A project name recommendation method based on NLP and knowledge graph

Abstract

The application discloses a project name recommendation method based on NLP and a knowledge graph, relates to the fields of artificial intelligence and data mining technology, and comprises the following steps: step 1, collecting multidimensional data of related projects and preprocessing to construct a project name corpus; step 2, constructing a project knowledge graph according to the project name corpus; step 3, performing core keyword recognition, potential keyword recognition and trend keyword prediction according to the project knowledge graph and the project name corpus to obtain a keyword cluster; and step 4, updating the project name corpus according to the keyword cluster, training a constructed project name generation model by using the updated project name corpus, inputting user input information into the trained project name generation model, and obtaining a recommended project name. The application combines multidimensional data processing, has the functions of keyword recognition and application landing of project funding keyword analysis, and provides technical and data support for scientific research project establishment.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, computer science, and data mining, and more specifically to a project name recommendation method based on NLP and knowledge graphs. Background Technology

[0002] Natural Language Processing (NLP) is a core interdisciplinary field at the intersection of computer science, artificial intelligence, and linguistics, aiming to enable computers to understand, generate, and interact with human language. Currently, NLP technology has been applied and achieved results in fields such as language, finance, and medicine. The combination of NLP and knowledge graphs has also become a development trend in recent years. NLP can provide entities and relationships for the construction of knowledge graphs, while knowledge graphs provide prior knowledge for NLP, thereby improving the effectiveness of NLP.

[0003] Therefore, how to utilize natural language processing and knowledge graphs to achieve accurate text analysis and project recommendation is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of the above problems, this invention is proposed to provide a project name recommendation method based on NLP and knowledge graphs to overcome or at least partially solve the above problems. It adopts a method combining natural language processing and knowledge graphs, and proposes a solution that integrates multi-source data, mines project keywords, and provides intelligent recommendation, taking into account the characteristics of diversified disciplines, decentralized project themes, and rapid iteration of research keywords. It integrates multi-dimensional data processing and project keyword analysis that combines keyword recognition and practical application, providing technical and data support for scientific research project initiation.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, embodiments of the present invention provide a project name recommendation method based on NLP and knowledge graphs, comprising the following steps:

[0007] Step 1: Collect multi-dimensional data of relevant projects and preprocess them to build a project name corpus;

[0008] Step 2: Construct a project knowledge graph based on the project name corpus;

[0009] Step 3: Based on the project knowledge graph and project name corpus, identify core keywords, potential keywords, and predict trend keywords to obtain keyword clusters;

[0010] Step 4: Update the project name corpus based on keyword clusters, and train the constructed project name generation model using the updated project name corpus. Collect user input information and input it into the trained project name generation model to obtain recommended project names.

[0011] Preferably, the multi-dimensional data collected in step 1 includes structured and unstructured data, including project name, region, user name, project approval time, project subject, amount, etc.

[0012] Preferably, the preprocessing of structured data includes deduplication, standardization, and supplementary information; deduplication is performed based on project name, user name, and project approval time; corresponding area codes, location attributes, user types, and user locations are supplemented based on the region; the project approval time is standardized into a numerical format, and the amount is standardized into a currency unit to achieve standardization.

[0013] Preferably, the preprocessing of unstructured data involves extracting technical keywords and application scenario keywords from the project name; the specific process is as follows:

[0014] Step 11: Use a pre-trained word segmentation model to segment the project name to obtain several name word blocks;

[0015] Step 12: Remove stop words such as articles, prepositions, conjunctions, and punctuation marks from each name block to obtain valid word blocks;

[0016] Step 13: Use the fusion method of TF-IDF and TextRank to extract technical keywords and application scenario keywords from effective word blocks.

