. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present disclosure difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present disclosure will be implemented (that is, these details should be fully within the scope of understanding of those skilled in the art). Where specific details (e.g., circuits) are set forth to describe exemplary embodiments of the present disclosure, it is apparent to those skilled in the art that the embodiments of the present disclosure can be implemented without these specific details or with changes in these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0166] Although the present disclosure has been described in conjunction with specific embodiments of the present disclosure, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0167] The embodiments of the present disclosure are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present disclosure. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A character recommendation method, characterized in that: The method comprises: Get the items to be reviewed; Performing vectorization processing on the items to be reviewed to obtain a vector of items to be reviewed; Vectorizing the pre-stored multiple characters to obtain multiple character vectors; Determining, from the plurality of persons, a person to be selected that is consistent with the category of the item to be reviewed according to the vector of the item to be reviewed and the plurality of person vectors; A target person is determined from the candidate persons according to a preset screening criterion, and a person recommendation result is generated according to the target person.

2. The method according to claim 1, characterized in that The pre-storage process of each character among the plurality of characters comprises: Obtaining the character information of the character; wherein the character information includes identity information and scientific research information; Determining the annotation content according to the identity information and the scientific research information, and annotating the character information based on the annotation content to obtain pre-annotated character information; Extracting the scientific research information to obtain scientific research entity information and scientific research relationship information, and annotating the pre-annotated person information based on the scientific research entity information and the scientific research relationship information to obtain target annotated person information; A character information relationship graph is constructed based on the character information annotated by the target, the character information relationship graph is stored, and the character information relationship graph is displayed on a character information storage interface.

3. The method according to claim 1, characterized in that The vectorization of the items to be reviewed to obtain the vector of the items to be reviewed includes: Based on the language representation model, extracting a project name field and a project scenario field from the project to be reviewed; Based on the convolutional neural network for text, the project name field and the project scenario field are matched to obtain a vector of the project to be reviewed.

4. The method according to claim 1, characterized in that The step of vectorizing the pre-stored multiple characters to obtain multiple character vectors includes: Determine a character information relationship graph corresponding to each character according to the pre-stored multiple characters; Based on the language representation model, extract the scientific research field of each character from the character information relationship diagram; Based on the text embedding model, each character is matched with the corresponding scientific research field to obtain each character vector; Each character vector is aggregated to obtain multiple character vectors.

5. The method according to claim 1, characterized in that The step of determining, from a plurality of persons, a person to be selected that is consistent with the category of the item to be reviewed based on the item to be reviewed vector and the plurality of person vectors includes: Determine the similarity between the item to be reviewed and each person according to the item to be reviewed vector and the multiple person vectors; Based on a fuzzy clustering algorithm, a candidate person with high similarity is determined from the multiple persons according to the similarity.

6. The method according to claim 1, characterized in that The step of determining the target person from the candidate persons according to the preset screening criteria includes: Obtain the unit to which the candidate person belongs and the unit to which the project to be reviewed belongs; In response to determining that the unit to which the person belongs is inconsistent with the unit to which the project belongs, the person to be selected is used as a target person.

7. The method according to claim 1, characterized in that Generating a character recommendation result according to the target character includes: Determine the identity information, scientific research information and recommendation score information of the target person; Based on the identity information, the scientific research information and the recommendation score information, a character recommendation result is generated, and the character recommendation result is displayed on a character recommendation interface.

8. A character recommendation device, characterized in that: include: An acquisition module, configured to acquire items to be reviewed; A first vectorization processing module is configured to perform vectorization processing on the item to be reviewed to obtain an item to be reviewed vector; A second vectorization processing module is configured to perform vectorization processing on a plurality of pre-stored characters to obtain a plurality of character vectors; A first determination module is configured to determine a candidate person who is consistent with the category of the item to be reviewed from a plurality of persons according to the item to be reviewed vector and the plurality of person vectors; The second determination module is configured to determine a target person from the candidate persons according to a preset screening criterion, and generate a person recommendation result according to the target person.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and running on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to cause a computer to execute the method according to any one of claims 1 to 7.

Images