[0017] Preferably, the process of extracting technical keywords and application scenario keywords in step 13 is as follows:

[0018] Step 131: Calculate the TF-IDF value for each valid term in each project name based on the valid term chunks and project name. Select the valid term chunk with the highest TF-IDF value as the technical keyword for that project name; the expression is:

[0019] TF-IDF=TF IDF;

[0020] TF = Number of valid word chunks appearing in the item name / Total number of valid word chunks in the item name;

[0021] IDF = log(total number of item names / (number of item names containing valid keywords + 1));

[0022] Step 132: For each item name after word segmentation, use a window of size 2, moving from left to right. Record the number of times any valid word block appears together with other valid word blocks in each window, constructing an application scenario keyword graph G=(V,E), where V is the set of nodes composed of valid word blocks, and E is an edge of G, representing the number of times two valid word blocks appear together; the expression is:

[0023]

[0024] in, Indicates the currently valid word block to be calculated. The final weight value; d represents the damping coefficient, with a value of 0.85; Indicates the preceding word block, i.e., the word block that precedes the valid word block. Valid word blocks that appear simultaneously in one window, and valid word blocks They have a co-occurrence relationship; Indicates a preceding word block The out-degree is the total number of other valid word blocks that the word block can be associated with. Indicates a preceding word block The final weight values ​​that have already been calculated; Indicates all words related to the currently valid word block. The set of preceding word blocks that have a common occurrence relationship; the number of iterations is set to 50 to make the formula converge; the effective word block with the largest final weight value is taken as the application scenario keyword.

[0025] Preferably, a project name corpus is constructed by associating the project name with corresponding technical keywords, application scenario keywords, user types, location attributes, and the topic to which the project belongs.

[0026] Preferably, entities, attributes, and relationships are established based on the project name corpus to construct a project knowledge graph; the entities include project entities, user entities, region entities, and topic entities, with the project name as the project entity, the user name as the user entity, the region as the region entity, and the topic to which the project belongs as the topic entity;

[0027] The entity attributes corresponding to project entities include project approval time, amount, technical keywords, and application scenario keywords. The entity attributes corresponding to user entities include user type, user location, and user preferences. The entity attributes corresponding to region entities include region code, location attribute, and cooperation activity level. User preferences are obtained by statistically analyzing the topic to which the user name belongs. The number of projects participated in is counted based on the region code, and the number of projects participated in corresponding to a region is used as the cooperation activity level.

[0028] Relationships include funding relationships, partnership relationships, attribution relationships, time-related relationships, and monetary relationships. Funding relationships are between user entities and project entities, with the monetary attribute. Partnership relationships are between regional entities, with the number of partnership projects as the attribute. Attribution relationships are between project entities and subject entities, with domain-related attributes. Time-related relationships are between project entities and time, with the project approval time as the attribute. Monetary relationships are between subject entities and project entities, with the average monetary value as the attribute. Based on the correspondence between project names and regions, the number of identical project names corresponding to any two sets of regions is taken as the number of partnership projects between the entities in those two sets of regions. Domain-related relationships are manually defined based on project entities and subject entities. The total monetary value is calculated by summing the monetary values ​​of all project names related to the project's subject, and the average monetary value is calculated as: total monetary value / number of project names.

[0029] Preferably, the project knowledge graph is imported into the Neo4j graph database for storage.

[0030] Preferably, step 3 includes the following steps:

[0031] Step 31: Identify core keywords based on project names in the project knowledge graph and project name corpus to obtain core keyword clusters;

[0032] Step 32: Based on the project knowledge graph and core keyword clusters, use the association rule mining algorithm to identify potential keywords and filter out potential keyword clusters;

[0033] Step 33: Based on the core keyword cluster and potential keyword cluster, use a pre-trained LSTM neural network model to predict trend keywords and output trend keyword clusters.

[0034] The preferred process for identifying core keywords is as follows:

[0035] Step 311: Based on the project knowledge graph, search and statistically analyze the number of associated projects, total amount, and average amount for each topic entity; perform search and statistical analysis based on the entity attributes and relationship attributes corresponding to each topic entity in the project knowledge graph;

[0036] Step 312: Calculate the weights corresponding to the number of associated items, total amount, and average amount for each subject entity using the entropy weight method;

[0037] Step 313: Calculate the weighted sum of the number of related projects, total amount, and average amount based on their weights to obtain the topic popularity value for each topic entity;

[0038] Step 314: Using the K-means clustering algorithm, cluster semantically similar keywords into one category based on the technical keywords and application scenario keywords corresponding to the project names in the project name corpus, and obtain several keyword clusters;

[0039] Step 315: Based on the topic popularity value of the topic entity corresponding to all project names in each keyword cluster, select the keyword with the highest popularity value in each keyword cluster as the core keyword to form a core keyword cluster.

[0040] The preferred process for identifying potential keywords is as follows:

[0041] Step 321: Divide the project approval time corresponding to the core keyword cluster within the preset time period threshold by quarter, count the number of project names in each quarter to obtain the number of projects in each quarter, calculate the quarterly growth rate of each core keyword in the core keyword cluster based on the number of projects in each quarter, sort them in descending order, and select the top k core keywords with the highest quarterly growth rate, where k is a positive integer greater than or equal to 1; the time period threshold is the most recent 3 years.

[0042] The quarterly growth rate of core keywords = (number of projects in the current quarter - number of projects in the previous quarter) / number of projects in the previous quarter;

[0043] Step 322: Search and statistically analyze the historical technical keywords and application scenario keywords of each user entity based on the project knowledge graph, and use them as historical project keywords;

[0044] Step 323: Use association rule mining algorithm and preset similarity threshold to select core keywords that are associated with historical project keywords from k core keywords to form potential keyword clusters; the similarity threshold can be 0.8, and select keyword clusters with similarity greater than 0.8.

[0045] The preferred method for predicting trend keywords is as follows:

[0046] A pre-trained LSTM neural network model is used to predict and output trend keyword clusters for future periods based on core keyword clusters, potential keyword clusters, and data from the project name corpus. The future period can be set to 1-2 years. The data in the project name corpus is time series data.

[0047] Preferably, step 4 employs a generation strategy of keyword fusion, semantic logic construction, and feasibility optimization to generate recommended project names based on user input information; the specific steps are as follows:

[0048] Step 41: Extract core elements from the core keyword cluster, potential keyword cluster, and trending keyword cluster, including technical keywords, application scenario keywords, and user preference keywords;

[0049] Step 42: Label the project name according to the format of technical keywords - application scenario keywords - user preference keywords, and update the project name corpus by selecting a high-frequency semantic framework and making unified adjustments;

[0050] Step 43: Use the updated project name corpus to construct training sample data with the input as the core elements and the output as project names, and train the project name generation model based on LLaMA3.

[0051] Step 44: Collect user input information and input it into the trained project name generation model to output recommended project names.

[0052] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a project name recommendation method based on NLP and knowledge graph. First, it integrates multi-dimensional project data to construct a comprehensive keyword analysis graph database; then, based on user input information, it accurately identifies the core keywords, potential keywords, and trend changes of the current project funding; finally, based on the identified keywords, it autonomously generates targeted and feasible new project recommendation names, providing a reference for project applications. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0054] Figure 1 This is a schematic diagram of a project name recommendation method based on NLP and knowledge graph provided in an embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] This invention discloses a project name recommendation method based on NLP and knowledge graphs, comprising the following steps:

[0057] S1: Collect multi-dimensional data of relevant projects and preprocess them to build a project name corpus;

[0058] S2: Construct a project knowledge graph based on a project name corpus;

[0059] S3: Based on the project knowledge graph and project name corpus, identify core keywords, potential keywords, and predict trend keywords to obtain keyword clusters;

[0060] S4: Update the project name corpus based on keyword clusters, and train the constructed project name generation model using the updated corpus. Collect user input information and input it into the trained project name generation model to obtain recommended project names. Keywords can be high-popularity keywords to represent the hot topics of the project.

[0061] In one specific embodiment, a project name recommendation method based on NLP and knowledge graphs is as follows: Figure 1 As shown, it includes the following steps:

[0062] S1: Data acquisition and preprocessing;

[0063] S11: Collect multidimensional data related to the project, including structured and unstructured data, specifically including project name, region, user name, project approval time, project theme, amount, etc.; the region can be different countries;

[0064] S12: Preprocessing of structured data includes: deduplication based on project name, user name, and project approval time; supplementing the corresponding region code, location attribute, user type, and user location based on the region; standardizing the project approval time to a "XXXX-XX-XX" numerical format and the amount to a unified currency unit; the location attribute is the project location, the user can be the funder, and the region code can be an ISO code;

[0065] The specific preprocessing process for unstructured data is as follows:

[0066] Using natural language processing technology, based on the PyTorch+NLTK module of Python, a word segmentation model based on Bidirectional Long Short-Term Memory Network (BiLSTM) is established. The word segmentation model is used to segment the project name to obtain several name word blocks.

[0067] Remove stop words such as articles, prepositions, conjunctions, and punctuation marks from each name block to obtain valid word blocks;

[0068] A fusion method of TF-IDF and TextRank is used to extract technical keywords and application scenario keywords from effective word blocks;

[0069] The process of extracting technical keywords and application scenario keywords is as follows:

[0070] Calculate the TF-IDF value for each valid term in each project name based on the valid term chunks and project name, and select the valid term chunk with the largest TF-IDF value as the technical keyword of the project name; the expression is:

[0071] TF-IDF=TF IDF;

[0072] TF = Number of valid word chunks appearing in the item name / Total number of valid word chunks in the item name;

[0073] IDF = log(total number of item names / (number of item names containing valid keywords + 1));

[0074] For each item name after word segmentation, a window of size 2 is used, moving from left to right. Each window records the number of times any valid word block co-occurs with other valid word blocks, constructing an application scenario keyword graph G=(V,E), where V is the node set composed of valid word blocks, and E is an edge of G, representing the number of times two valid word blocks co-occur. The expression is:

[0075]

[0076] in, Indicates the currently valid word block to be calculated. The final weight value; d represents the damping coefficient, with a value of 0.85; Indicates the preceding word block, i.e., the word block that precedes the valid word block. Valid word blocks that appear simultaneously in one window, and valid word blocks They have a co-occurrence relationship; Indicates a preceding word block The out-degree is the total number of other valid word blocks that the word block can be associated with. Indicates a preceding word block The final weight values ​​that have already been calculated; Indicates all words related to the currently valid word block. The set of preceding word blocks that have a common occurrence relationship; the number of iterations is set to 50 to make the formula converge; the effective word block with the largest final weight value is taken as the application scenario keyword;

[0077] S13: Associate project names with corresponding technical keywords, application scenario keywords, user types, location attributes, and the topic to which the project belongs, and build a project name corpus.

[0078] S2: Based on the project name corpus, establish entities, attributes, and relationships, construct a project knowledge graph, and import it into the Neo4j graph database for storage; entities include project entities, user entities, region entities, and topic entities, with project name as project entity, user name as user entity, region as region entity, and the topic to which the project belongs as topic entity;

[0079] The entity attributes corresponding to project entities include project approval time, amount, technical keywords, and application scenario keywords. The entity attributes corresponding to user entities include user type, user location, and user preferences. The entity attributes corresponding to region entities include region code, location attribute, and cooperation activity level. User preferences are obtained by statistically analyzing the topic to which the user name belongs. The number of projects participated in is counted based on the region code, and the number of projects participated in corresponding to a region is used as the cooperation activity level.

[0080] Relationships include funding relationships, partnership relationships, attribution relationships, time-related relationships, and monetary relationships. Funding relationships are between user entities and project entities, with the monetary attribute. Partnership relationships are between regional entities, with the number of partnership projects as the attribute. Attribution relationships are between project entities and subject entities, with domain-related attributes. Time-related relationships are between project entities and time, with the project approval time as the attribute. Monetary relationships are between subject entities and project entities, with the average monetary value as the attribute. Based on the correspondence between project names and regions, the number of identical project names corresponding to any two sets of regions is taken as the number of partnership projects between the entities in those two sets of regions. Domain-related relationships are manually defined based on project entities and subject entities. The total monetary value is calculated by summing the monetary values ​​of all project names related to the project's subject, and the average monetary value is calculated as: total monetary value / number of project names.

[0081] S3: Perform keyword identification;

[0082] S31: Based on the project knowledge graph and the project name corpus, identify core keywords to obtain core keyword clusters;

[0083] S311: Based on the project knowledge graph, search and statistically analyze the number of associated projects, total amount, and average amount for each topic entity; perform search and statistical analysis based on the entity attributes and relationship attributes corresponding to each topic entity in the project knowledge graph; the total amount can be the total amount of funding.

[0084] S312: Use the entropy weight method to calculate the number of associated projects, total amount, and average funding weight for each subject entity;

[0085] S313: Based on the weights, the number of related projects, the total amount, and the average amount are weighted and summed to obtain the topic popularity value corresponding to each topic entity;

[0086] Theme popularity score = (project quantity percentage × w1) + (total amount percentage × w2) + (average amount percentage × w3);

[0087] Where w1, w2, and w3 are the weights corresponding to the number of related projects, the total amount, and the average amount, respectively; w1=0.4, w2=0.35, and w3=0.25.

[0088] S314: Using the K-means clustering algorithm, the technical keywords and application scenario keywords corresponding to the project names in the project name corpus are clustered into a class based on semantically similar keywords to obtain several keyword clusters.

[0089] S315: Based on the topic popularity value of the topic entity corresponding to all project names in each keyword cluster, select the keyword with the highest popularity value in each keyword cluster as the core keyword to form a core keyword cluster;

[0090] S32: Based on the project knowledge graph and core keywords, use the association rule mining algorithm to identify potential keywords and filter out potential keyword clusters;

[0091] S321: Divide the annotation time of the projects corresponding to the core keyword clusters in the past 3 years into quarters, count the number of project names in each quarter to obtain the number of projects in each quarter, calculate the quarterly growth rate of each core keyword in the core keyword cluster based on the number of projects in each quarter, sort them in descending order, and select the top 5 core keywords with the highest quarterly growth rate.

[0092] The quarterly growth rate of core keywords = (number of projects in the current quarter - number of projects in the previous quarter) / number of projects in the previous quarter;

[0093] S322: Search and statistically analyze the historical technical keywords and application scenario keywords of each user entity based on the project knowledge graph, and use them as historical project keywords; historical project keywords can be keywords of projects funded by the user in the past.

[0094] S323: Using the Apriori association rule mining algorithm and a preset similarity threshold of 0.8, core keywords associated with historical project keywords are selected from 5 core keywords as potential keyword clusters; keyword clusters with a similarity greater than 0.8 are selected.

[0095] S33: Based on the core keyword cluster and potential keyword cluster, a pre-trained LSTM neural network model is used to predict trend keywords and output trend keyword clusters;

[0096] A pre-trained LSTM neural network model is used to predict the keyword development trend for the next 1-2 years based on the core keyword cluster, potential keyword cluster, and project name corpus, generating trend keyword clusters. The data in the project name corpus is time series data.

[0097] S4: Generate recommended project names; using a generation strategy of keyword fusion, semantic logic construction, and feasibility optimization, generate recommended project names based on user input information;

[0098] S41: Extract core elements from the core keyword cluster, potential keyword cluster, and trend keyword cluster, including technical keywords, application scenario keywords, and user preference keywords;

[0099] S42: Annotate and mine the semantic structure of project names according to the format of technical keywords-application scenario keywords-user preference keywords, and update the project name corpus by selecting high-frequency semantic frameworks and making unified adjustments;

[0100] S43: Use the updated project name corpus to construct training sample data with the input as the core element and the output as the project name, and train the project name generation model based on LLaMA3.

[0101] S44: Collect user input information and input it into the trained project name generation model to output recommended project names; user input information includes keywords or research titles.

[0102] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0103] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A project name recommendation method based on NLP and knowledge graph, characterized in that, Includes the following steps: Step 1: Collect multi-dimensional data of relevant projects and preprocess them to build a project name corpus; Step 2: Construct a project knowledge graph based on the project name corpus; Step 3: Based on the project knowledge graph and project name corpus, identify core keywords, potential keywords, and predict trend keywords to obtain keyword clusters; Step 4: Update the project name corpus based on keyword clusters, and train the constructed project name generation model using the updated project name corpus. Collect user input information and input it into the trained project name generation model to obtain recommended project names. Step 3 includes the following steps: Step 31: Identify core keywords based on project names in the project knowledge graph and project name corpus to obtain core keyword clusters; Step 32: Based on the project knowledge graph and core keyword clusters, use the association rule mining algorithm to identify potential keywords and filter out potential keyword clusters; Step 33: Predict trend keywords based on core keyword clusters and potential keyword clusters, and output trend keyword clusters; The process of identifying core keywords is as follows: Step 311: Based on the project knowledge graph, search and statistically analyze the number of associated projects, total amount, and average amount for each topic entity; Step 312: Calculate the weights corresponding to the number of associated items, total amount, and average amount for each subject entity using the entropy weight method; Step 313: Calculate the weighted sum of the number of related projects, total amount, and average amount based on their weights to obtain the topic popularity value for each topic entity; Step 314: Using the K-means clustering algorithm, cluster semantically similar keywords into one category based on the technical keywords and application scenario keywords corresponding to the project names in the project name corpus, and obtain several keyword clusters; Step 315: Based on the topic popularity value of the topic entity corresponding to all project names in each keyword cluster, select the keyword with the highest popularity value in each keyword cluster as the core keyword to form a core keyword cluster; The process of identifying potential keywords is as follows: Step 321: Divide the project approval time corresponding to the core keyword cluster within the preset time period threshold by quarter, count the number of project names corresponding to each quarter to obtain the number of projects in each quarter, calculate the quarterly growth rate of each core keyword in the core keyword cluster based on the number of projects in each quarter, sort them in descending order, and select the top k core keywords with the highest quarterly growth rate, where k is a positive integer greater than or equal to 1. Step 322: Search and statistically analyze the historical technical keywords and application scenario keywords of each user entity based on the project knowledge graph, and use them as historical project keywords; Step 323: Use association rule mining algorithm and preset similarity threshold to filter out core keywords associated with historical project keywords from k core keywords, and form potential keyword clusters; The specific process of trend keyword prediction is as follows: A pre-trained LSTM neural network model is used to predict and output trend keyword clusters for future periods based on data from the core keyword cluster, potential keyword cluster, and project name corpus.

2. The project name recommendation method based on NLP and knowledge graph as described in claim 1, characterized in that, The multi-dimensional data collected in step 1 includes structured and unstructured data, including project name, region, user name, project approval time, project subject, and amount.

3. The project name recommendation method based on NLP and knowledge graph as described in claim 2, characterized in that, Preprocessing of structured data includes deduplication, standardization, and information supplementation. Supplementation includes adding corresponding region codes, location attributes, user types, and user locations based on the region. Preprocessing of unstructured data involves: The project name is segmented using a pre-trained word segmentation model to obtain several name word blocks; Remove stop words from each name block to obtain valid word blocks; A fusion method of TF-IDF and TextRank is used to extract technical keywords and application scenario keywords from effective word blocks.

4. The project name recommendation method based on NLP and knowledge graph as described in claim 3, characterized in that, By associating project names with corresponding technical keywords, application scenario keywords, user types, location attributes, and the topic to which the project belongs, a project name corpus is constructed.

5. The project name recommendation method based on NLP and knowledge graph as described in claim 4, characterized in that, Based on the project name corpus, entities, attributes, and relationships are established to construct a project knowledge graph. Entities include project entities, user entities, region entities, and topic entities. The project name is used as the project entity, the user name as the user entity, the region as the region entity, and the topic to which the project belongs as the topic entity. The entity attributes corresponding to the project entity include project approval time, amount, technical keywords and application scenario keywords; the entity attributes corresponding to the user entity include user type, user location and user preferences; and the entity attributes corresponding to the region entity include region code, location attribute and cooperation activity level. Relationships include funding relationships, cooperative relationships, affiliation relationships, temporal relationships, and monetary relationships; Funding relationships are between user entities and project entities, with the relationship attribute being the amount; cooperation relationships are between regional entities, with the relationship attribute being the number of cooperative projects; affiliation relationships are between project entities and subject entities, with the relationship attribute being domain association. Time-related relationships are the relationship between project entities and time, with the relationship attribute being the project approval time; monetary relationships are the relationship between subject entities and project entities, with the relationship attribute being the average amount.

6. The project name recommendation method based on NLP and knowledge graph as described in claim 1, characterized in that, Step 4 consists of the following steps: Step 41: Extract core elements from the core keyword cluster, potential keyword cluster, and trending keyword cluster, including technical keywords, application scenario keywords, and user preference keywords; Step 42: Label project names according to the format of technical keywords - application scenario keywords - user preference keywords, and update the project name corpus; Step 43: Use the updated project name corpus to construct training sample data with the input as the core elements and the output as project names, and train the project generation model based on LLaMA3. Step 44: Collect user input information and input it into the trained project name generation model to output recommended project names.

Images